diff --git a/comfy/latent_formats.py b/comfy/latent_formats.py index 12a934d71..15b5497c8 100644 --- a/comfy/latent_formats.py +++ b/comfy/latent_formats.py @@ -4,6 +4,7 @@ class LatentFormat: scale_factor = 1.0 latent_channels = 4 latent_dimensions = 2 + preserve_empty_channel_multiples = False latent_rgb_factors = None latent_rgb_factors_bias = None latent_rgb_factors_reshape = None @@ -769,6 +770,10 @@ class ACEAudio(LatentFormat): latent_channels = 8 latent_dimensions = 2 +class SeedVR2(LatentFormat): + latent_channels = 16 + preserve_empty_channel_multiples = True + class ACEAudio15(LatentFormat): latent_channels = 64 latent_dimensions = 1 diff --git a/comfy/ldm/modules/attention.py b/comfy/ldm/modules/attention.py index 55360535a..8507557d5 100644 --- a/comfy/ldm/modules/attention.py +++ b/comfy/ldm/modules/attention.py @@ -32,6 +32,14 @@ except ImportError as e: raise e exit(-1) +SAGE_ATTENTION_VARLEN_IS_AVAILABLE = False +try: + from sageattention import sageattn_varlen + SAGE_ATTENTION_VARLEN_IS_AVAILABLE = True +except ImportError: + if model_management.sage_attention_enabled(): + logging.warning("SageAttention variable-length attention is unavailable, using pytorch var-len attention instead.") + SAGE_ATTENTION3_IS_AVAILABLE = False try: from sageattn3 import sageattn3_blackwell @@ -40,6 +48,7 @@ except ImportError: pass FLASH_ATTENTION_IS_AVAILABLE = False +FLASH_ATTENTION_VARLEN_IS_AVAILABLE = False try: from flash_attn import flash_attn_func FLASH_ATTENTION_IS_AVAILABLE = True @@ -48,6 +57,20 @@ except ImportError: logging.error(f"\n\nTo use the `--use-flash-attention` feature, the `flash-attn` package must be installed first.\ncommand:\n\t{sys.executable} -m pip install flash-attn") exit(-1) +try: + from flash_attn import flash_attn_varlen_func + FLASH_ATTENTION_VARLEN_IS_AVAILABLE = True +except ImportError: + if model_management.flash_attention_enabled() and FLASH_ATTENTION_IS_AVAILABLE: + logging.warning("Flash Attention variable-length attention is unavailable, using pytorch var-len attention instead.") + +FLASH_ATTENTION3_IS_AVAILABLE = False +try: + from flash_attn_interface import flash_attn_varlen_func as flash_attn3_varlen_func + FLASH_ATTENTION3_IS_AVAILABLE = True +except ImportError: + pass + REGISTERED_ATTENTION_FUNCTIONS = {} def register_attention_function(name: str, func: Callable): # avoid replacing existing functions @@ -735,28 +758,434 @@ def attention_flash(q, k, v, heads, mask=None, attn_precision=None, skip_reshape ) return out +_VAR_ATTENTION_NESTED_API_NAME = "nested_tensor_from_jagged" +_VAR_ATTENTION_GUARD_MESSAGE = ( + "SeedVR2 var_attention_pytorch: torch.nested.nested_tensor_from_jagged " + "is required by this attention path; the installed PyTorch build " + "does not provide it" +) + +def _var_attention_max_seqlen(cu_seqlens): + return int((cu_seqlens[1:] - cu_seqlens[:-1]).max().item()) + + +def _var_attention_qkv(q, k, v, heads, skip_reshape): + if skip_reshape: + return q, k, v, q.shape[-1] + total_tokens, embed_dim = q.shape + head_dim = embed_dim // heads + return ( + q.view(total_tokens, heads, head_dim), + k.view(k.shape[0], heads, head_dim), + v.view(v.shape[0], heads, head_dim), + head_dim, + ) + + +def _var_attention_output(out, heads, head_dim, skip_output_reshape): + if skip_output_reshape: + return out + return out.reshape(-1, heads * head_dim) + + +def _use_blackwell_attention(): + device = model_management.get_torch_device() + if device.type != "cuda": + return False + major, minor = torch.cuda.get_device_capability(device) + return (major, minor) >= (12, 0) + + +def var_attention_pytorch(q, k, v, heads, cu_seqlens_q, cu_seqlens_k, skip_reshape=False, skip_output_reshape=False): + _nested = getattr(torch, "nested", None) + if _nested is None or not hasattr(_nested, _VAR_ATTENTION_NESTED_API_NAME): + raise RuntimeError(_VAR_ATTENTION_GUARD_MESSAGE) + + if not skip_reshape: + # assumes 2D q, k,v [total_tokens, embed_dim] + total_tokens, embed_dim = q.shape + head_dim = embed_dim // heads + q = q.view(total_tokens, heads, head_dim) + k = k.view(k.shape[0], heads, head_dim) + v = v.view(v.shape[0], heads, head_dim) + + q = torch.nested.nested_tensor_from_jagged(q, offsets=cu_seqlens_q.long()) + k = torch.nested.nested_tensor_from_jagged(k, offsets=cu_seqlens_k.long()) + v = torch.nested.nested_tensor_from_jagged(v, offsets=cu_seqlens_k.long()) + + q = q.transpose(1, 2) + k = k.transpose(1, 2) + v = v.transpose(1, 2) + out = comfy.ops.scaled_dot_product_attention(q, k, v, attn_mask=None, dropout_p=0.0, is_causal=False) + + out = out.transpose(1, 2) + if not skip_output_reshape: + return out.values().reshape(-1, heads * (q.shape[-1])) + return out.values() + + +def _validate_split_cu_seqlens(name, cu_seqlens, token_count): + if cu_seqlens.dtype not in (torch.int32, torch.int64): + raise ValueError(f"{name} must use an integer dtype") + if cu_seqlens.ndim != 1 or cu_seqlens.numel() < 2: + raise ValueError(f"{name} must be a 1D tensor with at least two offsets") + if cu_seqlens[0].item() != 0: + raise ValueError(f"{name} must start at 0") + if (cu_seqlens[1:] <= cu_seqlens[:-1]).any().item(): + raise ValueError(f"{name} must be strictly increasing") + if cu_seqlens[-1].item() != token_count: + raise ValueError(f"{name} does not match token count") + + +def _split_indices(cu_seqlens): + return cu_seqlens[1:-1].to(device="cpu", dtype=torch.long) + + +def var_attention_pytorch_split(q, k, v, heads, cu_seqlens_q, cu_seqlens_k, skip_reshape=False, skip_output_reshape=False): + q, k, v, head_dim = _var_attention_qkv(q, k, v, heads, skip_reshape) + + _validate_split_cu_seqlens("cu_seqlens_q", cu_seqlens_q, q.shape[0]) + _validate_split_cu_seqlens("cu_seqlens_k", cu_seqlens_k, k.shape[0]) + if cu_seqlens_k[-1].item() != v.shape[0]: + raise ValueError("cu_seqlens_k does not match v token count") + + q_split_indices = _split_indices(cu_seqlens_q) + k_split_indices = _split_indices(cu_seqlens_k) + q_splits = torch.tensor_split(q, q_split_indices, dim=0) + k_splits = torch.tensor_split(k, k_split_indices, dim=0) + v_splits = torch.tensor_split(v, k_split_indices, dim=0) + if len(q_splits) != len(k_splits) or len(q_splits) != len(v_splits): + raise ValueError("cu_seqlens_q and cu_seqlens_k must describe the same sequence count") + + out = [] + for q_i, k_i, v_i in zip(q_splits, k_splits, v_splits): + q_i = q_i.permute(1, 0, 2).unsqueeze(0) + k_i = k_i.permute(1, 0, 2).unsqueeze(0) + v_i = v_i.permute(1, 0, 2).unsqueeze(0) + out_i = comfy.ops.scaled_dot_product_attention(q_i, k_i, v_i, attn_mask=None, dropout_p=0.0, is_causal=False) + out.append(out_i.squeeze(0).permute(1, 0, 2)) + + out = torch.cat(out, dim=0) + return _var_attention_output(out, heads, head_dim, skip_output_reshape) + +@torch._dynamo.disable +def var_attention_sage(q, k, v, heads, cu_seqlens_q, cu_seqlens_k, *args, skip_reshape=False, skip_output_reshape=False, **kwargs): + if not SAGE_ATTENTION_VARLEN_IS_AVAILABLE: + return var_attention_pytorch( + q, + k, + v, + heads, + cu_seqlens_q, + cu_seqlens_k, + skip_reshape=skip_reshape, + skip_output_reshape=skip_output_reshape, + ) + q, k, v, head_dim = _var_attention_qkv(q, k, v, heads, skip_reshape) + out_dtype = q.dtype + if not (q.dtype == k.dtype == v.dtype): + k = k.to(q.dtype) + v = v.to(q.dtype) + fallback_q, fallback_k, fallback_v = q, k, v + if q.dtype not in (torch.float16, torch.bfloat16): + q = q.to(torch.bfloat16) + k = k.to(torch.bfloat16) + v = v.to(torch.bfloat16) + sm_scale = kwargs.get("softmax_scale") + if sm_scale is None: + sm_scale = 1.0 / math.sqrt(head_dim) + try: + out = sageattn_varlen( + q, + k, + v, + cu_seqlens_q.int(), + cu_seqlens_k.int(), + _var_attention_max_seqlen(cu_seqlens_q), + _var_attention_max_seqlen(cu_seqlens_k), + kwargs.get("causal", False), + sm_scale, + ) + except Exception as e: + logging.error("Error running sage var-len attention: %s, using pytorch var-len attention instead.", e) + out = var_attention_pytorch( + fallback_q, + fallback_k, + fallback_v, + heads, + cu_seqlens_q, + cu_seqlens_k, + skip_reshape=True, + skip_output_reshape=skip_output_reshape, + ) + if out.dtype != out_dtype: + out = out.to(out_dtype) + return out + if out.dtype != out_dtype: + out = out.to(out_dtype) + return _var_attention_output(out, heads, head_dim, skip_output_reshape) + + +@torch._dynamo.disable +def var_attention_sage3(q, k, v, heads, cu_seqlens_q, cu_seqlens_k, *args, skip_reshape=False, skip_output_reshape=False, **kwargs): + if not SAGE_ATTENTION3_IS_AVAILABLE: + if SAGE_ATTENTION_VARLEN_IS_AVAILABLE: + return var_attention_sage( + q, + k, + v, + heads, + cu_seqlens_q, + cu_seqlens_k, + skip_reshape=skip_reshape, + skip_output_reshape=skip_output_reshape, + **kwargs, + ) + return var_attention_pytorch( + q, + k, + v, + heads, + cu_seqlens_q, + cu_seqlens_k, + skip_reshape=skip_reshape, + skip_output_reshape=skip_output_reshape, + ) + q, k, v, head_dim = _var_attention_qkv(q, k, v, heads, skip_reshape) + seq_lens_q = cu_seqlens_q[1:] - cu_seqlens_q[:-1] + seq_lens_k = cu_seqlens_k[1:] - cu_seqlens_k[:-1] + uniform_q = bool((seq_lens_q == seq_lens_q[0]).all().item()) + uniform_k = bool((seq_lens_k == seq_lens_k[0]).all().item()) + if not (uniform_q and uniform_k and seq_lens_q[0] == seq_lens_k[0]): + if SAGE_ATTENTION_VARLEN_IS_AVAILABLE: + return var_attention_sage( + q, + k, + v, + heads, + cu_seqlens_q, + cu_seqlens_k, + skip_reshape=True, + skip_output_reshape=skip_output_reshape, + **kwargs, + ) + return var_attention_pytorch( + q, + k, + v, + heads, + cu_seqlens_q, + cu_seqlens_k, + skip_reshape=True, + skip_output_reshape=skip_output_reshape, + ) + out_dtype = q.dtype + if not (q.dtype == k.dtype == v.dtype): + k = k.to(q.dtype) + v = v.to(q.dtype) + fallback_q, fallback_k, fallback_v = q, k, v + if q.dtype not in (torch.float16, torch.bfloat16): + q = q.to(torch.bfloat16) + k = k.to(torch.bfloat16) + v = v.to(torch.bfloat16) + batch_size = len(cu_seqlens_q) - 1 + seq_len = int(seq_lens_q[0].item()) + q = q.view(batch_size, seq_len, heads, head_dim).transpose(1, 2) + k = k.view(batch_size, seq_len, heads, head_dim).transpose(1, 2) + v = v.view(batch_size, seq_len, heads, head_dim).transpose(1, 2) + try: + out = sageattn3_blackwell(q, k, v, is_causal=kwargs.get("causal", False)) + except Exception as e: + logging.error("Error running SageAttention3 var-len attention: %s, using fallback var-len attention instead.", e) + if SAGE_ATTENTION_VARLEN_IS_AVAILABLE: + return var_attention_sage( + fallback_q, + fallback_k, + fallback_v, + heads, + cu_seqlens_q, + cu_seqlens_k, + skip_reshape=True, + skip_output_reshape=skip_output_reshape, + **kwargs, + ) + return var_attention_pytorch( + fallback_q, + fallback_k, + fallback_v, + heads, + cu_seqlens_q, + cu_seqlens_k, + skip_reshape=True, + skip_output_reshape=skip_output_reshape, + ) + out = out.transpose(1, 2).reshape(-1, heads, head_dim).contiguous() + if out.dtype != out_dtype: + out = out.to(out_dtype) + return _var_attention_output(out, heads, head_dim, skip_output_reshape) + + +@torch._dynamo.disable +def var_attention_flash(q, k, v, heads, cu_seqlens_q, cu_seqlens_k, *args, skip_reshape=False, skip_output_reshape=False, **kwargs): + if not FLASH_ATTENTION_VARLEN_IS_AVAILABLE: + return var_attention_pytorch( + q, + k, + v, + heads, + cu_seqlens_q, + cu_seqlens_k, + skip_reshape=skip_reshape, + skip_output_reshape=skip_output_reshape, + ) + q, k, v, head_dim = _var_attention_qkv(q, k, v, heads, skip_reshape) + max_seqlen_q = _var_attention_max_seqlen(cu_seqlens_q) + max_seqlen_k = _var_attention_max_seqlen(cu_seqlens_k) + try: + out = flash_attn_varlen_func( + q=q, + k=k, + v=v, + cu_seqlens_q=cu_seqlens_q.int(), + cu_seqlens_k=cu_seqlens_k.int(), + max_seqlen_q=max_seqlen_q, + max_seqlen_k=max_seqlen_k, + dropout_p=kwargs.get("dropout_p", 0.0), + causal=kwargs.get("causal", False), + deterministic=torch.are_deterministic_algorithms_enabled(), + ) + except Exception as e: + logging.error("Error running Flash Attention var-len attention: %s, using pytorch var-len attention instead.", e) + return var_attention_pytorch( + q, + k, + v, + heads, + cu_seqlens_q, + cu_seqlens_k, + skip_reshape=True, + skip_output_reshape=skip_output_reshape, + ) + return _var_attention_output(out, heads, head_dim, skip_output_reshape) + + +@torch._dynamo.disable +def var_attention_flash3(q, k, v, heads, cu_seqlens_q, cu_seqlens_k, *args, skip_reshape=False, skip_output_reshape=False, **kwargs): + if not FLASH_ATTENTION3_IS_AVAILABLE: + return var_attention_pytorch( + q, + k, + v, + heads, + cu_seqlens_q, + cu_seqlens_k, + skip_reshape=skip_reshape, + skip_output_reshape=skip_output_reshape, + ) + q, k, v, head_dim = _var_attention_qkv(q, k, v, heads, skip_reshape) + max_seqlen_q = _var_attention_max_seqlen(cu_seqlens_q) + max_seqlen_k = _var_attention_max_seqlen(cu_seqlens_k) + try: + out = flash_attn3_varlen_func( + q=q, + k=k, + v=v, + cu_seqlens_q=cu_seqlens_q.int(), + cu_seqlens_k=cu_seqlens_k.int(), + max_seqlen_q=max_seqlen_q, + max_seqlen_k=max_seqlen_k, + seqused_q=None, + seqused_k=None, + softmax_scale=kwargs.get("softmax_scale"), + causal=kwargs.get("causal", False), + deterministic=torch.are_deterministic_algorithms_enabled(), + ) + except Exception as e: + logging.error("Error running Flash Attention 3 var-len attention: %s, using pytorch var-len attention instead.", e) + return var_attention_pytorch( + q, + k, + v, + heads, + cu_seqlens_q, + cu_seqlens_k, + skip_reshape=True, + skip_output_reshape=skip_output_reshape, + ) + if isinstance(out, tuple): + out = out[0] + return _var_attention_output(out, heads, head_dim, skip_output_reshape) + + +@torch._dynamo.disable +def var_attention_sub_quad(q, k, v, heads, cu_seqlens_q, cu_seqlens_k, *args, skip_reshape=False, skip_output_reshape=False, **kwargs): + return var_attention_pytorch( + q, + k, + v, + heads, + cu_seqlens_q, + cu_seqlens_k, + skip_reshape=skip_reshape, + skip_output_reshape=skip_output_reshape, + ) + + +@torch._dynamo.disable +def var_attention_split(q, k, v, heads, cu_seqlens_q, cu_seqlens_k, *args, skip_reshape=False, skip_output_reshape=False, **kwargs): + return var_attention_pytorch_split( + q, + k, + v, + heads, + cu_seqlens_q, + cu_seqlens_k, + skip_reshape=skip_reshape, + skip_output_reshape=skip_output_reshape, + ) + + +optimized_var_attention = var_attention_pytorch optimized_attention = attention_basic if model_management.sage_attention_enabled(): logging.info("Using sage attention") optimized_attention = attention_sage + if SAGE_ATTENTION3_IS_AVAILABLE and _use_blackwell_attention(): + logging.info("Using SageAttention3 for variable-length attention") + optimized_var_attention = var_attention_sage3 + elif SAGE_ATTENTION_VARLEN_IS_AVAILABLE: + logging.info("Using SageAttention for variable-length attention") + optimized_var_attention = var_attention_sage + else: + logging.info("Using pytorch attention for variable-length attention") + optimized_var_attention = var_attention_pytorch elif model_management.flash_attention_enabled(): logging.info("Using Flash Attention") optimized_attention = attention_flash + if FLASH_ATTENTION_VARLEN_IS_AVAILABLE and model_management.get_torch_device().type == "cuda": + logging.info("Using Flash Attention 2 for variable-length attention") + optimized_var_attention = var_attention_flash + else: + logging.info("Using pytorch attention for variable-length attention") + optimized_var_attention = var_attention_pytorch elif model_management.xformers_enabled(): logging.info("Using xformers attention") optimized_attention = attention_xformers elif model_management.pytorch_attention_enabled(): logging.info("Using pytorch attention") optimized_attention = attention_pytorch + optimized_var_attention = var_attention_pytorch else: if args.use_split_cross_attention: logging.info("Using split optimization for attention") optimized_attention = attention_split + optimized_var_attention = var_attention_split else: logging.info("Using sub quadratic optimization for attention, if you have memory or speed issues try using: --use-split-cross-attention") optimized_attention = attention_sub_quad + optimized_var_attention = var_attention_sub_quad optimized_attention_masked = optimized_attention @@ -764,15 +1193,25 @@ optimized_attention_masked = optimized_attention # register core-supported attention functions if SAGE_ATTENTION_IS_AVAILABLE: register_attention_function("sage", attention_sage) +if SAGE_ATTENTION_VARLEN_IS_AVAILABLE: + register_attention_function("var_attention_sage", var_attention_sage) if SAGE_ATTENTION3_IS_AVAILABLE: register_attention_function("sage3", attention3_sage) + register_attention_function("var_attention_sage3", var_attention_sage3) if FLASH_ATTENTION_IS_AVAILABLE: register_attention_function("flash", attention_flash) +if FLASH_ATTENTION_VARLEN_IS_AVAILABLE: + register_attention_function("var_attention_flash", var_attention_flash) +if FLASH_ATTENTION3_IS_AVAILABLE: + register_attention_function("var_attention_flash3", var_attention_flash3) if model_management.xformers_enabled(): register_attention_function("xformers", attention_xformers) register_attention_function("pytorch", attention_pytorch) +register_attention_function("var_attention_pytorch", var_attention_pytorch) register_attention_function("sub_quad", attention_sub_quad) +register_attention_function("var_attention_sub_quad", var_attention_sub_quad) register_attention_function("split", attention_split) +register_attention_function("var_attention_split", var_attention_split) def optimized_attention_for_device(device, mask=False, small_input=False): @@ -1209,5 +1648,3 @@ class SpatialVideoTransformer(SpatialTransformer): x = self.proj_out(x) out = x + x_in return out - - diff --git a/comfy/ldm/modules/diffusionmodules/model.py b/comfy/ldm/modules/diffusionmodules/model.py index fcbaa074f..235df0b83 100644 --- a/comfy/ldm/modules/diffusionmodules/model.py +++ b/comfy/ldm/modules/diffusionmodules/model.py @@ -13,6 +13,7 @@ if model_management.xformers_enabled_vae(): import xformers import xformers.ops + def torch_cat_if_needed(xl, dim): xl = [x for x in xl if x is not None and x.shape[dim] > 0] if len(xl) > 1: @@ -22,7 +23,8 @@ def torch_cat_if_needed(xl, dim): else: return None -def get_timestep_embedding(timesteps, embedding_dim): + +def get_timestep_embedding(timesteps, embedding_dim, flip_sin_to_cos=False, downscale_freq_shift=1): """ This matches the implementation in Denoising Diffusion Probabilistic Models: From Fairseq. @@ -33,11 +35,13 @@ def get_timestep_embedding(timesteps, embedding_dim): assert len(timesteps.shape) == 1 half_dim = embedding_dim // 2 - emb = math.log(10000) / (half_dim - 1) + emb = math.log(10000) / (half_dim - downscale_freq_shift) emb = torch.exp(torch.arange(half_dim, dtype=torch.float32) * -emb) emb = emb.to(device=timesteps.device) emb = timesteps.float()[:, None] * emb[None, :] emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=1) + if flip_sin_to_cos: + emb = torch.cat([emb[:, half_dim:], emb[:, :half_dim]], dim=-1) if embedding_dim % 2 == 1: # zero pad emb = torch.nn.functional.pad(emb, (0,1,0,0)) return emb diff --git a/comfy/ldm/seedvr/model.py b/comfy/ldm/seedvr/model.py new file mode 100644 index 000000000..32a1c2134 --- /dev/null +++ b/comfy/ldm/seedvr/model.py @@ -0,0 +1,1629 @@ +from dataclasses import dataclass +from typing import Optional, Tuple, Union, List, Dict, Any, Callable +import einops +from einops import rearrange +import torch.nn.functional as F +from math import ceil, pi +import torch +from itertools import chain +from comfy.ldm.modules.diffusionmodules.model import get_timestep_embedding +from comfy.ldm.modules.attention import optimized_var_attention +from torch.nn.modules.utils import _triple +from torch import nn +import math +from comfy.ldm.flux.math import apply_rope1 +import comfy.model_management +import numbers + +def _torch_float8_types(): + return tuple( + getattr(torch, name) + for name in ( + "float8_e4m3fn", + "float8_e4m3fnuz", + "float8_e5m2", + "float8_e5m2fnuz", + "float8_e8m0fnu", + ) + if hasattr(torch, name) + ) + +class CustomRMSNorm(nn.Module): + + def __init__(self, normalized_shape, eps=1e-5, elementwise_affine=True, device=None, dtype=None): + super(CustomRMSNorm, self).__init__() + + if isinstance(normalized_shape, numbers.Integral): + normalized_shape = (normalized_shape,) + self.normalized_shape = torch.Size(normalized_shape) + self.eps = eps + self.elementwise_affine = elementwise_affine + + if self.elementwise_affine: + self.weight = nn.Parameter(torch.ones(*normalized_shape, device=device, dtype=dtype)) + else: + self.register_parameter('weight', None) + + def forward(self, input): + + dims = tuple(range(-len(self.normalized_shape), 0)) + + variance = input.pow(2).mean(dim=dims, keepdim=True) + rms = torch.sqrt(variance + self.eps) + + normalized = input / rms + + if self.elementwise_affine: + return normalized * self.weight.to(input.dtype) + return normalized + +class Cache: + def __init__(self, disable=False, prefix="", cache=None): + self.cache = cache if cache is not None else {} + self.disable = disable + self.prefix = prefix + + def __call__(self, key: str, fn: Callable): + if self.disable: + return fn() + + key = self.prefix + key + try: + result = self.cache[key] + except KeyError: + result = fn() + self.cache[key] = result + return result + + def namespace(self, namespace: str): + return Cache( + disable=self.disable, + prefix=self.prefix + namespace + ".", + cache=self.cache, + ) + + def get(self, key: str): + key = self.prefix + key + return self.cache[key] + +def repeat_concat( + vid: torch.FloatTensor, # (VL ... c) + txt: torch.FloatTensor, # (TL ... c) + vid_len: torch.LongTensor, # (n*b) + txt_len: torch.LongTensor, # (b) + txt_repeat: List, # (n) +) -> torch.FloatTensor: # (L ... c) + vid = torch.split(vid, vid_len.tolist()) + txt = torch.split(txt, txt_len.tolist()) + txt = [[x] * n for x, n in zip(txt, txt_repeat)] + txt = list(chain(*txt)) + return torch.cat(list(chain(*zip(vid, txt)))) + +def concat( + vid: torch.FloatTensor, # (VL ... c) + txt: torch.FloatTensor, # (TL ... c) + vid_len: torch.LongTensor, # (b) + txt_len: torch.LongTensor, # (b) +) -> torch.FloatTensor: # (L ... c) + vid = torch.split(vid, vid_len.tolist()) + txt = torch.split(txt, txt_len.tolist()) + return torch.cat(list(chain(*zip(vid, txt)))) + +def concat_idx( + vid_len: torch.LongTensor, # (b) + txt_len: torch.LongTensor, # (b) +) -> Tuple[ + Callable, + Callable, +]: + device = vid_len.device + vid_idx = torch.arange(vid_len.sum(), device=device) + txt_idx = torch.arange(len(vid_idx), len(vid_idx) + txt_len.sum(), device=device) + tgt_idx = concat(vid_idx, txt_idx, vid_len, txt_len) + src_idx = torch.argsort(tgt_idx) + return ( + lambda vid, txt: torch.index_select(torch.cat([vid, txt]), 0, tgt_idx), + lambda all: torch.index_select(all, 0, src_idx).split([len(vid_idx), len(txt_idx)]), + ) + + +def repeat_concat_idx( + vid_len: torch.LongTensor, # (n*b) + txt_len: torch.LongTensor, # (b) + txt_repeat: torch.LongTensor, # (n) +) -> Tuple[ + Callable, + Callable, +]: + device = vid_len.device + vid_idx = torch.arange(vid_len.sum(), device=device) + txt_idx = torch.arange(len(vid_idx), len(vid_idx) + txt_len.sum(), device=device) + txt_repeat_list = txt_repeat.tolist() + tgt_idx = repeat_concat(vid_idx, txt_idx, vid_len, txt_len, txt_repeat) + src_idx = torch.argsort(tgt_idx) + txt_idx_len = len(tgt_idx) - len(vid_idx) + repeat_txt_len = (txt_len * txt_repeat).tolist() + + def unconcat_coalesce(all): + vid_out, txt_out = all[src_idx].split([len(vid_idx), txt_idx_len]) + txt_out_coalesced = [] + for txt, repeat_time in zip(txt_out.split(repeat_txt_len), txt_repeat_list): + txt = txt.reshape(-1, repeat_time, *txt.shape[1:]).mean(1) + txt_out_coalesced.append(txt) + return vid_out, torch.cat(txt_out_coalesced) + + return ( + lambda vid, txt: torch.cat([vid, txt])[tgt_idx], + lambda all: unconcat_coalesce(all), + ) + + +@dataclass +class MMArg: + vid: Any + txt: Any + +def safe_pad_operation(x, padding, mode='constant', value=0.0): + """Safe padding operation that handles Half precision only for problematic modes""" + # Modes qui nécessitent le fix Half precision + problematic_modes = ['replicate', 'reflect', 'circular'] + + if mode in problematic_modes: + try: + return F.pad(x, padding, mode=mode, value=value) + except RuntimeError as e: + if "not implemented for 'Half'" in str(e): + original_dtype = x.dtype + return F.pad(x.float(), padding, mode=mode, value=value).to(original_dtype) + else: + raise e + else: + # Pour 'constant' et autres modes compatibles, pas de fix nécessaire + return F.pad(x, padding, mode=mode, value=value) + + +def get_args(key: str, args: List[Any]) -> List[Any]: + return [getattr(v, key) if isinstance(v, MMArg) else v for v in args] + + +def get_kwargs(key: str, kwargs: Dict[str, Any]) -> Dict[str, Any]: + return {k: getattr(v, key) if isinstance(v, MMArg) else v for k, v in kwargs.items()} + + +def get_window_op(name: str): + if name == "720pwin_by_size_bysize": + return make_720Pwindows_bysize + if name == "720pswin_by_size_bysize": + return make_shifted_720Pwindows_bysize + raise ValueError(f"Unknown windowing method: {name}") + + +# -------------------------------- Windowing -------------------------------- # +def make_720Pwindows_bysize(size: Tuple[int, int, int], num_windows: Tuple[int, int, int]): + t, h, w = size + resized_nt, resized_nh, resized_nw = num_windows + #cal windows under 720p + scale = math.sqrt((45 * 80) / (h * w)) + resized_h, resized_w = round(h * scale), round(w * scale) + wh, ww = ceil(resized_h / resized_nh), ceil(resized_w / resized_nw) # window size. + wt = ceil(min(t, 30) / resized_nt) # window size. + nt, nh, nw = ceil(t / wt), ceil(h / wh), ceil(w / ww) # window size. + return [ + ( + slice(it * wt, min((it + 1) * wt, t)), + slice(ih * wh, min((ih + 1) * wh, h)), + slice(iw * ww, min((iw + 1) * ww, w)), + ) + for iw in range(nw) + if min((iw + 1) * ww, w) > iw * ww + for ih in range(nh) + if min((ih + 1) * wh, h) > ih * wh + for it in range(nt) + if min((it + 1) * wt, t) > it * wt + ] + +def make_shifted_720Pwindows_bysize(size: Tuple[int, int, int], num_windows: Tuple[int, int, int]): + t, h, w = size + resized_nt, resized_nh, resized_nw = num_windows + #cal windows under 720p + scale = math.sqrt((45 * 80) / (h * w)) + resized_h, resized_w = round(h * scale), round(w * scale) + wh, ww = ceil(resized_h / resized_nh), ceil(resized_w / resized_nw) # window size. + wt = ceil(min(t, 30) / resized_nt) # window size. + + st, sh, sw = ( # shift size. + 0.5 if wt < t else 0, + 0.5 if wh < h else 0, + 0.5 if ww < w else 0, + ) + nt, nh, nw = ceil((t - st) / wt), ceil((h - sh) / wh), ceil((w - sw) / ww) # window size. + nt, nh, nw = ( # number of window. + nt + 1 if st > 0 else 1, + nh + 1 if sh > 0 else 1, + nw + 1 if sw > 0 else 1, + ) + return [ + ( + slice(max(int((it - st) * wt), 0), min(int((it - st + 1) * wt), t)), + slice(max(int((ih - sh) * wh), 0), min(int((ih - sh + 1) * wh), h)), + slice(max(int((iw - sw) * ww), 0), min(int((iw - sw + 1) * ww), w)), + ) + for iw in range(nw) + if min(int((iw - sw + 1) * ww), w) > max(int((iw - sw) * ww), 0) + for ih in range(nh) + if min(int((ih - sh + 1) * wh), h) > max(int((ih - sh) * wh), 0) + for it in range(nt) + if min(int((it - st + 1) * wt), t) > max(int((it - st) * wt), 0) + ] + +class RotaryEmbedding(nn.Module): + def __init__( + self, + dim, + custom_freqs = None, + freqs_for = 'lang', + theta = 10000, + max_freq = 10, + num_freqs = 1, + learned_freq = False, + use_xpos = False, + xpos_scale_base = 512, + interpolate_factor = 1., + theta_rescale_factor = 1., + seq_before_head_dim = False, + cache_if_possible = True, + cache_max_seq_len = 8192 + ): + super().__init__() + + theta *= theta_rescale_factor ** (dim / (dim - 2)) + + self.freqs_for = freqs_for + + if exists(custom_freqs): + freqs = custom_freqs + elif freqs_for == 'lang': + freqs = 1. / (theta ** (torch.arange(0, dim, 2)[:(dim // 2)].float() / dim)) + elif freqs_for == 'pixel': + freqs = torch.linspace(1., max_freq / 2, dim // 2) * pi + elif freqs_for == 'constant': + freqs = torch.ones(num_freqs).float() + + self.cache_if_possible = cache_if_possible + self.cache_max_seq_len = cache_max_seq_len + + self.register_buffer('cached_freqs', torch.zeros(cache_max_seq_len, dim), persistent = False) + self.cached_freqs_seq_len = 0 + + self.freqs = nn.Parameter(freqs, requires_grad = learned_freq) + + self.learned_freq = learned_freq + + # dummy for device + + self.register_buffer('dummy', torch.tensor(0), persistent = False) + + # default sequence dimension + + self.seq_before_head_dim = seq_before_head_dim + self.default_seq_dim = -3 if seq_before_head_dim else -2 + + # interpolation factors + + assert interpolate_factor >= 1. + self.interpolate_factor = interpolate_factor + + # xpos + + self.use_xpos = use_xpos + + if not use_xpos: + return + + scale = (torch.arange(0, dim, 2) + 0.4 * dim) / (1.4 * dim) + self.scale_base = xpos_scale_base + + self.register_buffer('scale', scale, persistent = False) + self.register_buffer('cached_scales', torch.zeros(cache_max_seq_len, dim), persistent = False) + self.cached_scales_seq_len = 0 + + # add apply_rotary_emb as static method + + self.apply_rotary_emb = staticmethod(apply_rotary_emb) + + @property + def device(self): + return self.dummy.device + + def get_axial_freqs( + self, + *dims, + offsets = None + ): + Colon = slice(None) + all_freqs = [] + + # handle offset + + if exists(offsets): + assert len(offsets) == len(dims) + + for ind, dim in enumerate(dims): + + offset = 0 + if exists(offsets): + offset = offsets[ind] + + if self.freqs_for == 'pixel': + pos = torch.linspace(-1, 1, steps = dim, device = self.device) + else: + pos = torch.arange(dim, device = self.device) + + pos = pos + offset + + freqs = self.forward(pos, seq_len = dim) + + all_axis = [None] * len(dims) + all_axis[ind] = Colon + + new_axis_slice = (Ellipsis, *all_axis, Colon) + all_freqs.append(freqs[new_axis_slice]) + + # concat all freqs + + all_freqs = torch.broadcast_tensors(*all_freqs) + return torch.cat(all_freqs, dim = -1) + + def forward( + self, + t, + seq_len: int | None = None, + offset = 0 + ): + should_cache = ( + self.cache_if_possible and + not self.learned_freq and + exists(seq_len) and + self.freqs_for != 'pixel' and + (offset + seq_len) <= self.cache_max_seq_len + ) + + if ( + should_cache and \ + exists(self.cached_freqs) and \ + (offset + seq_len) <= self.cached_freqs_seq_len + ): + return self.cached_freqs[offset:(offset + seq_len)].detach() + + freqs = self.freqs + + freqs = torch.einsum('..., f -> ... f', t.type(freqs.dtype), freqs) + freqs = einops.repeat(freqs, '... n -> ... (n r)', r = 2) + + if should_cache and offset == 0: + self.cached_freqs[:seq_len] = freqs.detach() + self.cached_freqs_seq_len = seq_len + + return freqs + +class RotaryEmbeddingBase(nn.Module): + def __init__(self, dim: int, rope_dim: int): + super().__init__() + self.rope = RotaryEmbedding( + dim=dim // rope_dim, + freqs_for="pixel", + max_freq=256, + ) + freqs = self.rope.freqs + del self.rope.freqs + self.rope.register_buffer("freqs", freqs.data) + + def get_axial_freqs(self, *dims): + return self.rope.get_axial_freqs(*dims) + + +class RotaryEmbedding3d(RotaryEmbeddingBase): + def __init__(self, dim: int): + super().__init__(dim, rope_dim=3) + self.mm = False + + def forward( + self, + q: torch.FloatTensor, # b h l d + k: torch.FloatTensor, # b h l d + size: Tuple[int, int, int], + ) -> Tuple[ + torch.FloatTensor, + torch.FloatTensor, + ]: + T, H, W = size + freqs = self.get_axial_freqs(T, H, W) + q = rearrange(q, "b h (T H W) d -> b h T H W d", T=T, H=H, W=W) + k = rearrange(k, "b h (T H W) d -> b h T H W d", T=T, H=H, W=W) + q = apply_rotary_emb(freqs, q.float()).to(q.dtype) + k = apply_rotary_emb(freqs, k.float()).to(k.dtype) + q = rearrange(q, "b h T H W d -> b h (T H W) d") + k = rearrange(k, "b h T H W d -> b h (T H W) d") + return q, k + + +class NaRotaryEmbedding3d(RotaryEmbedding3d): + def forward( + self, + q: torch.FloatTensor, + k: torch.FloatTensor, + shape: torch.LongTensor, + cache: Cache, + ) -> Tuple[ + torch.FloatTensor, + torch.FloatTensor, + ]: + freqs = cache("rope_freqs_3d", lambda: self.get_freqs(shape)) + freqs = freqs.to(device=q.device) + q = rearrange(q, "L h d -> h L d") + k = rearrange(k, "L h d -> h L d") + q = _apply_rope1_partial(q, freqs) + k = _apply_rope1_partial(k, freqs) + q = rearrange(q, "h L d -> L h d") + k = rearrange(k, "h L d -> L h d") + return q, k + + @torch._dynamo.disable + def get_freqs( + self, + shape: torch.LongTensor, + ) -> torch.Tensor: + freq_list = [] + for f, h, w in shape.tolist(): + freqs = self.get_axial_freqs(f, h, w) + freq_list.append(freqs.view(-1, freqs.size(-1))) + return _to_flux_freqs_cis(torch.cat(freq_list, dim=0)) + + +class MMRotaryEmbeddingBase(RotaryEmbeddingBase): + def __init__(self, dim: int, rope_dim: int): + super().__init__(dim, rope_dim) + self.rope = RotaryEmbedding( + dim=dim // rope_dim, + freqs_for="lang", + theta=10000, + cache_if_possible=False, + ) + freqs = self.rope.freqs + del self.rope.freqs + self.rope.register_buffer("freqs", freqs.data) + self.mm = True + +def slice_at_dim(t, dim_slice: slice, *, dim): + dim += (t.ndim if dim < 0 else 0) + colons = [slice(None)] * t.ndim + colons[dim] = dim_slice + return t[tuple(colons)] + +# rotary embedding helper functions + +def rotate_half(x): + x = rearrange(x, '... (d r) -> ... d r', r = 2) + x1, x2 = x.unbind(dim = -1) + x = torch.stack((-x2, x1), dim = -1) + return rearrange(x, '... d r -> ... (d r)') +def exists(val): + return val is not None + +def apply_rotary_emb( + freqs, + t, + start_index = 0, + scale = 1., + seq_dim = -2, + freqs_seq_dim = None +): + dtype = t.dtype + if not exists(freqs_seq_dim): + if freqs.ndim == 2 or t.ndim == 3: + freqs_seq_dim = 0 + + if t.ndim == 3 or exists(freqs_seq_dim): + seq_len = t.shape[seq_dim] + freqs = slice_at_dim(freqs, slice(-seq_len, None), dim = freqs_seq_dim) + + rot_feats = freqs.shape[-1] + end_index = start_index + rot_feats + + t_left = t[..., :start_index] + t_middle = t[..., start_index:end_index] + t_right = t[..., end_index:] + + angles = freqs.to(t_middle.device)[..., ::2] + cos = torch.cos(angles) * scale + sin = torch.sin(angles) * scale + + col0 = torch.stack([cos, sin], dim=-1) + col1 = torch.stack([-sin, cos], dim=-1) + freqs_mat = torch.stack([col0, col1], dim=-1) + + t_middle_out = apply_rope1(t_middle, freqs_mat) + out = torch.cat((t_left, t_middle_out, t_right), dim=-1) + return out.type(dtype) + +def _to_flux_freqs_cis(freqs_interleaved: torch.Tensor) -> torch.Tensor: + """Convert lucidrains-interleaved freqs `[..., d]` (`[θ0, θ0, θ1, θ1, ...]` + from `RotaryEmbedding.forward`'s `repeat(freqs, '... n -> ... (n r)', r=2)`) + into flux-canonical `freqs_cis` of shape `[..., d/2, 2, 2]` with the + `cos/-sin/sin/cos` rotation matrix baked in. Output dtype is fp32 to + match `comfy/ldm/flux/math.py:rope` precision; `apply_rope1` consumes + the matrix layout via `freqs_cis[..., 0]` (column 0) and + `freqs_cis[..., 1]` (column 1) of the 2x2 rotation matrix. + """ + angles = freqs_interleaved[..., ::2].float() + cos = torch.cos(angles) + sin = torch.sin(angles) + out = torch.stack([cos, -sin, sin, cos], dim=-1) + return rearrange(out, "... d (i j) -> ... d i j", i=2, j=2) + + +_ROPE1_PARTIAL_CHUNK_TOKENS = 4096 +SEEDVR2_7B_MLP_CHUNK = 8192 + + +def _apply_rope1_partial(t: torch.Tensor, freqs_cis: torch.Tensor) -> torch.Tensor: + """Apply ``apply_rope1`` to the leading ``rot_d = 2 * freqs_cis.shape[-3]`` + components of ``t``'s last dim, passing through the remaining dims + untouched in-place for inference tensors. Training tensors are cloned + before slice assignment to preserve autograd correctness. Mirrors the partial-rope contract of the legacy + ``apply_rotary_emb`` wrapper at line 470 (``t_left``/``t_middle``/``t_right`` + split). For SeedVR2-3B this matters because ``rope_dim=128`` integer- + divides into 3 axes as ``128 // 3 = 42`` per-axis, total ``42 * 3 = 126``; + head_dim is 128, so the trailing 2 dims are unrotated. The fast path + triggers when ``rot_d == t.shape[-1]`` (e.g. test rigs where dim is + chosen divisible by 6) and avoids the cat entirely. + """ + out = t.clone() if t.requires_grad or comfy.model_management.in_training else t + rot_d = 2 * freqs_cis.shape[-3] + seq_len = out.shape[-2] + for start in range(0, seq_len, _ROPE1_PARTIAL_CHUNK_TOKENS): + end = min(start + _ROPE1_PARTIAL_CHUNK_TOKENS, seq_len) + freqs_chunk = freqs_cis[start:end] + if rot_d == out.shape[-1]: + out[..., start:end, :] = apply_rope1(out[..., start:end, :], freqs_chunk).to(out.dtype) + else: + out[..., start:end, :rot_d] = apply_rope1(out[..., start:end, :rot_d], freqs_chunk).to(out.dtype) + return out + + +class NaMMRotaryEmbedding3d(MMRotaryEmbeddingBase): + def __init__(self, dim: int): + super().__init__(dim, rope_dim=3) + + def forward( + self, + vid_q: torch.FloatTensor, # L h d + vid_k: torch.FloatTensor, # L h d + vid_shape: torch.LongTensor, # B 3 + txt_q: torch.FloatTensor, # L h d + txt_k: torch.FloatTensor, # L h d + txt_shape: torch.LongTensor, # B 1 + cache: Cache, + ) -> Tuple[ + torch.FloatTensor, + torch.FloatTensor, + torch.FloatTensor, + torch.FloatTensor, + ]: + vid_freqs, txt_freqs = cache( + "mmrope_freqs_3d", + lambda: self.get_freqs(vid_shape, txt_shape), + ) + target_device = vid_q.device + if vid_freqs.device != target_device: + vid_freqs = vid_freqs.to(target_device) + if txt_freqs.device != target_device: + txt_freqs = txt_freqs.to(target_device) + vid_q = rearrange(vid_q, "L h d -> h L d") + vid_k = rearrange(vid_k, "L h d -> h L d") + vid_q = _apply_rope1_partial(vid_q, vid_freqs) + vid_k = _apply_rope1_partial(vid_k, vid_freqs) + vid_q = rearrange(vid_q, "h L d -> L h d") + vid_k = rearrange(vid_k, "h L d -> L h d") + + txt_q = rearrange(txt_q, "L h d -> h L d") + txt_k = rearrange(txt_k, "L h d -> h L d") + txt_q = _apply_rope1_partial(txt_q, txt_freqs) + txt_k = _apply_rope1_partial(txt_k, txt_freqs) + txt_q = rearrange(txt_q, "h L d -> L h d") + txt_k = rearrange(txt_k, "h L d -> L h d") + return vid_q, vid_k, txt_q, txt_k + + @torch._dynamo.disable # Disable compilation: .tolist() is data-dependent and causes graph breaks + def get_freqs( + self, + vid_shape: torch.LongTensor, + txt_shape: torch.LongTensor, + ) -> Tuple[ + torch.Tensor, + torch.Tensor, + ]: + + # Calculate actual max dimensions needed for this batch + max_temporal = 0 + max_height = 0 + max_width = 0 + max_txt_len = 0 + + for (f, h, w), l in zip(vid_shape.tolist(), txt_shape[:, 0].tolist()): + max_temporal = max(max_temporal, l + f) # Need up to l+f for temporal + max_height = max(max_height, h) + max_width = max(max_width, w) + max_txt_len = max(max_txt_len, l) + + autocast_device = "cuda" if torch.cuda.is_available() else "cpu" + with torch.amp.autocast(autocast_device, enabled=False): + vid_freqs = self.get_axial_freqs( + max_temporal + 16, + max_height + 4, + max_width + 4, + ).float() + txt_freqs = self.get_axial_freqs(max_txt_len + 16) + + # Now slice as before + vid_freq_list, txt_freq_list = [], [] + for (f, h, w), l in zip(vid_shape.tolist(), txt_shape[:, 0].tolist()): + vid_freq = vid_freqs[l : l + f, :h, :w].reshape(-1, vid_freqs.size(-1)) + txt_freq = txt_freqs[:l].repeat(1, 3).reshape(-1, vid_freqs.size(-1)) + vid_freq_list.append(vid_freq) + txt_freq_list.append(txt_freq) + vid_freqs_interleaved = torch.cat(vid_freq_list, dim=0) + txt_freqs_interleaved = torch.cat(txt_freq_list, dim=0) + + # Convert from lucidrains-interleaved layout `[θ0, θ0, θ1, θ1, ...]` + # (produced by `repeat(freqs, '... n -> ... (n r)', r=2)` in the + # upstream `RotaryEmbedding.forward`) to flux-canonical `freqs_cis` + # in shape `[..., d/2, 2, 2]` with `cos/-sin/sin/cos` baked in. + # Mirrors `comfy/ldm/flux/math.py:rope` (line 27) so the trailing + # 2x2 is the per-frequency rotation matrix that + # `comfy.ldm.flux.math.apply_rope1` expects. + return _to_flux_freqs_cis(vid_freqs_interleaved), _to_flux_freqs_cis(txt_freqs_interleaved) + +class MMModule(nn.Module): + def __init__( + self, + module: Callable[..., nn.Module], + *args, + shared_weights: bool = False, + vid_only: bool = False, + **kwargs, + ): + super().__init__() + self.shared_weights = shared_weights + self.vid_only = vid_only + if self.shared_weights: + assert get_args("vid", args) == get_args("txt", args) + assert get_kwargs("vid", kwargs) == get_kwargs("txt", kwargs) + self.all = module(*get_args("vid", args), **get_kwargs("vid", kwargs)) + else: + self.vid = module(*get_args("vid", args), **get_kwargs("vid", kwargs)) + self.txt = ( + module(*get_args("txt", args), **get_kwargs("txt", kwargs)) + if not vid_only + else None + ) + + def forward( + self, + vid: torch.FloatTensor, + txt: torch.FloatTensor, + *args, + **kwargs, + ) -> Tuple[ + torch.FloatTensor, + torch.FloatTensor, + ]: + vid_module = self.vid if not self.shared_weights else self.all + vid = vid_module(vid, *get_args("vid", args), **get_kwargs("vid", kwargs)) + if not self.vid_only: + txt_module = self.txt if not self.shared_weights else self.all + txt = txt.to(device=vid.device, dtype=vid.dtype) + txt = txt_module(txt, *get_args("txt", args), **get_kwargs("txt", kwargs)) + return vid, txt + +def get_na_rope(rope_type: Optional[str], dim: int): + if rope_type is None: + return None + if rope_type == "rope3d": + return NaRotaryEmbedding3d(dim=dim) + if rope_type == "mmrope3d": + return NaMMRotaryEmbedding3d(dim=dim) + +class NaMMAttention(nn.Module): + def __init__( + self, + vid_dim: int, + txt_dim: int, + heads: int, + head_dim: int, + qk_bias: bool, + qk_norm, + qk_norm_eps: float, + rope_type: Optional[str], + rope_dim: int, + shared_weights: bool, + device, dtype, operations, + **kwargs, + ): + super().__init__() + dim = MMArg(vid_dim, txt_dim) + self.heads = heads + inner_dim = heads * head_dim + qkv_dim = inner_dim * 3 + self.head_dim = head_dim + self.proj_qkv = MMModule( + operations.Linear, dim, qkv_dim, bias=qk_bias, shared_weights=shared_weights, device=device, dtype=dtype + ) + self.proj_out = MMModule(operations.Linear, inner_dim, dim, shared_weights=shared_weights, device=device, dtype=dtype) + self.norm_q = MMModule( + qk_norm, + normalized_shape=head_dim, + eps=qk_norm_eps, + elementwise_affine=True, + shared_weights=shared_weights, + device=device, dtype=dtype + ) + self.norm_k = MMModule( + qk_norm, + normalized_shape=head_dim, + eps=qk_norm_eps, + elementwise_affine=True, + shared_weights=shared_weights, + device=device, dtype=dtype + ) + + + self.rope = get_na_rope(rope_type=rope_type, dim=rope_dim) + + def forward(self): + pass + +def window( + hid: torch.FloatTensor, # (L c) + hid_shape: torch.LongTensor, # (b n) + window_fn: Callable[[torch.Tensor], List[torch.Tensor]], +): + hid = unflatten(hid, hid_shape) + hid = list(map(window_fn, hid)) + hid_windows = torch.tensor(list(map(len, hid)), device=hid_shape.device) + hid, hid_shape = flatten(list(chain(*hid))) + return hid, hid_shape, hid_windows + +def window_idx( + hid_shape: torch.LongTensor, # (b n) + window_fn: Callable[[torch.Tensor], List[torch.Tensor]], +): + hid_idx = torch.arange(hid_shape.prod(-1).sum(), device=hid_shape.device).unsqueeze(-1) + tgt_idx, tgt_shape, tgt_windows = window(hid_idx, hid_shape, window_fn) + tgt_idx = tgt_idx.squeeze(-1) + src_idx = torch.argsort(tgt_idx) + return ( + lambda hid: torch.index_select(hid, 0, tgt_idx), + lambda hid: torch.index_select(hid, 0, src_idx), + tgt_shape, + tgt_windows, + ) + +class NaSwinAttention(NaMMAttention): + def __init__( + self, + *args, + window: Union[int, Tuple[int, int, int]], + window_method: bool, # shifted or not + **kwargs, + ): + super().__init__(*args, **kwargs) + self.version_7b = kwargs.get("version", False) + self.window = _triple(window) + self.window_method = window_method + assert all(map(lambda v: isinstance(v, int) and v >= 0, self.window)) + + self.window_op = get_window_op(window_method) + + def forward( + self, + vid: torch.FloatTensor, # l c + txt: torch.FloatTensor, # l c + vid_shape: torch.LongTensor, # b 3 + txt_shape: torch.LongTensor, # b 1 + cache: Cache, + ) -> Tuple[ + torch.FloatTensor, + torch.FloatTensor, + ]: + + vid_qkv, txt_qkv = self.proj_qkv(vid, txt) + + # re-org the input seq for window attn + cache_win = cache.namespace(f"{self.window_method}_{self.window}_sd3") + + def make_window(x: torch.Tensor): + t, h, w, _ = x.shape + window_slices = self.window_op((t, h, w), self.window) + return [x[st, sh, sw] for (st, sh, sw) in window_slices] + + window_partition, window_reverse, window_shape, window_count = cache_win( + "win_transform", + lambda: window_idx(vid_shape, make_window), + ) + vid_qkv_win = window_partition(vid_qkv) + + vid_qkv_win = rearrange(vid_qkv_win, "l (o h d) -> l o h d", o=3, d=self.head_dim) + txt_qkv = rearrange(txt_qkv, "l (o h d) -> l o h d", o=3, d=self.head_dim) + + vid_q, vid_k, vid_v = vid_qkv_win.unbind(1) + txt_q, txt_k, txt_v = txt_qkv.unbind(1) + + vid_q, txt_q = self.norm_q(vid_q, txt_q) + vid_k, txt_k = self.norm_k(vid_k, txt_k) + + txt_len = cache("txt_len", lambda: txt_shape.prod(-1)) + + vid_len_win = cache_win("vid_len", lambda: window_shape.prod(-1)) + txt_len = txt_len.to(window_count.device) + + # window rope + if self.rope: + if self.version_7b: + vid_q, vid_k = self.rope(vid_q, vid_k, window_shape, cache_win) + elif self.rope.mm: + # repeat text q and k for window mmrope + _, num_h, _ = txt_q.shape + txt_q_repeat = rearrange(txt_q, "l h d -> l (h d)") + txt_q_repeat = unflatten(txt_q_repeat, txt_shape) + txt_q_repeat = [[x] * n for x, n in zip(txt_q_repeat, window_count)] + txt_q_repeat = list(chain(*txt_q_repeat)) + txt_q_repeat, txt_shape_repeat = flatten(txt_q_repeat) + txt_q_repeat = rearrange(txt_q_repeat, "l (h d) -> l h d", h=num_h) + + txt_k_repeat = rearrange(txt_k, "l h d -> l (h d)") + txt_k_repeat = unflatten(txt_k_repeat, txt_shape) + txt_k_repeat = [[x] * n for x, n in zip(txt_k_repeat, window_count)] + txt_k_repeat = list(chain(*txt_k_repeat)) + txt_k_repeat, _ = flatten(txt_k_repeat) + txt_k_repeat = rearrange(txt_k_repeat, "l (h d) -> l h d", h=num_h) + + vid_q, vid_k, txt_q, txt_k = self.rope( + vid_q, vid_k, window_shape, txt_q_repeat, txt_k_repeat, txt_shape_repeat, cache_win + ) + else: + vid_q, vid_k = self.rope(vid_q, vid_k, window_shape, cache_win) + + txt_len_win = cache_win("txt_len", lambda: txt_len.repeat_interleave(window_count)) + all_len_win = cache_win("all_len", lambda: vid_len_win + txt_len_win) + concat_win, unconcat_win = cache_win( + "mm_pnp", lambda: repeat_concat_idx(vid_len_win, txt_len, window_count) + ) + out = optimized_var_attention( + q=concat_win(vid_q, txt_q), + k=concat_win(vid_k, txt_k), + v=concat_win(vid_v, txt_v), + heads=self.heads, skip_reshape=True, skip_output_reshape=True, + cu_seqlens_q=cache_win( + "vid_seqlens_q", lambda: safe_pad_operation(all_len_win.cumsum(0), (1, 0)).int() + ), + cu_seqlens_k=cache_win( + "vid_seqlens_k", lambda: safe_pad_operation(all_len_win.cumsum(0), (1, 0)).int() + ), + ) + vid_out, txt_out = unconcat_win(out) + + vid_out = rearrange(vid_out, "l h d -> l (h d)") + txt_out = rearrange(txt_out, "l h d -> l (h d)") + vid_out = window_reverse(vid_out) + + vid_out, txt_out = self.proj_out(vid_out, txt_out) + + return vid_out, txt_out + +class MLP(nn.Module): + def __init__( + self, + dim: int, + expand_ratio: int, + device, dtype, operations + ): + super().__init__() + self.proj_in = operations.Linear(dim, dim * expand_ratio, device=device, dtype=dtype) + self.act = nn.GELU("tanh") + self.proj_out = operations.Linear(dim * expand_ratio, dim, device=device, dtype=dtype) + + def forward(self, x: torch.FloatTensor) -> torch.FloatTensor: + x = self.proj_in(x) + x = self.act(x) + x = self.proj_out(x) + return x + + +class SwiGLUMLP(nn.Module): + def __init__( + self, + dim: int, + expand_ratio: int, + multiple_of: int = 256, + device=None, dtype=None, operations=None + ): + super().__init__() + hidden_dim = int(2 * dim * expand_ratio / 3) + hidden_dim = multiple_of * ((hidden_dim + multiple_of - 1) // multiple_of) + self.proj_in_gate = operations.Linear(dim, hidden_dim, bias=False, device=device, dtype=dtype) + self.proj_out = operations.Linear(hidden_dim, dim, bias=False, device=device, dtype=dtype) + self.proj_in = operations.Linear(dim, hidden_dim, bias=False, device=device, dtype=dtype) + + def forward(self, x: torch.FloatTensor) -> torch.FloatTensor: + return self.proj_out(F.silu(self.proj_in_gate(x)) * self.proj_in(x)) + +def get_mlp(mlp_type: Optional[str] = "normal"): + # 3b and 7b uses different mlp types + if mlp_type == "normal": + return MLP + elif mlp_type == "swiglu": + return SwiGLUMLP + +class NaMMSRTransformerBlock(nn.Module): + def __init__( + self, + *, + vid_dim: int, + txt_dim: int, + emb_dim: int, + heads: int, + head_dim: int, + expand_ratio: int, + norm, + norm_eps: float, + ada, + qk_bias: bool, + qk_norm, + mlp_type: str, + shared_weights: bool, + rope_type: str, + rope_dim: int, + is_last_layer: bool, + device, dtype, operations, + **kwargs, + ): + super().__init__() + version = kwargs.get("version", False) + dim = MMArg(vid_dim, txt_dim) + self.attn_norm = MMModule(norm, normalized_shape=dim, eps=norm_eps, elementwise_affine=False, shared_weights=shared_weights, device=device, dtype=dtype) + + self.attn = NaSwinAttention( + vid_dim=vid_dim, + txt_dim=txt_dim, + heads=heads, + head_dim=head_dim, + qk_bias=qk_bias, + qk_norm=qk_norm, + qk_norm_eps=norm_eps, + rope_type=rope_type, + rope_dim=rope_dim, + shared_weights=shared_weights, + window=kwargs.pop("window", None), + window_method=kwargs.pop("window_method", None), + version=version, + device=device, dtype=dtype, operations=operations + ) + + self.mlp_norm = MMModule(norm, normalized_shape=dim, eps=norm_eps, elementwise_affine=False, shared_weights=shared_weights, vid_only=is_last_layer, device=device, dtype=dtype) + self.mlp = MMModule( + get_mlp(mlp_type), + dim=dim, + expand_ratio=expand_ratio, + shared_weights=shared_weights, + vid_only=is_last_layer, + device=device, dtype=dtype, operations=operations + ) + self.ada = MMModule(ada, dim=dim, emb_dim=emb_dim, layers=["attn", "mlp"], shared_weights=shared_weights, vid_only=is_last_layer, device=device, dtype=dtype) + self.is_last_layer = is_last_layer + self.version = version + + def _seedvr2_7b_mlp( + self, + vid: torch.FloatTensor, + txt: torch.FloatTensor, + ) -> Tuple[ + torch.FloatTensor, + torch.FloatTensor, + ]: + vid_module = self.mlp.vid if not self.mlp.shared_weights else self.mlp.all + if comfy.model_management.in_training or vid.requires_grad: + vid = torch.cat([vid_module(chunk) for chunk in vid.split(SEEDVR2_7B_MLP_CHUNK, dim=0)], dim=0) + else: + vid_out = None + offset = 0 + for chunk in vid.split(SEEDVR2_7B_MLP_CHUNK, dim=0): + chunk_out = vid_module(chunk) + if vid_out is None: + vid_out = chunk_out.new_empty((vid.shape[0], *chunk_out.shape[1:])) + vid_out[offset:offset + chunk_out.shape[0]] = chunk_out + offset += chunk_out.shape[0] + vid = vid_out + if not self.mlp.vid_only: + txt_module = self.mlp.txt if not self.mlp.shared_weights else self.mlp.all + txt = txt.to(device=vid.device, dtype=vid.dtype) + txt = txt_module(txt) + return vid, txt + + def forward( + self, + vid: torch.FloatTensor, # l c + txt: torch.FloatTensor, # l c + vid_shape: torch.LongTensor, # b 3 + txt_shape: torch.LongTensor, # b 1 + emb: torch.FloatTensor, + cache: Cache, + ) -> Tuple[ + torch.FloatTensor, + torch.FloatTensor, + torch.LongTensor, + torch.LongTensor, + ]: + hid_len = MMArg( + cache("vid_len", lambda: vid_shape.prod(-1)), + cache("txt_len", lambda: txt_shape.prod(-1)), + ) + ada_kwargs = { + "emb": emb, + "hid_len": hid_len, + "cache": cache, + "branch_tag": MMArg("vid", "txt"), + } + + vid_attn, txt_attn = self.attn_norm(vid, txt) + vid_attn, txt_attn = self.ada(vid_attn, txt_attn, layer="attn", mode="in", **ada_kwargs) + vid_attn, txt_attn = self.attn(vid_attn, txt_attn, vid_shape, txt_shape, cache) + vid_attn, txt_attn = self.ada(vid_attn, txt_attn, layer="attn", mode="out", **ada_kwargs) + vid_attn, txt_attn = (vid_attn + vid), (txt_attn + txt) + + vid_mlp, txt_mlp = self.mlp_norm(vid_attn, txt_attn) + vid_mlp, txt_mlp = self.ada(vid_mlp, txt_mlp, layer="mlp", mode="in", **ada_kwargs) + if self.version: + vid_mlp, txt_mlp = self._seedvr2_7b_mlp(vid_mlp, txt_mlp) + else: + vid_mlp, txt_mlp = self.mlp(vid_mlp, txt_mlp) + vid_mlp, txt_mlp = self.ada(vid_mlp, txt_mlp, layer="mlp", mode="out", **ada_kwargs) + vid_mlp, txt_mlp = (vid_mlp + vid_attn), (txt_mlp + txt_attn) + + return vid_mlp, txt_mlp, vid_shape, txt_shape + +class PatchOut(nn.Module): + def __init__( + self, + out_channels: int, + patch_size: Union[int, Tuple[int, int, int]], + dim: int, + device, dtype, operations + ): + super().__init__() + t, h, w = _triple(patch_size) + self.patch_size = t, h, w + self.proj = operations.Linear(dim, out_channels * t * h * w, device=device, dtype=dtype) + + def forward( + self, + vid: torch.Tensor, + ) -> torch.Tensor: + t, h, w = self.patch_size + vid = self.proj(vid) + vid = rearrange(vid, "b T H W (t h w c) -> b c (T t) (H h) (W w)", t=t, h=h, w=w) + if t > 1: + vid = vid[:, :, (t - 1) :] + return vid + +class NaPatchOut(PatchOut): + def forward( + self, + vid: torch.FloatTensor, # l c + vid_shape: torch.LongTensor, + cache: Cache = Cache(disable=True), # for test + vid_shape_before_patchify = None + ) -> Tuple[ + torch.FloatTensor, + torch.LongTensor, + ]: + + t, h, w = self.patch_size + vid = self.proj(vid) + + if not (t == h == w == 1): + vid = unflatten(vid, vid_shape) + for i in range(len(vid)): + vid[i] = rearrange(vid[i], "T H W (t h w c) -> (T t) (H h) (W w) c", t=t, h=h, w=w) + if t > 1 and vid_shape_before_patchify[i, 0] % t != 0: + vid[i] = vid[i][(t - vid_shape_before_patchify[i, 0] % t) :] + vid, vid_shape = flatten(vid) + + return vid, vid_shape + +class PatchIn(nn.Module): + def __init__( + self, + in_channels: int, + patch_size: Union[int, Tuple[int, int, int]], + dim: int, + device, dtype, operations + ): + super().__init__() + t, h, w = _triple(patch_size) + self.patch_size = t, h, w + self.proj = operations.Linear(in_channels * t * h * w, dim, device=device, dtype=dtype) + + def forward( + self, + vid: torch.Tensor, + ) -> torch.Tensor: + t, h, w = self.patch_size + if t > 1: + assert vid.size(2) % t == 1 + vid = torch.cat([vid[:, :, :1]] * (t - 1) + [vid], dim=2) + vid = rearrange(vid, "b c (T t) (H h) (W w) -> b T H W (t h w c)", t=t, h=h, w=w) + vid = self.proj(vid) + return vid + +class NaPatchIn(PatchIn): + def forward( + self, + vid: torch.Tensor, # l c + vid_shape: torch.LongTensor, + cache: Cache = Cache(disable=True), # for test + ) -> torch.Tensor: + cache = cache.namespace("patch") + vid_shape_before_patchify = cache("vid_shape_before_patchify", lambda: vid_shape) + t, h, w = self.patch_size + if not (t == h == w == 1): + vid = unflatten(vid, vid_shape) + for i in range(len(vid)): + if t > 1 and vid_shape_before_patchify[i, 0] % t != 0: + vid[i] = torch.cat([vid[i][:1]] * (t - vid[i].size(0) % t) + [vid[i]], dim=0) + vid[i] = rearrange(vid[i], "(T t) (H h) (W w) c -> T H W (t h w c)", t=t, h=h, w=w) + vid, vid_shape = flatten(vid) + + vid = self.proj(vid) + return vid, vid_shape + +def expand_dims(x: torch.Tensor, dim: int, ndim: int): + shape = x.shape + shape = shape[:dim] + (1,) * (ndim - len(shape)) + shape[dim:] + return x.reshape(shape) + + +class AdaSingle(nn.Module): + def __init__( + self, + dim: int, + emb_dim: int, + layers: List[str], + modes: List[str] = ["in", "out"], + device = None, dtype = None, + ): + assert emb_dim == 6 * dim, "AdaSingle requires emb_dim == 6 * dim" + super().__init__() + self.dim = dim + self.emb_dim = emb_dim + self.layers = layers + + randn_kwargs = {"device": device} + fp8_types = _torch_float8_types() + if dtype is not None and dtype not in fp8_types: + randn_kwargs["dtype"] = dtype + + for l in layers: + if "in" in modes: + # Passing fp8 ``dtype=`` here would break CPU weight + # loads: CPU has no ``normal_kernel_cpu`` for fp8. + self.register_parameter(f"{l}_shift", nn.Parameter(torch.randn(dim, **randn_kwargs) / dim**0.5)) + self.register_parameter( + f"{l}_scale", nn.Parameter(torch.randn(dim, **randn_kwargs) / dim**0.5 + 1) + ) + if "out" in modes: + self.register_parameter(f"{l}_gate", nn.Parameter(torch.randn(dim, **randn_kwargs) / dim**0.5)) + + def forward( + self, + hid: torch.FloatTensor, # b ... c + emb: torch.FloatTensor, # b d + layer: str, + mode: str, + cache: Cache = Cache(disable=True), + branch_tag: str = "", + hid_len: Optional[torch.LongTensor] = None, # b + ) -> torch.FloatTensor: + idx = self.layers.index(layer) + emb = rearrange(emb, "b (d l g) -> b d l g", l=len(self.layers), g=3)[..., idx, :] + emb = expand_dims(emb, 1, hid.ndim + 1) + + if hid_len is not None: + slice_inputs = lambda x, dim: x + emb = cache( + f"emb_repeat_{idx}_{branch_tag}", + lambda: slice_inputs( + torch.repeat_interleave(emb, hid_len, dim=0), + dim=0, + ), + ) + + shiftA, scaleA, gateA = emb.unbind(-1) + shiftB, scaleB, gateB = ( + getattr(self, f"{layer}_shift", None), + getattr(self, f"{layer}_scale", None), + getattr(self, f"{layer}_gate", None), + ) + + fp8_types = _torch_float8_types() + if fp8_types: + target_dtype = hid.dtype + + if shiftB is not None and shiftB.dtype in fp8_types: + shiftB = shiftB.to(target_dtype) + if scaleB is not None and scaleB.dtype in fp8_types: + scaleB = scaleB.to(target_dtype) + if gateB is not None and gateB.dtype in fp8_types: + gateB = gateB.to(target_dtype) + + if mode == "in": + return hid.mul_(scaleA + scaleB).add_(shiftA + shiftB) + if mode == "out": + if gateB is not None: + return hid.mul_(gateA + gateB) + else: + return hid.mul_(gateA) + + raise NotImplementedError + + +def emb_add(emb1: torch.Tensor, emb2: Optional[torch.Tensor]): + return emb1 if emb2 is None else emb1 + emb2 + + +class TimeEmbedding(nn.Module): + def __init__( + self, + sinusoidal_dim: int, + hidden_dim: int, + output_dim: int, + device, dtype, operations + ): + super().__init__() + self.sinusoidal_dim = sinusoidal_dim + self.proj_in = operations.Linear(sinusoidal_dim, hidden_dim, device=device, dtype=dtype) + self.proj_hid = operations.Linear(hidden_dim, hidden_dim, device=device, dtype=dtype) + self.proj_out = operations.Linear(hidden_dim, output_dim, device=device, dtype=dtype) + self.act = nn.SiLU() + + def forward( + self, + timestep: Union[int, float, torch.IntTensor, torch.FloatTensor], + device: torch.device, + dtype: torch.dtype, + ) -> torch.FloatTensor: + if not torch.is_tensor(timestep): + timestep = torch.tensor([timestep], device=device, dtype=dtype) + if timestep.ndim == 0: + timestep = timestep[None] + + emb = get_timestep_embedding( + timesteps=timestep, + embedding_dim=self.sinusoidal_dim, + flip_sin_to_cos=False, + downscale_freq_shift=0, + ).to(dtype) + emb = self.proj_in(emb) + emb = self.act(emb) + emb = self.proj_hid(emb) + emb = self.act(emb) + emb = self.proj_out(emb) + return emb + +def flatten( + hid: List[torch.FloatTensor], # List of (*** c) +) -> Tuple[ + torch.FloatTensor, # (L c) + torch.LongTensor, # (b n) +]: + assert len(hid) > 0 + shape = torch.stack([torch.tensor(x.shape[:-1], device=hid[0].device) for x in hid]) + hid = torch.cat([x.flatten(0, -2) for x in hid]) + return hid, shape + + +def unflatten( + hid: torch.FloatTensor, # (L c) or (L ... c) + hid_shape: torch.LongTensor, # (b n) +) -> List[torch.Tensor]: # List of (*** c) or (*** ... c) + hid_len = hid_shape.prod(-1) + hid = hid.split(hid_len.tolist()) + hid = [x.unflatten(0, s.tolist()) for x, s in zip(hid, hid_shape)] + return hid + +def repeat( + hid: torch.FloatTensor, # (L c) + hid_shape: torch.LongTensor, # (b n) + pattern: str, + **kwargs: Dict[str, torch.LongTensor], # (b) +) -> Tuple[ + torch.FloatTensor, + torch.LongTensor, +]: + hid = unflatten(hid, hid_shape) + kwargs = [{k: v[i].item() for k, v in kwargs.items()} for i in range(len(hid))] + return flatten([einops.repeat(h, pattern, **a) for h, a in zip(hid, kwargs)]) + +class NaDiT(nn.Module): + + def __init__( + self, + norm_eps, + qk_rope, + num_layers, + mlp_type, + vid_in_channels = 33, + vid_out_channels = 16, + vid_dim = 2560, + txt_in_dim = 5120, + heads = 20, + head_dim = 128, + mm_layers = 10, + expand_ratio = 4, + qk_bias = False, + patch_size = [ 1,2,2 ], + shared_qkv: bool = False, + shared_mlp: bool = False, + window_method: Optional[Tuple[str]] = None, + temporal_window_size: int = None, + temporal_shifted: bool = False, + rope_dim = 128, + rope_type = "mmrope3d", + vid_out_norm: Optional[str] = None, + device = None, + dtype = None, + operations = None, + **kwargs, + ): + self._7b_version = vid_dim == 3072 + self.dtype = dtype + factory_kwargs = {"device": device, "dtype": dtype} + window_method = num_layers // 2 * ["720pwin_by_size_bysize","720pswin_by_size_bysize"] + txt_dim = vid_dim + emb_dim = vid_dim * 6 + block_type = ["mmdit_sr"] * num_layers + window = num_layers * [(4,3,3)] + ada = AdaSingle + norm = CustomRMSNorm + qk_norm = CustomRMSNorm + if isinstance(block_type, str): + block_type = [block_type] * num_layers + elif len(block_type) != num_layers: + raise ValueError("The ``block_type`` list should equal to ``num_layers``.") + super().__init__() + # ``torch.empty`` returns uninitialized memory, not zeros. The + # SeedVR2Conditioning fail-loud guard at + # ``comfy_extras/nodes_seedvr.py`` distinguishes "buffer was loaded" + # from "buffer was never populated by the file" by checking + # ``positive_conditioning.abs().sum() == 0``. That sentinel is only + # reliable if the post-construction buffer state is deterministically + # zero, so explicitly zero-fill here rather than relying on the + # allocator's zero-on-alloc behavior (allocator-dependent and not + # contractual). When ``load_state_dict`` populates these buffers + # from a properly-baked SeedVR2 .safetensors, the in-place copy + # overwrites the zeros with the universal SeedVR2 conditioning + # tensors (shape (58, 5120) and (64, 5120) bf16). + self.register_buffer("positive_conditioning", torch.zeros((58, 5120), device=device, dtype=dtype)) + self.register_buffer("negative_conditioning", torch.zeros((64, 5120), device=device, dtype=dtype)) + self.vid_in = NaPatchIn( + in_channels=vid_in_channels, + patch_size=patch_size, + dim=vid_dim, + device=device, dtype=dtype, operations=operations + ) + self.txt_in = ( + operations.Linear(txt_in_dim, txt_dim, **factory_kwargs) + if txt_in_dim and txt_in_dim != txt_dim + else nn.Identity() + ) + self.emb_in = TimeEmbedding( + sinusoidal_dim=256, + hidden_dim=max(vid_dim, txt_dim), + output_dim=emb_dim, + device=device, dtype=dtype, operations=operations + ) + + if window is None or isinstance(window[0], int): + window = [window] * num_layers + if window_method is None or isinstance(window_method, str): + window_method = [window_method] * num_layers + if temporal_window_size is None or isinstance(temporal_window_size, int): + temporal_window_size = [temporal_window_size] * num_layers + if temporal_shifted is None or isinstance(temporal_shifted, bool): + temporal_shifted = [temporal_shifted] * num_layers + + rope_dim = rope_dim if rope_dim is not None else head_dim // 2 + self.blocks = nn.ModuleList( + [ + NaMMSRTransformerBlock( + vid_dim=vid_dim, + txt_dim=txt_dim, + emb_dim=emb_dim, + heads=heads, + head_dim=head_dim, + expand_ratio=expand_ratio, + norm=norm, + norm_eps=norm_eps, + ada=ada, + qk_bias=qk_bias, + qk_rope=qk_rope, + qk_norm=qk_norm, + shared_qkv=shared_qkv, + shared_mlp=shared_mlp, + mlp_type=mlp_type, + rope_dim = rope_dim, + window=window[i], + window_method=window_method[i], + temporal_window_size=temporal_window_size[i], + temporal_shifted=temporal_shifted[i], + is_last_layer=(i == num_layers - 1) and not self._7b_version, + rope_type = rope_type, + shared_weights=not ( + (i < mm_layers) if isinstance(mm_layers, int) else mm_layers[i] + ), + version = self._7b_version, + operations = operations, + **kwargs, + **factory_kwargs + ) + for i in range(num_layers) + ] + ) + self.vid_out = NaPatchOut( + out_channels=vid_out_channels, + patch_size=patch_size, + dim=vid_dim, + device=device, dtype=dtype, operations=operations + ) + + self.need_txt_repeat = block_type[0] in [ + "mmdit_stwin", + "mmdit_stwin_spatial", + "mmdit_stwin_3d_spatial", + ] + + self.vid_out_norm = None + if vid_out_norm is not None: + self.vid_out_norm = CustomRMSNorm( + normalized_shape=vid_dim, + eps=norm_eps, + elementwise_affine=True, + device=device, dtype=dtype + ) + self.vid_out_ada = ada( + dim=vid_dim, + emb_dim=emb_dim, + layers=["out"], + modes=["in"], + device=device, dtype=dtype + ) + + def _resolve_text_conditioning(self, context, cond_or_uncond=None): + if context is None or getattr(context, "numel", lambda: None)() == 0: + context = self.positive_conditioning + return flatten([context]) + if NaDiT._seedvr2_is_single_conditioning_branch(cond_or_uncond): + if context.shape[0] == 1: + context = context.squeeze(0) + return flatten([context]) + return flatten(context.unbind(0)) + if context.shape[0] % 2 != 0: + raise ValueError(f"SeedVR2 expected an even text-conditioning batch, got shape {tuple(context.shape)}") + neg_cond, pos_cond = context.chunk(2, dim=0) + if pos_cond.shape[0] == 1: + pos_cond, neg_cond = pos_cond.squeeze(0), neg_cond.squeeze(0) + return flatten([pos_cond, neg_cond]) + return flatten((*pos_cond.unbind(0), *neg_cond.unbind(0))) + + @staticmethod + def _seedvr2_is_single_conditioning_branch(cond_or_uncond): + if cond_or_uncond is None or len(cond_or_uncond) == 0: + return False + first = cond_or_uncond[0] + return all(entry == first for entry in cond_or_uncond) + + def _swap_pos_neg_halves(self, out, cond_or_uncond=None): + if NaDiT._seedvr2_is_single_conditioning_branch(cond_or_uncond): + return out + # ``dim=0`` is explicit on both calls. The contract is "split + # the batch axis into two halves and swap them"; making the + # axis load-bearing in source guards against silent drift if a + # future refactor reorders tensor axes. + pos, neg = out.chunk(2, dim=0) + return torch.cat([neg, pos], dim=0) + + def forward( + self, + x, + timestep, + context, # l c + disable_cache: bool = False, # for test # TODO ? // gives an error when set to True + **kwargs + ): + transformer_options = kwargs.get("transformer_options", {}) + patches_replace = transformer_options.get("patches_replace", {}) + blocks_replace = patches_replace.get("dit", {}) + conditions = kwargs.get("condition") + b, tc, h, w = x.shape + x = x.view(b, 16, -1, h, w) + conditions = conditions.view(b, 17, -1, h, w) + x = x.movedim(1, -1) + conditions = conditions.movedim(1, -1) + cache = Cache(disable=disable_cache) + + txt, txt_shape = self._resolve_text_conditioning(context, transformer_options.get("cond_or_uncond")) + + vid, vid_shape = flatten(x) + cond_latent, _ = flatten(conditions) + + vid = torch.cat([vid, cond_latent], dim=-1) + if txt_shape.size(-1) == 1 and self.need_txt_repeat: + txt, txt_shape = repeat(txt, txt_shape, "l c -> t l c", t=vid_shape[:, 0]) + + txt = self.txt_in(txt) + + vid_shape_before_patchify = vid_shape + vid, vid_shape = self.vid_in(vid, vid_shape, cache=cache) + + emb = self.emb_in(timestep, device=vid.device, dtype=vid.dtype) + + for i, block in enumerate(self.blocks): + if ("block", i) in blocks_replace: + def block_wrap(args): + out = {} + out["vid"], out["txt"], out["vid_shape"], out["txt_shape"] = block( + vid=args["vid"], + txt=args["txt"], + vid_shape=args["vid_shape"], + txt_shape=args["txt_shape"], + emb=args["emb"], + cache=args["cache"], + ) + return out + out = blocks_replace[("block", i)]({ + "vid":vid, + "txt":txt, + "vid_shape":vid_shape, + "txt_shape":txt_shape, + "emb":emb, + "cache":cache, + }, {"original_block": block_wrap}) + vid, txt, vid_shape, txt_shape = out["vid"], out["txt"], out["vid_shape"], out["txt_shape"] + else: + vid, txt, vid_shape, txt_shape = block( + vid=vid, + txt=txt, + vid_shape=vid_shape, + txt_shape=txt_shape, + emb=emb, + cache=cache, + ) + + if self.vid_out_norm: + vid = self.vid_out_norm(vid) + vid = self.vid_out_ada( + vid, + emb=emb, + layer="out", + mode="in", + hid_len=cache("vid_len", lambda: vid_shape.prod(-1)), + cache=cache, + branch_tag="vid", + ) + + vid, vid_shape = self.vid_out(vid, vid_shape, cache, vid_shape_before_patchify = vid_shape_before_patchify) + vid = unflatten(vid, vid_shape) + out = torch.stack(vid) + out = out.movedim(-1, 1) + out = rearrange(out, "b c t h w -> b (c t) h w") + return self._swap_pos_neg_halves(out, transformer_options.get("cond_or_uncond")) diff --git a/comfy/ldm/seedvr/vae.py b/comfy/ldm/seedvr/vae.py new file mode 100644 index 000000000..0593fa547 --- /dev/null +++ b/comfy/ldm/seedvr/vae.py @@ -0,0 +1,2421 @@ +from contextlib import nullcontext +from typing import Literal, Optional, Tuple +import gc +import torch +import torch.nn as nn +import torch.nn.functional as F +from einops import rearrange +from torch import Tensor +from contextlib import contextmanager +from comfy.utils import ProgressBar + +from comfy.ldm.seedvr.model import safe_pad_operation +from comfy.ldm.modules.attention import optimized_attention +from comfy.ldm.modules.diffusionmodules.model import vae_attention + +import math +from enum import Enum +from comfy.ops import NVIDIA_MEMORY_CONV_BUG_WORKAROUND + +import logging +import comfy.model_management +import comfy.ops +ops = comfy.ops.disable_weight_init + + +def _seedvr2_temporal_slicing_min_size(temporal_size, temporal_overlap, temporal_scale=1): + if temporal_size is None: + return None + + temporal_size = int(temporal_size) + if temporal_size <= 0: + return 0 + + temporal_overlap = max(0, int(temporal_overlap or 0)) + temporal_overlap = min(temporal_overlap, temporal_size - 1) + temporal_step = temporal_size - temporal_overlap + temporal_scale = max(1, int(temporal_scale)) + return max(1, math.ceil(temporal_step / temporal_scale)) + + +def _seedvr2_clamped_spatial_overlap(overlap, tile_size): + overlap = max(0, int(overlap)) + tile_size = max(1, int(tile_size)) + return min(overlap, tile_size - 1) + + +def _seedvr2_clear_temporal_memory(model): + for module in model.modules(): + if hasattr(module, "memory"): + module.memory = None + + +@torch.inference_mode() +def tiled_vae( + x, + vae_model, + tile_size=(512, 512), + tile_overlap=(64, 64), + temporal_size=16, + temporal_overlap=0, + encode=True, + **kwargs, +): + gc.collect() + comfy.model_management.soft_empty_cache() + + x = x.to(next(vae_model.parameters()).dtype) + if x.ndim != 5: + x = x.unsqueeze(2) + + _, _, d, h, w = x.shape + + sf_s = getattr(vae_model, "spatial_downsample_factor", 8) + sf_t = getattr(vae_model, "temporal_downsample_factor", 4) + if encode: + slicing_attr = "slicing_sample_min_size" + slicing_min_size = _seedvr2_temporal_slicing_min_size(temporal_size, temporal_overlap) + else: + slicing_attr = "slicing_latent_min_size" + slicing_min_size = _seedvr2_temporal_slicing_min_size(temporal_size, temporal_overlap, sf_t) + if encode: + ti_h, ti_w = tile_size + ov_h = _seedvr2_clamped_spatial_overlap(tile_overlap[0], ti_h) + ov_w = _seedvr2_clamped_spatial_overlap(tile_overlap[1], ti_w) + blend_ov_h = max(0, ov_h // sf_s) + blend_ov_w = max(0, ov_w // sf_s) + target_d = (d + sf_t - 1) // sf_t + target_h = (h + sf_s - 1) // sf_s + target_w = (w + sf_s - 1) // sf_s + else: + ti_h = max(1, tile_size[0] // sf_s) + ti_w = max(1, tile_size[1] // sf_s) + ov_h = _seedvr2_clamped_spatial_overlap(tile_overlap[0] // sf_s, ti_h) + ov_w = _seedvr2_clamped_spatial_overlap(tile_overlap[1] // sf_s, ti_w) + blend_ov_h = ov_h * sf_s + blend_ov_w = ov_w * sf_s + + target_d = max(1, d * sf_t - (sf_t - 1)) + target_h = h * sf_s + target_w = w * sf_s + + stride_h = max(1, ti_h - ov_h) + stride_w = max(1, ti_w - ov_w) + + storage_device = vae_model.device + result = None + count = None + def run_temporal_chunks(spatial_tile, model=vae_model, device=storage_device): + device = torch.device(device) + _seedvr2_clear_temporal_memory(model) + t_chunk = spatial_tile.to(device=device, dtype=next(model.parameters()).dtype, non_blocking=True).contiguous() + old_device = getattr(model, "device", None) + model.device = device + old_slicing_min_size = getattr(model, slicing_attr, None) + if old_slicing_min_size is not None and slicing_min_size is not None: + if slicing_min_size <= 0: + setattr(model, slicing_attr, t_chunk.shape[2]) + else: + setattr(model, slicing_attr, slicing_min_size) + try: + if encode: + out = model.encode(t_chunk)[0] + else: + out = model.decode_(t_chunk) + finally: + if old_slicing_min_size is not None and slicing_min_size is not None: + setattr(model, slicing_attr, old_slicing_min_size) + if old_device is not None: + model.device = old_device + if isinstance(out, (tuple, list)): + out = out[0] + if out.ndim == 4: + out = out.unsqueeze(2) + return out.to(storage_device) + + ramp_cache = {} + def get_ramp(steps): + if steps not in ramp_cache: + t = torch.linspace(0, 1, steps=steps, device=storage_device, dtype=torch.float32) + ramp_cache[steps] = 0.5 - 0.5 * torch.cos(t * torch.pi) + return ramp_cache[steps] + + tile_ranges = [] + for y_idx in range(0, h, stride_h): + y_end = min(y_idx + ti_h, h) + if y_idx > 0 and (y_end - y_idx) <= ov_h: + continue + for x_idx in range(0, w, stride_w): + x_end = min(x_idx + ti_w, w) + if x_idx > 0 and (x_end - x_idx) <= ov_w: + continue + tile_ranges.append((y_idx, y_end, x_idx, x_end)) + + total_tiles = len(tile_ranges) + bar = ProgressBar(total_tiles) + single_spatial_tile = h <= ti_h and w <= ti_w + + _seedvr2_clear_temporal_memory(vae_model) + + def run_tile(tile_index, tile_range): + y_idx, y_end, x_idx, x_end = tile_range + tile_x = x[:, :, :, y_idx:y_end, x_idx:x_end] + tile_out = run_temporal_chunks(tile_x) + return tile_index, y_idx, y_end, x_idx, x_end, tile_out + + ordered_tile_outputs = ( + run_tile(tile_index, tile_range) + for tile_index, tile_range in enumerate(tile_ranges) + ) + + for _, y_idx, y_end, x_idx, x_end, tile_out in ordered_tile_outputs: + + if single_spatial_tile: + result = tile_out[:, :, :target_d, :target_h, :target_w] + if result.device != x.device: + result = result.to(x.device).to(x.dtype) + if x.shape[2] == 1 and sf_t == 1: + result = result.squeeze(2) + bar.update(1) + return result + + if result is None: + b_out, c_out = tile_out.shape[0], tile_out.shape[1] + result = torch.zeros((b_out, c_out, target_d, target_h, target_w), device=storage_device, dtype=torch.float32) + count = torch.zeros((1, 1, 1, target_h, target_w), device=storage_device, dtype=torch.float32) + + if encode: + ys, ye = y_idx // sf_s, (y_idx // sf_s) + tile_out.shape[3] + xs, xe = x_idx // sf_s, (x_idx // sf_s) + tile_out.shape[4] + cur_ov_h = max(0, min(blend_ov_h, tile_out.shape[3] // 2)) + cur_ov_w = max(0, min(blend_ov_w, tile_out.shape[4] // 2)) + else: + ys, ye = y_idx * sf_s, (y_idx * sf_s) + tile_out.shape[3] + xs, xe = x_idx * sf_s, (x_idx * sf_s) + tile_out.shape[4] + cur_ov_h = max(0, min(blend_ov_h, tile_out.shape[3] // 2)) + cur_ov_w = max(0, min(blend_ov_w, tile_out.shape[4] // 2)) + + w_h = torch.ones((tile_out.shape[3],), device=storage_device) + w_w = torch.ones((tile_out.shape[4],), device=storage_device) + + if cur_ov_h > 0: + r = get_ramp(cur_ov_h) + if y_idx > 0: + w_h[:cur_ov_h] = r + if y_end < h: + w_h[-cur_ov_h:] = 1.0 - r + + if cur_ov_w > 0: + r = get_ramp(cur_ov_w) + if x_idx > 0: + w_w[:cur_ov_w] = r + if x_end < w: + w_w[-cur_ov_w:] = 1.0 - r + + final_weight = w_h.view(1,1,1,-1,1) * w_w.view(1,1,1,1,-1) + + valid_d = min(tile_out.shape[2], result.shape[2]) + tile_out = tile_out[:, :, :valid_d, :, :] + + tile_out.mul_(final_weight) + + result[:, :, :valid_d, ys:ye, xs:xe] += tile_out + count[:, :, :, ys:ye, xs:xe] += final_weight + + del tile_out, final_weight, w_h, w_w + bar.update(1) + + result.div_(count.clamp(min=1e-6)) + _seedvr2_clear_temporal_memory(vae_model) + + if result.device != x.device: + result = result.to(x.device).to(x.dtype) + + if x.shape[2] == 1 and sf_t == 1: + result = result.squeeze(2) + + return result + +_NORM_LIMIT = float("inf") +def get_norm_limit(): + return _NORM_LIMIT + + +def set_norm_limit(value: Optional[float] = None): + global _NORM_LIMIT + if value is None: + value = float("inf") + _NORM_LIMIT = value + +@contextmanager +def ignore_padding(model): + orig_padding = model.padding + model.padding = (0, 0, 0) + try: + yield + finally: + model.padding = orig_padding + +class MemoryState(Enum): + DISABLED = 0 + INITIALIZING = 1 + ACTIVE = 2 + UNSET = 3 + +def get_cache_size(conv_module, input_len, pad_len, dim=0): + dilated_kernerl_size = conv_module.dilation[dim] * (conv_module.kernel_size[dim] - 1) + 1 + output_len = (input_len + pad_len - dilated_kernerl_size) // conv_module.stride[dim] + 1 + remain_len = ( + input_len + pad_len - ((output_len - 1) * conv_module.stride[dim] + dilated_kernerl_size) + ) + overlap_len = dilated_kernerl_size - conv_module.stride[dim] + cache_len = overlap_len + remain_len # >= 0 + + assert output_len > 0 + return cache_len + +class DiagonalGaussianDistribution(object): + def __init__(self, parameters: torch.Tensor, deterministic: bool = False): + self.parameters = parameters + self.mean, self.logvar = torch.chunk(parameters, 2, dim=1) + self.logvar = torch.clamp(self.logvar, -30.0, 20.0) + self.deterministic = deterministic + self.std = torch.exp(0.5 * self.logvar) + self.var = torch.exp(self.logvar) + if self.deterministic: + self.var = self.std = torch.zeros_like( + self.mean, device=self.parameters.device, dtype=self.parameters.dtype + ) + + def sample(self, generator: Optional[torch.Generator] = None) -> torch.Tensor: + sample = torch.randn( + self.mean.shape, + generator=generator, + device=self.parameters.device, + dtype=self.parameters.dtype, + ) + x = self.mean + self.std * sample + return x + + def mode(self): + return self.mean + +class SpatialNorm(nn.Module): + def __init__( + self, + f_channels: int, + zq_channels: int, + ): + super().__init__() + self.norm_layer = ops.GroupNorm(num_channels=f_channels, num_groups=32, eps=1e-6, affine=True) + self.conv_y = ops.Conv2d(zq_channels, f_channels, kernel_size=1, stride=1, padding=0) + self.conv_b = ops.Conv2d(zq_channels, f_channels, kernel_size=1, stride=1, padding=0) + + def forward(self, f: torch.Tensor, zq: torch.Tensor) -> torch.Tensor: + f_size = f.shape[-2:] + zq = F.interpolate(zq, size=f_size, mode="nearest") + norm_f = self.norm_layer(f) + new_f = norm_f * self.conv_y(zq) + self.conv_b(zq) + return new_f + +# partial implementation of diffusers's Attention for comfyui +class Attention(nn.Module): + def __init__( + self, + query_dim: int, + cross_attention_dim: Optional[int] = None, + heads: int = 8, + kv_heads: Optional[int] = None, + dim_head: int = 64, + dropout: float = 0.0, + bias: bool = False, + upcast_attention: bool = False, + upcast_softmax: bool = False, + added_kv_proj_dim: Optional[int] = None, + added_proj_bias: Optional[bool] = True, + norm_num_groups: Optional[int] = None, + spatial_norm_dim: Optional[int] = None, + out_bias: bool = True, + scale_qk: bool = True, + only_cross_attention: bool = False, + eps: float = 1e-5, + rescale_output_factor: float = 1.0, + residual_connection: bool = False, + _from_deprecated_attn_block: bool = False, + out_dim: int = None, + out_context_dim: int = None, + context_pre_only=None, + pre_only=False, + is_causal: bool = False, + ): + super().__init__() + + self.inner_dim = out_dim if out_dim is not None else dim_head * heads + self.inner_kv_dim = self.inner_dim if kv_heads is None else dim_head * kv_heads + self.query_dim = query_dim + self.use_bias = bias + self.is_cross_attention = cross_attention_dim is not None + self.cross_attention_dim = cross_attention_dim if cross_attention_dim is not None else query_dim + self.upcast_attention = upcast_attention + self.upcast_softmax = upcast_softmax + self.rescale_output_factor = rescale_output_factor + self.residual_connection = residual_connection + self.dropout = dropout + self.fused_projections = False + self.out_dim = out_dim if out_dim is not None else query_dim + self.out_context_dim = out_context_dim if out_context_dim is not None else query_dim + self.context_pre_only = context_pre_only + self.pre_only = pre_only + self.is_causal = is_causal + + # we make use of this private variable to know whether this class is loaded + # with an deprecated state dict so that we can convert it on the fly + self._from_deprecated_attn_block = _from_deprecated_attn_block + + self.scale_qk = scale_qk + self.scale = dim_head**-0.5 if self.scale_qk else 1.0 + + self.heads = out_dim // dim_head if out_dim is not None else heads + self.sliceable_head_dim = heads + + self.added_kv_proj_dim = added_kv_proj_dim + self.only_cross_attention = only_cross_attention + + if norm_num_groups is not None: + self.group_norm = ops.GroupNorm(num_channels=query_dim, num_groups=norm_num_groups, eps=eps, affine=True) + else: + self.group_norm = None + + if spatial_norm_dim is not None: + self.spatial_norm = SpatialNorm(f_channels=query_dim, zq_channels=spatial_norm_dim) + else: + self.spatial_norm = None + + self.norm_q = None + self.norm_k = None + + self.norm_cross = None + self.to_q = ops.Linear(query_dim, self.inner_dim, bias=bias) + + if not self.only_cross_attention: + # only relevant for the `AddedKVProcessor` classes + self.to_k = ops.Linear(self.cross_attention_dim, self.inner_kv_dim, bias=bias) + self.to_v = ops.Linear(self.cross_attention_dim, self.inner_kv_dim, bias=bias) + else: + self.to_k = None + self.to_v = None + + self.added_proj_bias = added_proj_bias + if self.added_kv_proj_dim is not None: + self.add_k_proj = ops.Linear(added_kv_proj_dim, self.inner_kv_dim, bias=added_proj_bias) + self.add_v_proj = ops.Linear(added_kv_proj_dim, self.inner_kv_dim, bias=added_proj_bias) + if self.context_pre_only is not None: + self.add_q_proj = ops.Linear(added_kv_proj_dim, self.inner_dim, bias=added_proj_bias) + else: + self.add_q_proj = None + self.add_k_proj = None + self.add_v_proj = None + + if not self.pre_only: + self.to_out = nn.ModuleList([]) + self.to_out.append(ops.Linear(self.inner_dim, self.out_dim, bias=out_bias)) + self.to_out.append(nn.Dropout(dropout)) + else: + self.to_out = None + + if self.context_pre_only is not None and not self.context_pre_only: + self.to_add_out = ops.Linear(self.inner_dim, self.out_context_dim, bias=out_bias) + else: + self.to_add_out = None + + self.norm_added_q = None + self.norm_added_k = None + self.optimized_vae_attention = vae_attention() + + def __call__( + self, + hidden_states: torch.Tensor, + encoder_hidden_states: Optional[torch.Tensor] = None, + attention_mask: Optional[torch.Tensor] = None, + temb: Optional[torch.Tensor] = None, + *args, + **kwargs, + ) -> torch.Tensor: + + residual = hidden_states + if self.spatial_norm is not None: + hidden_states = self.spatial_norm(hidden_states, temb) + + input_ndim = hidden_states.ndim + + if input_ndim == 4: + batch_size, channel, height, width = hidden_states.shape + hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2) + + batch_size, sequence_length, _ = ( + hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape + ) + + if attention_mask is not None: + attention_mask = self.prepare_attention_mask(attention_mask, sequence_length, batch_size) + attention_mask = attention_mask.view(batch_size, self.heads, -1, attention_mask.shape[-1]) + + if self.group_norm is not None: + hidden_states = self.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2) + + query = self.to_q(hidden_states) + + if encoder_hidden_states is None: + encoder_hidden_states = hidden_states + elif self.norm_cross: + encoder_hidden_states = self.norm_encoder_hidden_states(encoder_hidden_states) + + key = self.to_k(encoder_hidden_states) + value = self.to_v(encoder_hidden_states) + + inner_dim = key.shape[-1] + head_dim = inner_dim // self.heads + + query = query.view(batch_size, -1, self.heads, head_dim).transpose(1, 2) + + key = key.view(batch_size, -1, self.heads, head_dim).transpose(1, 2) + value = value.view(batch_size, -1, self.heads, head_dim).transpose(1, 2) + + if self.norm_q is not None: + query = self.norm_q(query) + if self.norm_k is not None: + key = self.norm_k(key) + + if input_ndim == 4 and encoder_hidden_states is hidden_states and attention_mask is None and self.heads == 1: + query = query.squeeze(1).transpose(1, 2).reshape(batch_size, head_dim, height, width) + key = key.squeeze(1).transpose(1, 2).reshape(batch_size, head_dim, height, width) + value = value.squeeze(1).transpose(1, 2).reshape(batch_size, head_dim, height, width) + hidden_states = self.optimized_vae_attention(query, key, value).reshape(batch_size, self.heads, head_dim, height * width).transpose(2, 3) + else: + hidden_states = optimized_attention(query, key, value, heads = self.heads, mask = attention_mask, skip_reshape=True, skip_output_reshape=True) + + hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, self.heads * head_dim) + hidden_states = hidden_states.to(query.dtype) + + # linear proj + hidden_states = self.to_out[0](hidden_states) + # dropout + hidden_states = self.to_out[1](hidden_states) + + if input_ndim == 4: + hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width) + + if self.residual_connection: + hidden_states = hidden_states + residual + + hidden_states = hidden_states / self.rescale_output_factor + + return hidden_states + + +def inflate_weight(weight_2d: torch.Tensor, weight_3d: torch.Tensor): + with torch.no_grad(): + depth = weight_3d.size(2) + weight_3d.copy_(weight_2d.unsqueeze(2).repeat(1, 1, depth, 1, 1) / depth) + return weight_3d + +def inflate_bias(bias_2d: torch.Tensor, bias_3d: torch.Tensor): + with torch.no_grad(): + bias_3d.copy_(bias_2d) + return bias_3d + + +def modify_state_dict(layer, state_dict, prefix, inflate_weight_fn, inflate_bias_fn): + weight_name = prefix + "weight" + bias_name = prefix + "bias" + if weight_name in state_dict: + weight_2d = state_dict[weight_name] + if weight_2d.dim() == 4: + weight_3d = inflate_weight_fn( + weight_2d=weight_2d, + weight_3d=layer.weight, + ) + state_dict[weight_name] = weight_3d + else: + return state_dict + if bias_name in state_dict: + bias_2d = state_dict[bias_name] + if bias_2d.dim() == 1: + bias_3d = inflate_bias_fn( + bias_2d=bias_2d, + bias_3d=layer.bias, + ) + state_dict[bias_name] = bias_3d + return state_dict + +def causal_norm_wrapper(norm_layer: nn.Module, x: torch.Tensor) -> torch.Tensor: + input_dtype = x.dtype + if isinstance(norm_layer, (ops.LayerNorm, ops.RMSNorm)): + if x.ndim == 4: + x = rearrange(x, "b c h w -> b h w c") + x = norm_layer(x) + x = rearrange(x, "b h w c -> b c h w") + return x.to(input_dtype) + if x.ndim == 5: + x = rearrange(x, "b c t h w -> b t h w c") + x = norm_layer(x) + x = rearrange(x, "b t h w c -> b c t h w") + return x.to(input_dtype) + if isinstance(norm_layer, (ops.GroupNorm, nn.BatchNorm2d, nn.SyncBatchNorm)): + if x.ndim <= 4: + return norm_layer(x).to(input_dtype) + if x.ndim == 5: + t = x.size(2) + x = rearrange(x, "b c t h w -> (b t) c h w") + memory_occupy = x.numel() * x.element_size() / 1024**3 + if isinstance(norm_layer, ops.GroupNorm) and memory_occupy > get_norm_limit(): + num_chunks = min(4 if x.element_size() == 2 else 2, norm_layer.num_groups) + assert norm_layer.num_groups % num_chunks == 0 + num_groups_per_chunk = norm_layer.num_groups // num_chunks + + x = list(x.chunk(num_chunks, dim=1)) + weights = norm_layer.weight.chunk(num_chunks, dim=0) + biases = norm_layer.bias.chunk(num_chunks, dim=0) + for i, (w, b) in enumerate(zip(weights, biases)): + x[i] = F.group_norm(x[i], num_groups_per_chunk, w, b, norm_layer.eps) + x[i] = x[i].to(input_dtype) + x = torch.cat(x, dim=1) + else: + x = norm_layer(x) + x = rearrange(x, "(b t) c h w -> b c t h w", t=t) + return x.to(input_dtype) + raise NotImplementedError + +def safe_interpolate_operation(x, size=None, scale_factor=None, mode='nearest', align_corners=None, recompute_scale_factor=None): + problematic_modes = ['bilinear', 'bicubic', 'trilinear'] + + if mode in problematic_modes: + try: + return F.interpolate( + x, + size=size, + scale_factor=scale_factor, + mode=mode, + align_corners=align_corners, + recompute_scale_factor=recompute_scale_factor + ) + except RuntimeError as e: + if ("not implemented for 'Half'" in str(e) or + "compute_indices_weights" in str(e)): + original_dtype = x.dtype + return F.interpolate( + x.float(), + size=size, + scale_factor=scale_factor, + mode=mode, + align_corners=align_corners, + recompute_scale_factor=recompute_scale_factor + ).to(original_dtype) + else: + raise e + else: + # Pour 'nearest' et autres modes compatibles, pas de fix nécessaire + return F.interpolate( + x, + size=size, + scale_factor=scale_factor, + mode=mode, + align_corners=align_corners, + recompute_scale_factor=recompute_scale_factor + ) + +_receptive_field_t = Literal["half", "full"] + +def extend_head(tensor, times: int = 2, memory = None): + if memory is not None: + return torch.cat((memory.to(tensor), tensor), dim=2) + assert times >= 0, "Invalid input for function 'extend_head'!" + if times == 0: + return tensor + else: + tile_repeat = [1] * tensor.ndim + tile_repeat[2] = times + return torch.cat(tensors=(torch.tile(tensor[:, :, :1], tile_repeat), tensor), dim=2) + +def cache_send_recv(tensor, cache_size, times, memory=None): + recv_buffer = None + + if memory is not None: + recv_buffer = memory.to(tensor[0]) + elif times > 0: + tile_repeat = [1] * tensor[0].ndim + tile_repeat[2] = times + recv_buffer = torch.tile(tensor[0][:, :, :1], tile_repeat) + + return recv_buffer + +class InflatedCausalConv3d(ops.Conv3d): + def __init__( + self, + *args, + inflation_mode, + memory_device = "same", + **kwargs, + ): + self.inflation_mode = inflation_mode + self.memory = None + super().__init__(*args, **kwargs) + self.temporal_padding = self.padding[0] + self.memory_device = memory_device + self.padding = (0, *self.padding[1:]) + self.memory_limit = float("inf") + self.logged_once = False + + def set_memory_limit(self, value: float): + self.memory_limit = value + + def set_memory_device(self, memory_device): + self.memory_device = memory_device + + def _conv_forward(self, input, weight, bias, *args, **kwargs): + if (NVIDIA_MEMORY_CONV_BUG_WORKAROUND and + weight.dtype in (torch.float16, torch.bfloat16) and + hasattr(torch.backends.cudnn, 'is_available') and + torch.backends.cudnn.is_available() and + getattr(torch.backends.cudnn, 'enabled', True)): + try: + out = torch.cudnn_convolution( + input, weight, self.padding, self.stride, self.dilation, self.groups, + benchmark=False, deterministic=False, allow_tf32=True + ) + if bias is not None: + out += bias.reshape((1, -1) + (1,) * (out.ndim - 2)) + return out + except RuntimeError: + pass + except NotImplementedError: + pass + try: + return super()._conv_forward(input, weight, bias, *args, **kwargs) + except NotImplementedError: + # for: Could not run 'aten::cudnn_convolution' with arguments from the 'CPU' backend + if not self.logged_once: + logging.warning("VAE is on CPU for decoding. This is most likely due to not enough memory") + self.logged_once = True + return F.conv3d(input, weight, bias, *args, **kwargs) + + def memory_limit_conv( + self, + x, + *, + split_dim=3, + padding=(0, 0, 0, 0, 0, 0), + prev_cache=None, + ): + # Compatible with no limit. + if math.isinf(self.memory_limit): + if prev_cache is not None: + x = torch.cat([prev_cache, x], dim=split_dim - 1) + return super().forward(x) + + # Compute tensor shape after concat & padding. + shape = torch.tensor(x.size()) + if prev_cache is not None: + shape[split_dim - 1] += prev_cache.size(split_dim - 1) + shape[-3:] += torch.tensor(padding).view(3, 2).sum(-1).flip(0) + memory_occupy = shape.prod() * x.element_size() / 1024**3 # GiB + if memory_occupy < self.memory_limit or split_dim == x.ndim: + x_concat = x + if prev_cache is not None: + x_concat = torch.cat([prev_cache, x], dim=split_dim - 1) + + def pad_and_forward(): + padded = safe_pad_operation(x_concat, padding, mode='constant', value=0.0) + if not padded.is_contiguous(): + padded = padded.contiguous() + with ignore_padding(self): + return torch.nn.Conv3d.forward(self, padded) + + return pad_and_forward() + + num_splits = math.ceil(memory_occupy / self.memory_limit) + size_per_split = x.size(split_dim) // num_splits + split_sizes = [size_per_split] * (num_splits - 1) + split_sizes += [x.size(split_dim) - sum(split_sizes)] + + x = list(x.split(split_sizes, dim=split_dim)) + if prev_cache is not None: + prev_cache = list(prev_cache.split(split_sizes, dim=split_dim)) + cache = None + for idx in range(len(x)): + if prev_cache is not None: + x[idx] = torch.cat([prev_cache[idx], x[idx]], dim=split_dim - 1) + + lpad_dim = (x[idx].ndim - split_dim - 1) * 2 + rpad_dim = lpad_dim + 1 + padding = list(padding) + padding[lpad_dim] = self.padding[split_dim - 2] if idx == 0 else 0 + padding[rpad_dim] = self.padding[split_dim - 2] if idx == len(x) - 1 else 0 + pad_len = padding[lpad_dim] + padding[rpad_dim] + padding = tuple(padding) + + next_cache = None + cache_len = cache.size(split_dim) if cache is not None else 0 + next_catch_size = get_cache_size( + conv_module=self, + input_len=x[idx].size(split_dim) + cache_len, + pad_len=pad_len, + dim=split_dim - 2, + ) + if next_catch_size != 0: + assert next_catch_size <= x[idx].size(split_dim) + next_cache = ( + x[idx].transpose(0, split_dim)[-next_catch_size:].transpose(0, split_dim) + ) + + x[idx] = self.memory_limit_conv( + x[idx], + split_dim=split_dim + 1, + padding=padding, + prev_cache=cache + ) + + cache = next_cache + + output = torch.cat(x, dim=split_dim) + return output + + def forward( + self, + input, + memory_state: MemoryState = MemoryState.UNSET + ) -> Tensor: + assert memory_state != MemoryState.UNSET + if memory_state != MemoryState.ACTIVE: + self.memory = None + if ( + math.isinf(self.memory_limit) + and torch.is_tensor(input) + ): + return self.basic_forward(input, memory_state) + return self.slicing_forward(input, memory_state) + + def basic_forward(self, input: Tensor, memory_state: MemoryState = MemoryState.UNSET): + mem_size = self.stride[0] - self.kernel_size[0] + if (self.memory is not None) and (memory_state == MemoryState.ACTIVE): + input = extend_head(input, memory=self.memory, times=-1) + else: + input = extend_head(input, times=self.temporal_padding * 2) + memory = ( + input[:, :, mem_size:].detach() + if (mem_size != 0 and memory_state != MemoryState.DISABLED) + else None + ) + if ( + memory_state != MemoryState.DISABLED + and not self.training + and (self.memory_device is not None) + ): + self.memory = memory + if self.memory_device == "cpu" and self.memory is not None: + self.memory = self.memory.to("cpu") + return super().forward(input) + + def slicing_forward( + self, + input, + memory_state: MemoryState = MemoryState.UNSET, + ) -> Tensor: + squeeze_out = False + if torch.is_tensor(input): + input = [input] + squeeze_out = True + + cache_size = self.kernel_size[0] - self.stride[0] + cache = cache_send_recv( + input, cache_size=cache_size, memory=self.memory, times=self.temporal_padding * 2 + ) + + # Single GPU inference - simplified memory management + if ( + memory_state in [MemoryState.INITIALIZING, MemoryState.ACTIVE] # use_slicing + and not self.training + and (self.memory_device is not None) + and cache_size != 0 + ): + if cache_size > input[-1].size(2) and cache is not None and len(input) == 1: + input[0] = torch.cat([cache, input[0]], dim=2) + cache = None + if cache_size <= input[-1].size(2): + self.memory = input[-1][:, :, -cache_size:].detach().contiguous() + if self.memory_device == "cpu" and self.memory is not None: + self.memory = self.memory.to("cpu") + + padding = tuple(x for x in reversed(self.padding) for _ in range(2)) + for i in range(len(input)): + # Prepare cache for next input slice. + next_cache = None + cache_size = 0 + if i < len(input) - 1: + cache_len = cache.size(2) if cache is not None else 0 + cache_size = get_cache_size(self, input[i].size(2) + cache_len, pad_len=0) + if cache_size != 0: + if cache_size > input[i].size(2) and cache is not None: + input[i] = torch.cat([cache, input[i]], dim=2) + cache = None + assert cache_size <= input[i].size(2), f"{cache_size} > {input[i].size(2)}" + next_cache = input[i][:, :, -cache_size:] + + # Conv forward for this input slice. + input[i] = self.memory_limit_conv( + input[i], + padding=padding, + prev_cache=cache + ) + + # Update cache. + cache = next_cache + + return input[0] if squeeze_out else input + +def remove_head(tensor: Tensor, times: int = 1) -> Tensor: + if times == 0: + return tensor + return torch.cat(tensors=(tensor[:, :, :1], tensor[:, :, times + 1 :]), dim=2) + +class Upsample3D(nn.Module): + + def __init__( + self, + channels, + out_channels = None, + inflation_mode = "tail", + temporal_up: bool = False, + spatial_up: bool = True, + slicing: bool = False, + interpolate = True, + name: str = "conv", + use_conv_transpose = False, + use_conv: bool = False, + padding = 1, + bias = True, + kernel_size = None, + **kwargs, + ): + super().__init__() + self.interpolate = interpolate + self.channels = channels + self.out_channels = out_channels or channels + self.use_conv_transpose = use_conv_transpose + self.use_conv = use_conv + self.name = name + + self.conv = None + if use_conv_transpose: + if kernel_size is None: + kernel_size = 4 + self.conv = ops.ConvTranspose2d( + channels, self.out_channels, kernel_size=kernel_size, stride=2, padding=padding, bias=bias + ) + elif use_conv: + if kernel_size is None: + kernel_size = 3 + self.conv = ops.Conv2d(self.channels, self.out_channels, kernel_size=kernel_size, padding=padding, bias=bias) + + conv = self.conv if self.name == "conv" else self.Conv2d_0 + + # Note: lora_layer is not passed into constructor in the original implementation. + # So we make a simplification. + conv = InflatedCausalConv3d( + self.channels, + self.out_channels, + 3, + padding=1, + inflation_mode=inflation_mode, + ) + + self.temporal_up = temporal_up + self.spatial_up = spatial_up + self.temporal_ratio = 2 if temporal_up else 1 + self.spatial_ratio = 2 if spatial_up else 1 + self.slicing = slicing + + assert not self.interpolate + # [Override] MAGViT v2 implementation + if not self.interpolate: + upscale_ratio = (self.spatial_ratio**2) * self.temporal_ratio + self.upscale_conv = ops.Conv3d( + self.channels, self.channels * upscale_ratio, kernel_size=1, padding=0 + ) + identity = ( + torch.eye(self.channels) + .repeat(upscale_ratio, 1) + .reshape_as(self.upscale_conv.weight) + ) + self.upscale_conv.weight.data.copy_(identity) + + if self.name == "conv": + self.conv = conv + else: + self.Conv2d_0 = conv + + self.norm = None + + def forward( + self, + hidden_states: torch.FloatTensor, + memory_state=None, + **kwargs, + ) -> torch.FloatTensor: + assert hidden_states.shape[1] == self.channels + + if hasattr(self, "norm") and self.norm is not None: + # [Overridden] change to causal norm. + hidden_states = causal_norm_wrapper(self.norm, hidden_states) + + if self.use_conv_transpose: + return self.conv(hidden_states) + + if self.slicing: + split_size = hidden_states.size(2) // 2 + hidden_states = list( + hidden_states.split([split_size, hidden_states.size(2) - split_size], dim=2) + ) + else: + hidden_states = [hidden_states] + + for i in range(len(hidden_states)): + hidden_states[i] = self.upscale_conv(hidden_states[i]) + hidden_states[i] = rearrange( + hidden_states[i], + "b (x y z c) f h w -> b c (f z) (h x) (w y)", + x=self.spatial_ratio, + y=self.spatial_ratio, + z=self.temporal_ratio, + ) + + if self.temporal_up and memory_state != MemoryState.ACTIVE: + hidden_states[0] = remove_head(hidden_states[0]) + + if not self.slicing: + hidden_states = hidden_states[0] + + if self.use_conv: + if self.name == "conv": + hidden_states = self.conv(hidden_states, memory_state=memory_state) + else: + hidden_states = self.Conv2d_0(hidden_states, memory_state=memory_state) + + if not self.slicing: + return hidden_states + else: + return torch.cat(hidden_states, dim=2) + + +class Downsample3D(nn.Module): + """A 3D downsampling layer with an optional convolution.""" + + def __init__( + self, + channels, + out_channels = None, + inflation_mode = "tail", + spatial_down: bool = False, + temporal_down: bool = False, + name: str = "conv", + kernel_size=3, + use_conv: bool = False, + padding = 1, + bias=True, + **kwargs, + ): + super().__init__() + self.padding = padding + self.name = name + self.channels = channels + self.out_channels = out_channels or channels + self.temporal_down = temporal_down + self.spatial_down = spatial_down + self.use_conv = use_conv + self.padding = padding + + self.temporal_ratio = 2 if temporal_down else 1 + self.spatial_ratio = 2 if spatial_down else 1 + + self.temporal_kernel = 3 if temporal_down else 1 + self.spatial_kernel = 3 if spatial_down else 1 + + if use_conv: + conv = InflatedCausalConv3d( + self.channels, + self.out_channels, + kernel_size=(self.temporal_kernel, self.spatial_kernel, self.spatial_kernel), + stride=(self.temporal_ratio, self.spatial_ratio, self.spatial_ratio), + padding=( + 1 if self.temporal_down else 0, + self.padding if self.spatial_down else 0, + self.padding if self.spatial_down else 0, + ), + inflation_mode=inflation_mode, + ) + else: + assert self.channels == self.out_channels + conv = nn.AvgPool3d( + kernel_size=(self.temporal_ratio, self.spatial_ratio, self.spatial_ratio), + stride=(self.temporal_ratio, self.spatial_ratio, self.spatial_ratio), + ) + + self.conv = conv + + + def forward( + self, + hidden_states: torch.FloatTensor, + memory_state = None, + **kwargs, + ) -> torch.FloatTensor: + + assert hidden_states.shape[1] == self.channels + + if hasattr(self, "norm") and self.norm is not None: + # [Overridden] change to causal norm. + hidden_states = causal_norm_wrapper(self.norm, hidden_states) + + if self.use_conv and self.padding == 0 and self.spatial_down: + pad = (0, 1, 0, 1) + hidden_states = safe_pad_operation(hidden_states, pad, mode="constant", value=0) + + assert hidden_states.shape[1] == self.channels + + hidden_states = self.conv(hidden_states, memory_state=memory_state) + + return hidden_states + + +class ResnetBlock3D(nn.Module): + def __init__( + self, + in_channels: int, + out_channels: Optional[int] = None, + conv_shortcut: bool = False, + dropout: float = 0.0, + temb_channels: int = 512, + groups: int = 32, + groups_out: Optional[int] = None, + eps: float = 1e-6, + non_linearity: str = "swish", + time_embedding_norm: str = "default", + output_scale_factor: float = 1.0, + skip_time_act: bool = False, + use_in_shortcut: Optional[bool] = None, + up: bool = False, + down: bool = False, + conv_shortcut_bias: bool = True, + conv_2d_out_channels: Optional[int] = None, + inflation_mode = "tail", + time_receptive_field: _receptive_field_t = "half", + slicing: bool = False, + **kwargs, + ): + super().__init__() + self.up = up + self.down = down + self.in_channels = in_channels + self.out_channels = in_channels if out_channels is None else out_channels + conv_2d_out_channels = conv_2d_out_channels or out_channels + self.use_in_shortcut = use_in_shortcut + self.output_scale_factor = output_scale_factor + self.skip_time_act = skip_time_act + self.nonlinearity = nn.SiLU() + if temb_channels is not None: + self.time_emb_proj = ops.Linear(temb_channels, out_channels) + else: + self.time_emb_proj = None + self.norm1 = ops.GroupNorm(num_groups=groups, num_channels=in_channels, eps=eps, affine=True) + if groups_out is None: + groups_out = groups + self.norm2 = ops.GroupNorm(num_groups=groups_out, num_channels=out_channels, eps=eps, affine=True) + self.use_in_shortcut = self.in_channels != out_channels + self.dropout = torch.nn.Dropout(dropout) + self.conv1 = InflatedCausalConv3d( + self.in_channels, + self.out_channels, + kernel_size=(1, 3, 3) if time_receptive_field == "half" else (3, 3, 3), + stride=1, + padding=(0, 1, 1) if time_receptive_field == "half" else (1, 1, 1), + inflation_mode=inflation_mode, + ) + + self.conv2 = InflatedCausalConv3d( + self.out_channels, + conv_2d_out_channels, + kernel_size=3, + stride=1, + padding=1, + inflation_mode=inflation_mode, + ) + + self.upsample = self.downsample = None + if self.up: + self.upsample = Upsample3D( + self.in_channels, + use_conv=False, + inflation_mode=inflation_mode, + slicing=slicing, + ) + elif self.down: + self.downsample = Downsample3D( + self.in_channels, + use_conv=False, + padding=1, + name="op", + inflation_mode=inflation_mode, + ) + + self.conv_shortcut = None + if self.use_in_shortcut: + self.conv_shortcut = InflatedCausalConv3d( + self.in_channels, + conv_2d_out_channels, + kernel_size=1, + stride=1, + padding=0, + bias=True, + inflation_mode=inflation_mode, + ) + + def forward( + self, input_tensor, temb, memory_state = None, **kwargs + ): + hidden_states = input_tensor + + hidden_states = causal_norm_wrapper(self.norm1, hidden_states) + + hidden_states = self.nonlinearity(hidden_states) + + if self.upsample is not None: + if hidden_states.shape[0] >= 64: + input_tensor = input_tensor.contiguous() + hidden_states = hidden_states.contiguous() + input_tensor = self.upsample(input_tensor, memory_state=memory_state) + hidden_states = self.upsample(hidden_states, memory_state=memory_state) + elif self.downsample is not None: + input_tensor = self.downsample(input_tensor, memory_state=memory_state) + hidden_states = self.downsample(hidden_states, memory_state=memory_state) + + hidden_states = self.conv1(hidden_states, memory_state=memory_state) + + if self.time_emb_proj is not None: + if not self.skip_time_act: + temb = self.nonlinearity(temb) + temb = self.time_emb_proj(temb)[:, :, None, None] + + if temb is not None: + hidden_states = hidden_states + temb + + hidden_states = causal_norm_wrapper(self.norm2, hidden_states) + + hidden_states = self.nonlinearity(hidden_states) + + hidden_states = self.dropout(hidden_states) + hidden_states = self.conv2(hidden_states, memory_state=memory_state) + + if self.conv_shortcut is not None: + input_tensor = self.conv_shortcut(input_tensor, memory_state=memory_state) + + output_tensor = (input_tensor + hidden_states) / self.output_scale_factor + + return output_tensor + + +class DownEncoderBlock3D(nn.Module): + def __init__( + self, + in_channels: int, + out_channels: int, + dropout: float = 0.0, + num_layers: int = 1, + resnet_eps: float = 1e-6, + resnet_time_scale_shift: str = "default", + resnet_act_fn: str = "swish", + resnet_groups: int = 32, + resnet_pre_norm: bool = True, + output_scale_factor: float = 1.0, + add_downsample: bool = True, + downsample_padding: int = 1, + inflation_mode = "tail", + time_receptive_field: _receptive_field_t = "half", + temporal_down: bool = True, + spatial_down: bool = True, + ): + super().__init__() + resnets = [] + temporal_modules = [] + + for i in range(num_layers): + in_channels = in_channels if i == 0 else out_channels + resnets.append( + # [Override] Replace module. + ResnetBlock3D( + in_channels=in_channels, + out_channels=out_channels, + temb_channels=None, + eps=resnet_eps, + groups=resnet_groups, + dropout=dropout, + time_embedding_norm=resnet_time_scale_shift, + non_linearity=resnet_act_fn, + output_scale_factor=output_scale_factor, + pre_norm=resnet_pre_norm, + inflation_mode=inflation_mode, + time_receptive_field=time_receptive_field, + ) + ) + temporal_modules.append(nn.Identity()) + + self.resnets = nn.ModuleList(resnets) + self.temporal_modules = nn.ModuleList(temporal_modules) + + if add_downsample: + self.downsamplers = nn.ModuleList( + [ + Downsample3D( + out_channels, + use_conv=True, + out_channels=out_channels, + padding=downsample_padding, + name="op", + temporal_down=temporal_down, + spatial_down=spatial_down, + inflation_mode=inflation_mode, + ) + ] + ) + else: + self.downsamplers = None + + def forward( + self, + hidden_states: torch.FloatTensor, + memory_state = None, + **kwargs, + ) -> torch.FloatTensor: + for resnet, temporal in zip(self.resnets, self.temporal_modules): + hidden_states = resnet(hidden_states, temb=None, memory_state=memory_state) + hidden_states = temporal(hidden_states) + + if self.downsamplers is not None: + for downsampler in self.downsamplers: + hidden_states = downsampler(hidden_states, memory_state=memory_state) + + return hidden_states + + +class UpDecoderBlock3D(nn.Module): + def __init__( + self, + in_channels: int, + out_channels: int, + dropout: float = 0.0, + num_layers: int = 1, + resnet_eps: float = 1e-6, + resnet_time_scale_shift: str = "default", # default, spatial + resnet_act_fn: str = "swish", + resnet_groups: int = 32, + resnet_pre_norm: bool = True, + output_scale_factor: float = 1.0, + add_upsample: bool = True, + temb_channels: Optional[int] = None, + inflation_mode = "tail", + time_receptive_field: _receptive_field_t = "half", + temporal_up: bool = True, + spatial_up: bool = True, + slicing: bool = False, + ): + super().__init__() + resnets = [] + temporal_modules = [] + + for i in range(num_layers): + input_channels = in_channels if i == 0 else out_channels + + resnets.append( + # [Override] Replace module. + ResnetBlock3D( + in_channels=input_channels, + out_channels=out_channels, + temb_channels=temb_channels, + eps=resnet_eps, + groups=resnet_groups, + dropout=dropout, + time_embedding_norm=resnet_time_scale_shift, + non_linearity=resnet_act_fn, + output_scale_factor=output_scale_factor, + pre_norm=resnet_pre_norm, + inflation_mode=inflation_mode, + time_receptive_field=time_receptive_field, + slicing=slicing, + ) + ) + + temporal_modules.append(nn.Identity()) + + self.resnets = nn.ModuleList(resnets) + self.temporal_modules = nn.ModuleList(temporal_modules) + + if add_upsample: + # [Override] Replace module & use learnable upsample + self.upsamplers = nn.ModuleList( + [ + Upsample3D( + out_channels, + use_conv=True, + out_channels=out_channels, + temporal_up=temporal_up, + spatial_up=spatial_up, + interpolate=False, + inflation_mode=inflation_mode, + slicing=slicing, + ) + ] + ) + else: + self.upsamplers = None + + def forward( + self, + hidden_states: torch.FloatTensor, + temb: Optional[torch.FloatTensor] = None, + memory_state=None + ) -> torch.FloatTensor: + for resnet, temporal in zip(self.resnets, self.temporal_modules): + hidden_states = resnet(hidden_states, temb=None, memory_state=memory_state) + hidden_states = temporal(hidden_states) + + if self.upsamplers is not None: + for upsampler in self.upsamplers: + hidden_states = upsampler(hidden_states, memory_state=memory_state) + + return hidden_states + + +class UNetMidBlock3D(nn.Module): + def __init__( + self, + in_channels: int, + temb_channels: int, + dropout: float = 0.0, + num_layers: int = 1, + resnet_eps: float = 1e-6, + resnet_time_scale_shift: str = "default", # default, spatial + resnet_act_fn: str = "swish", + resnet_groups: int = 32, + resnet_pre_norm: bool = True, + add_attention: bool = True, + attention_head_dim: int = 1, + output_scale_factor: float = 1.0, + inflation_mode = "tail", + time_receptive_field: _receptive_field_t = "half", + ): + super().__init__() + resnet_groups = resnet_groups if resnet_groups is not None else min(in_channels // 4, 32) + self.add_attention = add_attention + + # there is always at least one resnet + resnets = [ + # [Override] Replace module. + ResnetBlock3D( + in_channels=in_channels, + out_channels=in_channels, + temb_channels=temb_channels, + eps=resnet_eps, + groups=resnet_groups, + dropout=dropout, + time_embedding_norm=resnet_time_scale_shift, + non_linearity=resnet_act_fn, + output_scale_factor=output_scale_factor, + pre_norm=resnet_pre_norm, + inflation_mode=inflation_mode, + time_receptive_field=time_receptive_field, + ) + ] + attentions = [] + + if attention_head_dim is None: + attention_head_dim = in_channels + + for _ in range(num_layers): + if self.add_attention: + attentions.append( + Attention( + in_channels, + heads=in_channels // attention_head_dim, + dim_head=attention_head_dim, + rescale_output_factor=output_scale_factor, + eps=resnet_eps, + norm_num_groups=( + resnet_groups if resnet_time_scale_shift == "default" else None + ), + spatial_norm_dim=( + temb_channels if resnet_time_scale_shift == "spatial" else None + ), + residual_connection=True, + bias=True, + upcast_softmax=True, + _from_deprecated_attn_block=True, + ) + ) + else: + attentions.append(None) + + resnets.append( + ResnetBlock3D( + in_channels=in_channels, + out_channels=in_channels, + temb_channels=temb_channels, + eps=resnet_eps, + groups=resnet_groups, + dropout=dropout, + time_embedding_norm=resnet_time_scale_shift, + non_linearity=resnet_act_fn, + output_scale_factor=output_scale_factor, + pre_norm=resnet_pre_norm, + inflation_mode=inflation_mode, + time_receptive_field=time_receptive_field, + ) + ) + + self.attentions = nn.ModuleList(attentions) + self.resnets = nn.ModuleList(resnets) + + def forward(self, hidden_states, temb=None, memory_state=None): + video_length, frame_height, frame_width = hidden_states.size()[-3:] + hidden_states = self.resnets[0](hidden_states, temb, memory_state=memory_state) + for attn, resnet in zip(self.attentions, self.resnets[1:]): + if attn is not None: + hidden_states = rearrange(hidden_states, "b c f h w -> (b f) c h w") + hidden_states = attn(hidden_states, temb=temb) + hidden_states = rearrange( + hidden_states, "(b f) c h w -> b c f h w", f=video_length + ) + hidden_states = resnet(hidden_states, temb, memory_state=memory_state) + + return hidden_states + + +class Encoder3D(nn.Module): + def __init__( + self, + in_channels: int = 3, + out_channels: int = 3, + down_block_types: Tuple[str, ...] = ("DownEncoderBlock3D",), + block_out_channels: Tuple[int, ...] = (64,), + layers_per_block: int = 2, + norm_num_groups: int = 32, + act_fn: str = "silu", + double_z: bool = True, + mid_block_add_attention=True, + # [Override] add extra_cond_dim, temporal down num + temporal_down_num: int = 2, + extra_cond_dim: int = None, + gradient_checkpoint: bool = False, + inflation_mode = "tail", + time_receptive_field: _receptive_field_t = "half", + ): + super().__init__() + self.layers_per_block = layers_per_block + self.temporal_down_num = temporal_down_num + + self.conv_in = InflatedCausalConv3d( + in_channels, + block_out_channels[0], + kernel_size=3, + stride=1, + padding=1, + inflation_mode=inflation_mode, + ) + + self.mid_block = None + self.down_blocks = nn.ModuleList([]) + self.extra_cond_dim = extra_cond_dim + + self.conv_extra_cond = nn.ModuleList([]) + + # down + output_channel = block_out_channels[0] + for i, down_block_type in enumerate(down_block_types): + input_channel = output_channel + output_channel = block_out_channels[i] + is_final_block = i == len(block_out_channels) - 1 + # [Override] to support temporal down block design + is_temporal_down_block = i >= len(block_out_channels) - self.temporal_down_num - 1 + # Note: take the last ones + + assert down_block_type == "DownEncoderBlock3D" + + down_block = DownEncoderBlock3D( + num_layers=self.layers_per_block, + in_channels=input_channel, + out_channels=output_channel, + add_downsample=not is_final_block, + resnet_eps=1e-6, + downsample_padding=0, + # Note: Don't know why set it as 0 + resnet_act_fn=act_fn, + resnet_groups=norm_num_groups, + temporal_down=is_temporal_down_block, + spatial_down=True, + inflation_mode=inflation_mode, + time_receptive_field=time_receptive_field, + ) + self.down_blocks.append(down_block) + + def zero_module(module): + # Zero out the parameters of a module and return it. + for p in module.parameters(): + p.detach().zero_() + return module + + self.conv_extra_cond.append( + zero_module( + ops.Conv3d(extra_cond_dim, output_channel, kernel_size=1, stride=1, padding=0) + ) + if self.extra_cond_dim is not None and self.extra_cond_dim > 0 + else None + ) + + # mid + self.mid_block = UNetMidBlock3D( + in_channels=block_out_channels[-1], + resnet_eps=1e-6, + resnet_act_fn=act_fn, + output_scale_factor=1, + resnet_time_scale_shift="default", + attention_head_dim=block_out_channels[-1], + resnet_groups=norm_num_groups, + temb_channels=None, + add_attention=mid_block_add_attention, + inflation_mode=inflation_mode, + time_receptive_field=time_receptive_field, + ) + + # out + self.conv_norm_out = ops.GroupNorm( + num_channels=block_out_channels[-1], num_groups=norm_num_groups, eps=1e-6 + ) + self.conv_act = nn.SiLU() + + conv_out_channels = 2 * out_channels if double_z else out_channels + self.conv_out = InflatedCausalConv3d( + block_out_channels[-1], conv_out_channels, 3, padding=1, inflation_mode=inflation_mode + ) + + self.gradient_checkpointing = gradient_checkpoint + + def forward( + self, + sample: torch.FloatTensor, + extra_cond=None, + memory_state = None + ) -> torch.FloatTensor: + r"""The forward method of the `Encoder` class.""" + sample = sample.to(next(self.parameters()).device) + sample = self.conv_in(sample, memory_state = memory_state) + if self.training and self.gradient_checkpointing: + + def create_custom_forward(module): + def custom_forward(*inputs): + return module(*inputs) + + return custom_forward + + # down + # [Override] add extra block and extra cond + for down_block, extra_block in zip(self.down_blocks, self.conv_extra_cond): + sample = torch.utils.checkpoint.checkpoint( + create_custom_forward(down_block), sample, use_reentrant=False + ) + if extra_block is not None: + sample = sample + safe_interpolate_operation(extra_block(extra_cond), size=sample.shape[2:]) + + # middle + sample = self.mid_block(sample) + + else: + # down + # [Override] add extra block and extra cond + for down_block, extra_block in zip(self.down_blocks, self.conv_extra_cond): + sample = down_block(sample, memory_state=memory_state) + if extra_block is not None: + sample = sample + safe_interpolate_operation(extra_block(extra_cond), size=sample.shape[2:]) + + # middle + sample = self.mid_block(sample, memory_state=memory_state) + + # post-process + sample = causal_norm_wrapper(self.conv_norm_out, sample) + sample = self.conv_act(sample) + sample = self.conv_out(sample, memory_state = memory_state) + + return sample + + +class Decoder3D(nn.Module): + + def __init__( + self, + in_channels: int = 3, + out_channels: int = 3, + up_block_types: Tuple[str, ...] = ("UpDecoderBlock3D",), + block_out_channels: Tuple[int, ...] = (64,), + layers_per_block: int = 2, + norm_num_groups: int = 32, + act_fn: str = "silu", + norm_type: str = "group", # group, spatial + mid_block_add_attention=True, + # [Override] add temporal up block + inflation_mode = "tail", + time_receptive_field: _receptive_field_t = "half", + temporal_up_num: int = 2, + slicing_up_num: int = 0, + gradient_checkpoint: bool = False, + ): + super().__init__() + self.layers_per_block = layers_per_block + self.temporal_up_num = temporal_up_num + + self.conv_in = InflatedCausalConv3d( + in_channels, + block_out_channels[-1], + kernel_size=3, + stride=1, + padding=1, + inflation_mode=inflation_mode, + ) + + self.mid_block = None + self.up_blocks = nn.ModuleList([]) + + temb_channels = in_channels if norm_type == "spatial" else None + + # mid + self.mid_block = UNetMidBlock3D( + in_channels=block_out_channels[-1], + resnet_eps=1e-6, + resnet_act_fn=act_fn, + output_scale_factor=1, + resnet_time_scale_shift="default" if norm_type == "group" else norm_type, + attention_head_dim=block_out_channels[-1], + resnet_groups=norm_num_groups, + temb_channels=temb_channels, + add_attention=mid_block_add_attention, + inflation_mode=inflation_mode, + time_receptive_field=time_receptive_field, + ) + + # up + reversed_block_out_channels = list(reversed(block_out_channels)) + output_channel = reversed_block_out_channels[0] + for i, up_block_type in enumerate(up_block_types): + prev_output_channel = output_channel + output_channel = reversed_block_out_channels[i] + + is_final_block = i == len(block_out_channels) - 1 + is_temporal_up_block = i < self.temporal_up_num + is_slicing_up_block = i >= len(block_out_channels) - slicing_up_num + # Note: Keep symmetric + + assert up_block_type == "UpDecoderBlock3D" + up_block = UpDecoderBlock3D( + num_layers=self.layers_per_block + 1, + in_channels=prev_output_channel, + out_channels=output_channel, + add_upsample=not is_final_block, + resnet_eps=1e-6, + resnet_act_fn=act_fn, + resnet_groups=norm_num_groups, + resnet_time_scale_shift=norm_type, + temb_channels=temb_channels, + temporal_up=is_temporal_up_block, + slicing=is_slicing_up_block, + inflation_mode=inflation_mode, + time_receptive_field=time_receptive_field, + ) + self.up_blocks.append(up_block) + prev_output_channel = output_channel + + # out + if norm_type == "spatial": + self.conv_norm_out = SpatialNorm(block_out_channels[0], temb_channels) + else: + self.conv_norm_out = ops.GroupNorm( + num_channels=block_out_channels[0], num_groups=norm_num_groups, eps=1e-6 + ) + self.conv_act = nn.SiLU() + self.conv_out = InflatedCausalConv3d( + block_out_channels[0], out_channels, 3, padding=1, inflation_mode=inflation_mode + ) + + self.gradient_checkpointing = gradient_checkpoint + + # Note: Just copy from Decoder. + def forward( + self, + sample: torch.FloatTensor, + latent_embeds: Optional[torch.FloatTensor] = None, + memory_state = None, + ) -> torch.FloatTensor: + + sample = sample.to(next(self.parameters()).device) + sample = self.conv_in(sample, memory_state=memory_state) + + upscale_dtype = next(iter(self.up_blocks.parameters())).dtype + # middle + sample = self.mid_block(sample, latent_embeds, memory_state=memory_state) + sample = sample.to(upscale_dtype) + + # up + for up_block in self.up_blocks: + sample = up_block(sample, latent_embeds, memory_state=memory_state) + + # post-process + sample = causal_norm_wrapper(self.conv_norm_out, sample) + sample = self.conv_act(sample) + sample = self.conv_out(sample, memory_state=memory_state) + + return sample + +def wavelet_blur(image: Tensor, radius): + max_safe_radius = max(1, min(image.shape[-2:]) // 8) + if radius > max_safe_radius: + radius = max_safe_radius + + num_channels = image.shape[1] + + kernel_vals = [ + [0.0625, 0.125, 0.0625], + [0.125, 0.25, 0.125], + [0.0625, 0.125, 0.0625], + ] + kernel = torch.tensor(kernel_vals, dtype=image.dtype, device=image.device) + kernel = kernel[None, None].repeat(num_channels, 1, 1, 1) + + image = safe_pad_operation(image, (radius, radius, radius, radius), mode='replicate') + output = F.conv2d(image, kernel, groups=num_channels, dilation=radius) + + return output + +def wavelet_decomposition(image: Tensor, levels: int = 5): + high_freq = torch.zeros_like(image) + + for i in range(levels): + radius = 2 ** i + low_freq = wavelet_blur(image, radius) + high_freq.add_(image).sub_(low_freq) + image = low_freq + + return high_freq, low_freq + +def wavelet_reconstruction(content_feat: Tensor, style_feat: Tensor) -> Tensor: + + if content_feat.shape != style_feat.shape: + # Resize style to match content spatial dimensions + if len(content_feat.shape) >= 3: + # safe_interpolate_operation handles FP16 conversion automatically + style_feat = safe_interpolate_operation( + style_feat, + size=content_feat.shape[-2:], + mode='bilinear', + align_corners=False + ) + + # Decompose both features into frequency components + content_high_freq, content_low_freq = wavelet_decomposition(content_feat) + del content_low_freq # Free memory immediately + + style_high_freq, style_low_freq = wavelet_decomposition(style_feat) + del style_high_freq # Free memory immediately + + if content_high_freq.shape != style_low_freq.shape: + style_low_freq = safe_interpolate_operation( + style_low_freq, + size=content_high_freq.shape[-2:], + mode='bilinear', + align_corners=False + ) + + content_high_freq.add_(style_low_freq) + + return content_high_freq.clamp_(-1.0, 1.0) + +def _histogram_matching_channel(source: Tensor, reference: Tensor, device: torch.device) -> Tensor: + original_shape = source.shape + + # Flatten + source_flat = source.flatten() + reference_flat = reference.flatten() + + # Sort both arrays + source_sorted, source_indices = torch.sort(source_flat) + reference_sorted, _ = torch.sort(reference_flat) + del reference_flat + + # Quantile mapping + n_source = len(source_sorted) + n_reference = len(reference_sorted) + + if n_source == n_reference: + matched_sorted = reference_sorted + else: + # Interpolate reference to match source quantiles + source_quantiles = torch.linspace(0, 1, n_source, device=device) + ref_indices = (source_quantiles * (n_reference - 1)).long() + ref_indices.clamp_(0, n_reference - 1) + matched_sorted = reference_sorted[ref_indices] + del source_quantiles, ref_indices, reference_sorted + + del source_sorted, source_flat + + # Reconstruct using argsort (portable across CUDA/ROCm/MPS) + inverse_indices = torch.argsort(source_indices) + del source_indices + matched_flat = matched_sorted[inverse_indices] + del matched_sorted, inverse_indices + + return matched_flat.reshape(original_shape) + +def _lab_to_rgb_batch(lab: Tensor, device: torch.device, matrix_inv: Tensor, epsilon: float, kappa: float) -> Tensor: + """Convert batch of CIELAB images to RGB color space.""" + L, a, b = lab[:, 0], lab[:, 1], lab[:, 2] + + # LAB to XYZ + fy = (L + 16.0) / 116.0 + fx = a.div(500.0).add_(fy) + fz = fy - b / 200.0 + del L, a, b + + # XYZ transformation + x = torch.where( + fx > epsilon, + torch.pow(fx, 3.0), + fx.mul(116.0).sub_(16.0).div_(kappa) + ) + y = torch.where( + fy > epsilon, + torch.pow(fy, 3.0), + fy.mul(116.0).sub_(16.0).div_(kappa) + ) + z = torch.where( + fz > epsilon, + torch.pow(fz, 3.0), + fz.mul(116.0).sub_(16.0).div_(kappa) + ) + del fx, fy, fz + + # Apply D65 white point (in-place) + x.mul_(0.95047) + # y *= 1.00000 # (no-op, skip) + z.mul_(1.08883) + + xyz = torch.stack([x, y, z], dim=1) + del x, y, z + + # Matrix multiplication: XYZ -> RGB + B, C, H, W = xyz.shape + xyz_flat = xyz.permute(0, 2, 3, 1).reshape(-1, 3) + del xyz + + # Ensure dtype consistency for matrix multiplication + xyz_flat = xyz_flat.to(dtype=matrix_inv.dtype) + rgb_linear_flat = torch.matmul(xyz_flat, matrix_inv.T) + del xyz_flat + + rgb_linear = rgb_linear_flat.reshape(B, H, W, 3).permute(0, 3, 1, 2) + del rgb_linear_flat + + # Apply inverse gamma correction (delinearize) + mask = rgb_linear > 0.0031308 + rgb = torch.where( + mask, + torch.pow(torch.clamp(rgb_linear, min=0.0), 1.0 / 2.4).mul_(1.055).sub_(0.055), + rgb_linear * 12.92 + ) + del mask, rgb_linear + + return torch.clamp(rgb, 0.0, 1.0) + +def _rgb_to_lab_batch(rgb: Tensor, device: torch.device, matrix: Tensor, epsilon: float, kappa: float) -> Tensor: + """Convert batch of RGB images to CIELAB color space using D65 illuminant.""" + # Apply sRGB gamma correction (linearize) + mask = rgb > 0.04045 + rgb_linear = torch.where( + mask, + torch.pow((rgb + 0.055) / 1.055, 2.4), + rgb / 12.92 + ) + del mask + + # Matrix multiplication: RGB -> XYZ + B, C, H, W = rgb_linear.shape + rgb_flat = rgb_linear.permute(0, 2, 3, 1).reshape(-1, 3) + del rgb_linear + + # Ensure dtype consistency for matrix multiplication + rgb_flat = rgb_flat.to(dtype=matrix.dtype) + xyz_flat = torch.matmul(rgb_flat, matrix.T) + del rgb_flat + + xyz = xyz_flat.reshape(B, H, W, 3).permute(0, 3, 1, 2) + del xyz_flat + + # Normalize by D65 white point (in-place) + xyz[:, 0].div_(0.95047) # X + # xyz[:, 1] /= 1.00000 # Y (no-op, skip) + xyz[:, 2].div_(1.08883) # Z + + # XYZ to LAB transformation + epsilon_cubed = epsilon ** 3 + mask = xyz > epsilon_cubed + f_xyz = torch.where( + mask, + torch.pow(xyz, 1.0 / 3.0), + xyz.mul(kappa).add_(16.0).div_(116.0) + ) + del xyz, mask + + # Extract channels and compute LAB + L = f_xyz[:, 1].mul(116.0).sub_(16.0) # Lightness [0, 100] + a = (f_xyz[:, 0] - f_xyz[:, 1]).mul_(500.0) # Green-Red [-128, 127] + b = (f_xyz[:, 1] - f_xyz[:, 2]).mul_(200.0) # Blue-Yellow [-128, 127] + del f_xyz + + return torch.stack([L, a, b], dim=1) + +def lab_color_transfer( + content_feat: Tensor, + style_feat: Tensor, + luminance_weight: float = 0.8 +) -> Tensor: + content_feat = wavelet_reconstruction(content_feat, style_feat) + + if content_feat.shape != style_feat.shape: + style_feat = safe_interpolate_operation( + style_feat, + size=content_feat.shape[-2:], + mode='bilinear', + align_corners=False + ) + + device = content_feat.device + + def ensure_float32_precision(c): + orig_dtype = c.dtype + c = c.float() + return c, orig_dtype + content_feat, original_dtype = ensure_float32_precision(content_feat) + style_feat, _ = ensure_float32_precision(style_feat) + + rgb_to_xyz_matrix = torch.tensor([ + [0.4124564, 0.3575761, 0.1804375], + [0.2126729, 0.7151522, 0.0721750], + [0.0193339, 0.1191920, 0.9503041] + ], dtype=torch.float32, device=device) + + xyz_to_rgb_matrix = torch.tensor([ + [ 3.2404542, -1.5371385, -0.4985314], + [-0.9692660, 1.8760108, 0.0415560], + [ 0.0556434, -0.2040259, 1.0572252] + ], dtype=torch.float32, device=device) + + epsilon = 6.0 / 29.0 + kappa = (29.0 / 3.0) ** 3 + + content_feat.add_(1.0).mul_(0.5).clamp_(0.0, 1.0) + style_feat.add_(1.0).mul_(0.5).clamp_(0.0, 1.0) + + # Convert to LAB color space + content_lab = _rgb_to_lab_batch(content_feat, device, rgb_to_xyz_matrix, epsilon, kappa) + del content_feat + + style_lab = _rgb_to_lab_batch(style_feat, device, rgb_to_xyz_matrix, epsilon, kappa) + del style_feat, rgb_to_xyz_matrix + + # Match chrominance channels (a*, b*) for accurate color transfer + matched_a = _histogram_matching_channel(content_lab[:, 1], style_lab[:, 1], device) + matched_b = _histogram_matching_channel(content_lab[:, 2], style_lab[:, 2], device) + + # Handle luminance with weighted blending + if luminance_weight < 1.0: + # Partially match luminance for better overall color accuracy + matched_L = _histogram_matching_channel(content_lab[:, 0], style_lab[:, 0], device) + # Blend: preserve some content L* for detail, adopt some style L* for color + result_L = content_lab[:, 0].mul(luminance_weight).add_(matched_L.mul(1.0 - luminance_weight)) + del matched_L + else: + # Fully preserve content luminance + result_L = content_lab[:, 0] + + del content_lab, style_lab + + # Reconstruct LAB with corrected channels + result_lab = torch.stack([result_L, matched_a, matched_b], dim=1) + del result_L, matched_a, matched_b + + # Convert back to RGB + result_rgb = _lab_to_rgb_batch(result_lab, device, xyz_to_rgb_matrix, epsilon, kappa) + del result_lab, xyz_to_rgb_matrix + + # Convert back to [-1, 1] range (in-place) + result = result_rgb.mul_(2.0).sub_(1.0) + del result_rgb + + result = result.to(original_dtype) + + return result + + +def wavelet_color_transfer(content_feat: Tensor, style_feat: Tensor) -> Tensor: + return wavelet_reconstruction(content_feat, style_feat) + + +def adain_color_transfer(content_feat: Tensor, style_feat: Tensor, eps: float = 1e-5) -> Tensor: + if content_feat.shape != style_feat.shape: + style_feat = safe_interpolate_operation( + style_feat, + size=content_feat.shape[-2:], + mode='bilinear', + align_corners=False, + ) + + original_dtype = content_feat.dtype + content_feat = content_feat.float() + style_feat = style_feat.float() + + b, c = content_feat.shape[:2] + content_flat = content_feat.reshape(b, c, -1) + style_flat = style_feat.reshape(b, c, -1) + + content_mean = content_flat.mean(dim=2).reshape(b, c, 1, 1) + content_std = (content_flat.var(dim=2, correction=0) + eps).sqrt().reshape(b, c, 1, 1) + style_mean = style_flat.mean(dim=2).reshape(b, c, 1, 1) + style_std = (style_flat.var(dim=2, correction=0) + eps).sqrt().reshape(b, c, 1, 1) + del content_flat, style_flat + + normalized = (content_feat - content_mean) / content_std + del content_mean, content_std + result = normalized * style_std + style_mean + del normalized, style_mean, style_std + + result = result.clamp_(-1.0, 1.0) + if result.dtype != original_dtype: + result = result.to(original_dtype) + return result + + +class VideoAutoencoderKL(nn.Module): + def __init__( + self, + in_channels: int = 3, + out_channels: int = 3, + layers_per_block: int = 2, + act_fn: str = "silu", + latent_channels: int = 16, + norm_num_groups: int = 32, + attention: bool = True, + temporal_scale_num: int = 2, + slicing_up_num: int = 0, + gradient_checkpoint: bool = False, + inflation_mode = "pad", + time_receptive_field: _receptive_field_t = "full", + use_quant_conv: bool = False, + use_post_quant_conv: bool = False, + slicing_sample_min_size = 4, + *args, + **kwargs, + ): + self.slicing_sample_min_size = slicing_sample_min_size + self.slicing_latent_min_size = slicing_sample_min_size // (2**temporal_scale_num) + extra_cond_dim = kwargs.pop("extra_cond_dim") if "extra_cond_dim" in kwargs else None + block_out_channels = (128, 256, 512, 512) + down_block_types = ("DownEncoderBlock3D",) * 4 + up_block_types = ("UpDecoderBlock3D",) * 4 + super().__init__() + + # pass init params to Encoder + self.encoder = Encoder3D( + in_channels=in_channels, + out_channels=latent_channels, + down_block_types=down_block_types, + block_out_channels=block_out_channels, + layers_per_block=layers_per_block, + act_fn=act_fn, + norm_num_groups=norm_num_groups, + double_z=True, + extra_cond_dim=extra_cond_dim, + # [Override] add temporal_down_num parameter + temporal_down_num=temporal_scale_num, + gradient_checkpoint=gradient_checkpoint, + inflation_mode=inflation_mode, + time_receptive_field=time_receptive_field, + ) + + # pass init params to Decoder + self.decoder = Decoder3D( + in_channels=latent_channels, + out_channels=out_channels, + up_block_types=up_block_types, + block_out_channels=block_out_channels, + layers_per_block=layers_per_block, + norm_num_groups=norm_num_groups, + act_fn=act_fn, + # [Override] add temporal_up_num parameter + temporal_up_num=temporal_scale_num, + slicing_up_num=slicing_up_num, + gradient_checkpoint=gradient_checkpoint, + inflation_mode=inflation_mode, + time_receptive_field=time_receptive_field, + ) + + self.quant_conv = ( + InflatedCausalConv3d( + in_channels=2 * latent_channels, + out_channels=2 * latent_channels, + kernel_size=1, + inflation_mode=inflation_mode, + ) + if use_quant_conv + else None + ) + self.post_quant_conv = ( + InflatedCausalConv3d( + in_channels=latent_channels, + out_channels=latent_channels, + kernel_size=1, + inflation_mode=inflation_mode, + ) + if use_post_quant_conv + else None + ) + + # A hacky way to remove attention. + if not attention: + self.encoder.mid_block.attentions = torch.nn.ModuleList([None]) + self.decoder.mid_block.attentions = torch.nn.ModuleList([None]) + + self.use_slicing = True + + def encode(self, x: torch.FloatTensor, return_dict: bool = True): + h = self.slicing_encode(x) + posterior = DiagonalGaussianDistribution(h).mode() + + if not return_dict: + return (posterior,) + + return posterior + + def decode_( + self, z: torch.Tensor, return_dict: bool = True + ): + decoded = self.slicing_decode(z) + + if not return_dict: + return (decoded,) + + return decoded + + def _encode( + self, x, memory_state = MemoryState.DISABLED + ) -> torch.Tensor: + _x = x.to(self.device) + h = self.encoder(_x, memory_state=memory_state) + if self.quant_conv is not None: + output = self.quant_conv(h, memory_state=memory_state) + else: + output = h + return output.to(x.device) + + def _decode( + self, z, memory_state = MemoryState.DISABLED + ) -> torch.Tensor: + _z = z.to(self.device) + + if self.post_quant_conv is not None: + _z = self.post_quant_conv(_z, memory_state=memory_state) + + output = self.decoder(_z, memory_state=memory_state) + return output.to(z.device) + + def slicing_encode(self, x: torch.Tensor) -> torch.Tensor: + sp_size =1 + if self.use_slicing and (x.shape[2] - 1) > self.slicing_sample_min_size * sp_size: + split_size = max( + self.slicing_sample_min_size * sp_size, + getattr(self, "temporal_downsample_factor", 1), + ) + x_slices = list(x[:, :, 1:].split(split_size=split_size, dim=2)) + min_active_len = getattr(self, "temporal_downsample_factor", 1) + if len(x_slices) > 1 and x_slices[-1].shape[2] < min_active_len: + x_slices[-2] = torch.cat((x_slices[-2], x_slices[-1]), dim=2) + x_slices.pop() + encoded_slices = [ + self._encode( + torch.cat((x[:, :, :1], x_slices[0]), dim=2), + memory_state=MemoryState.INITIALIZING, + ) + ] + for x_idx in range(1, len(x_slices)): + encoded_slices.append( + self._encode(x_slices[x_idx], memory_state=MemoryState.ACTIVE) + ) + out = torch.cat(encoded_slices, dim=2) + modules_with_memory = [m for m in self.modules() + if isinstance(m, InflatedCausalConv3d) and m.memory is not None] + for m in modules_with_memory: + m.memory = None + return out + else: + return self._encode(x) + + def slicing_decode(self, z: torch.Tensor) -> torch.Tensor: + sp_size = 1 + if self.use_slicing and (z.shape[2] - 1) > self.slicing_latent_min_size * sp_size: + z_slices = z[:, :, 1:].split(split_size=self.slicing_latent_min_size * sp_size, dim=2) + decoded_slices = [ + self._decode( + torch.cat((z[:, :, :1], z_slices[0]), dim=2), + memory_state=MemoryState.INITIALIZING + ) + ] + for z_idx in range(1, len(z_slices)): + decoded_slices.append( + self._decode(z_slices[z_idx], memory_state=MemoryState.ACTIVE) + ) + out = torch.cat(decoded_slices, dim=2) + modules_with_memory = [m for m in self.modules() + if isinstance(m, InflatedCausalConv3d) and m.memory is not None] + for m in modules_with_memory: + m.memory = None + return out + else: + return self._decode(z) + + def tiled_encode(self, x: torch.Tensor, **kwargs) -> torch.Tensor: + raise NotImplementedError + + def tiled_decode(self, z: torch.Tensor, **kwargs) -> torch.Tensor: + raise NotImplementedError + + def forward( + self, x: torch.FloatTensor, mode: Literal["encode", "decode", "all"] = "all", **kwargs + ): + # x: [b c t h w] + def _unwrap(value): + return value[0] if isinstance(value, tuple) else value + + if mode == "encode": + return _unwrap(self.encode(x)) + elif mode == "decode": + return _unwrap(self.decode_(x)) + else: + latent = _unwrap(self.encode(x)) + return _unwrap(self.decode_(latent)) + +class VideoAutoencoderKLWrapper(VideoAutoencoderKL): + def __init__( + self, + *args, + spatial_downsample_factor = 8, + temporal_downsample_factor = 4, + freeze_encoder = True, + **kwargs, + ): + self.spatial_downsample_factor = spatial_downsample_factor + self.temporal_downsample_factor = temporal_downsample_factor + self.freeze_encoder = freeze_encoder + self.enable_tiling = False + super().__init__(*args, **kwargs) + self.set_memory_limit(0.5, 0.5) + + def forward(self, x: torch.FloatTensor): + with torch.no_grad() if self.freeze_encoder else nullcontext(): + z, p = self.encode(x) + x = self.decode(z) + return x, z, p + + def encode(self, x, orig_dims=None): + if x.ndim == 4: + x = x.unsqueeze(2) + x = x.to(dtype=next(self.parameters()).dtype) + self.device = x.device + p = super().encode(x) + z = p.squeeze(2) + return z, p + + def decode(self, z, seedvr2_tiling=None): + seedvr2_tiling = {} if seedvr2_tiling is None else seedvr2_tiling + if not isinstance(seedvr2_tiling, dict): + raise RuntimeError( + "SeedVR2 VideoAutoencoderKLWrapper.decode: `seedvr2_tiling` must be a dict; " + f"got {type(seedvr2_tiling).__name__} with value {seedvr2_tiling!r}." + ) + + if z.ndim == 5: + b, c, t_latent, h, w = z.shape + if c != 16: + raise RuntimeError( + "SeedVR2 VideoAutoencoderKLWrapper.decode: 5-D latent input must " + f"have 16 channels; got shape {tuple(z.shape)}." + ) + latent = z + elif z.ndim == 4: + b, tc, h, w = z.shape + if tc % 16 != 0: + raise RuntimeError( + "SeedVR2 VideoAutoencoderKLWrapper.decode: 4-D latent input must " + "use collapsed channel layout (B, 16*T, H, W); " + f"got shape {tuple(z.shape)}." + ) + latent = z.reshape(b, 16, -1, h, w) + else: + raise RuntimeError( + "SeedVR2 VideoAutoencoderKLWrapper.decode: latent input must be " + "4-D collapsed (B, 16*T, H, W) or 5-D (B, 16, T, H, W); " + f"got shape {tuple(z.shape)}." + ) + scale = 0.9152 + shift = 0 + latent = latent / scale + shift + + self.device = latent.device + self.enable_tiling = seedvr2_tiling.get("enable_tiling", False) + + if self.enable_tiling: + decode_seedvr2_args = dict(seedvr2_tiling) + tile_h, tile_w = decode_seedvr2_args.get("tile_size", (512, 512)) + ov_h, ov_w = decode_seedvr2_args.get("tile_overlap", (64, 64)) + decode_seedvr2_args["tile_overlap"] = ( + min(ov_h, max(0, tile_h - 8)), + min(ov_w, max(0, tile_w - 8)), + ) + x = tiled_vae(latent, self, **decode_seedvr2_args, encode=False) + if x.ndim == 4: + # tiled_vae squeezes the temporal axis when + # temporal_downsample_factor == 1 AND latent T == 1 + # (see tiled_vae line 179-180); re-add it so the post-decode + # pipeline can keep batch and time distinct on the tiled path. + x = x.unsqueeze(2) + else: + x = super().decode_(latent) + + # ensure even dims for save video + h, w = x.shape[-2:] + w2 = w - (w % 2) + h2 = h - (h % 2) + x = x[..., :h2, :w2] + + return x + + def set_memory_limit(self, conv_max_mem: Optional[float], norm_max_mem: Optional[float], memory_device = "same"): + set_norm_limit(norm_max_mem) + for m in self.modules(): + if isinstance(m, InflatedCausalConv3d): + m.set_memory_limit(conv_max_mem if conv_max_mem is not None else float("inf")) + + for module in self.modules(): + if isinstance(module, InflatedCausalConv3d): + module.set_memory_device(memory_device) diff --git a/comfy/model_base.py b/comfy/model_base.py index 205178911..032eec96a 100644 --- a/comfy/model_base.py +++ b/comfy/model_base.py @@ -53,6 +53,8 @@ import comfy.ldm.pixeldit.model import comfy.ldm.pixeldit.pid import comfy.ldm.ace.model import comfy.ldm.omnigen.omnigen2 +import comfy.ldm.seedvr.model + import comfy.ldm.qwen_image.model import comfy.ldm.kandinsky5.model import comfy.ldm.anima.model @@ -926,6 +928,16 @@ class HunyuanDiT(BaseModel): out['image_meta_size'] = comfy.conds.CONDRegular(torch.FloatTensor([[height, width, target_height, target_width, 0, 0]])) return out +class SeedVR2(BaseModel): + def __init__(self, model_config, model_type=ModelType.FLOW, device=None): + super().__init__(model_config, model_type, device, comfy.ldm.seedvr.model.NaDiT) + def extra_conds(self, **kwargs): + out = super().extra_conds(**kwargs) + condition = kwargs.get("condition", None) + if condition is not None: + out["condition"] = comfy.conds.CONDRegular(condition) + return out + class PixArt(BaseModel): def __init__(self, model_config, model_type=ModelType.EPS, device=None): super().__init__(model_config, model_type, device=device, unet_model=comfy.ldm.pixart.pixartms.PixArtMS) diff --git a/comfy/model_detection.py b/comfy/model_detection.py index f0db7d388..0d221b2eb 100644 --- a/comfy/model_detection.py +++ b/comfy/model_detection.py @@ -594,6 +594,56 @@ def detect_unet_config(state_dict, key_prefix, metadata=None): return dit_config + if "{}blocks.35.mlp.vid.proj_in.weight".format(key_prefix) in state_dict_keys and state_dict["{}blocks.35.mlp.vid.proj_in.weight".format(key_prefix)].shape[1] == 3072: # seedvr2 7b + dit_config = {} + dit_config["image_model"] = "seedvr2" + dit_config["vid_dim"] = 3072 + dit_config["heads"] = 24 + dit_config["num_layers"] = 36 + # 7B uses non-shared MMModule layout (separate ``vid.`` / ``txt.`` + # submodules) at EVERY block — verified by inspecting the 7B + # state_dict at ``blocks.31.ada.txt.attn_gate`` (txt. prefix means + # ``MMModule.shared_weights=False``). Native NaDiT computes + # per-block ``shared_weights = not (i < mm_layers)``, so to keep + # every block non-shared we set ``mm_layers = num_layers``. + # Without this, blocks at index >= mm_layers (default 10) try to + # load ``blocks.N.*.all.*`` keys that don't exist in the file, + # silently miss-load → all-black output. + dit_config["mm_layers"] = 36 + dit_config["norm_eps"] = 1e-5 + dit_config["qk_rope"] = True + dit_config["rope_type"] = "rope3d" + dit_config["rope_dim"] = 64 + dit_config["mlp_type"] = "normal" + return dit_config + elif "{}blocks.35.mlp.all.proj_in_gate.weight".format(key_prefix) in state_dict_keys: # seedvr2 7b + dit_config = {} + dit_config["image_model"] = "seedvr2" + dit_config["vid_dim"] = 3072 + dit_config["heads"] = 24 + dit_config["num_layers"] = 36 + # This checkpoint layout carries shared ``all.`` MMModule keys. + # Preserve the historical split: the initial blocks use separate + # vid/txt modules, later blocks use shared modules. + dit_config["mm_layers"] = 10 + dit_config["norm_eps"] = 1e-5 + dit_config["qk_rope"] = True + dit_config["rope_type"] = "rope3d" + dit_config["rope_dim"] = 64 + dit_config["mlp_type"] = "swiglu" + return dit_config + elif "{}blocks.31.mlp.all.proj_in_gate.weight".format(key_prefix) in state_dict_keys: # seedvr2 3b + dit_config = {} + dit_config["image_model"] = "seedvr2" + dit_config["vid_dim"] = 2560 + dit_config["heads"] = 20 + dit_config["num_layers"] = 32 + dit_config["norm_eps"] = 1.0e-05 + dit_config["qk_rope"] = None + dit_config["mlp_type"] = "swiglu" + dit_config["vid_out_norm"] = True + return dit_config + if '{}head.modulation'.format(key_prefix) in state_dict_keys: # Wan 2.1 dit_config = {} dit_config["image_model"] = "wan2.1" diff --git a/comfy/sample.py b/comfy/sample.py index 2be0cae5f..de71596b3 100644 --- a/comfy/sample.py +++ b/comfy/sample.py @@ -44,7 +44,13 @@ def fix_empty_latent_channels(model, latent_image, downscale_ratio_spacial=None, is_empty = torch.count_nonzero(latent_image) == 0 if is_empty: if latent_format.latent_channels != latent_image.shape[1]: - latent_image = comfy.utils.repeat_to_batch_size(latent_image, latent_format.latent_channels, dim=1) + preserves_collapsed_channels = ( + getattr(latent_format, "preserve_empty_channel_multiples", False) + and latent_image.ndim == 4 + and latent_image.shape[1] % latent_format.latent_channels == 0 + ) + if not preserves_collapsed_channels: + latent_image = comfy.utils.repeat_to_batch_size(latent_image, latent_format.latent_channels, dim=1) if downscale_ratio_spacial is not None: if downscale_ratio_spacial != latent_format.spacial_downscale_ratio: ratio = downscale_ratio_spacial / latent_format.spacial_downscale_ratio diff --git a/comfy/sd.py b/comfy/sd.py index 30b877b85..c4be183ac 100644 --- a/comfy/sd.py +++ b/comfy/sd.py @@ -1,3 +1,4 @@ +import inspect import json import torch from enum import Enum @@ -16,6 +17,7 @@ import comfy.ldm.cosmos.vae import comfy.ldm.wan.vae import comfy.ldm.wan.vae2_2 import comfy.ldm.hunyuan3d.vae +import comfy.ldm.seedvr.vae import comfy.ldm.ace.vae.music_dcae_pipeline import comfy.ldm.cogvideo.vae import comfy.ldm.hunyuan_video.vae @@ -82,6 +84,36 @@ import comfy.latent_formats import comfy.ldm.flux.redux +SEEDVR2_VAE_DECODE_BYTES_PER_OUTPUT_PIXEL = 160 + + +def _seedvr2_vae_decode_output_pixels(latent_t, latent_h, latent_w): + output_t = max(1, (latent_t - 1) * 4 + 1) + return output_t * latent_h * 8 * latent_w * 8 + + +def _seedvr2_vae_decode_memory_used(shape): + if len(shape) == 5: + candidates = [] + if shape[1] == 16: + candidates.append((shape[2], shape[3], shape[4])) + if shape[-1] == 16: + candidates.append((shape[1], shape[2], shape[3])) + if len(candidates) == 0: + candidates.append((shape[2], shape[3], shape[4])) + output_pixels = max(_seedvr2_vae_decode_output_pixels(*candidate) for candidate in candidates) + elif len(shape) == 4: + latent_t = max(1, (shape[1] + 15) // 16) + latent_h, latent_w = shape[2], shape[3] + output_pixels = _seedvr2_vae_decode_output_pixels(latent_t, latent_h, latent_w) + else: + latent_t, latent_h, latent_w = 1, shape[-2], shape[-1] + output_pixels = _seedvr2_vae_decode_output_pixels(latent_t, latent_h, latent_w) + # SeedVR2 decode performs full-frame LAB histogram matching: fp32 channels + # plus int64 sort indices dominate peak memory, not the VAE weight dtype. + return output_pixels * SEEDVR2_VAE_DECODE_BYTES_PER_OUTPUT_PIXEL + + def load_lora_for_models(model, clip, lora, strength_model, strength_clip, lora_metadata=None): key_map = {} if model is not None: @@ -465,8 +497,10 @@ class CLIP: class VAE: def __init__(self, sd=None, device=None, config=None, dtype=None, metadata=None): - if 'decoder.up_blocks.0.resnets.0.norm1.weight' in sd.keys(): #diffusers format - sd = diffusers_convert.convert_vae_state_dict(sd) + is_seedvr2_vae = "decoder.up_blocks.2.upsamplers.0.upscale_conv.weight" in sd + if not is_seedvr2_vae and 'decoder.up_blocks.0.resnets.0.norm1.weight' in sd.keys(): #diffusers format + if metadata is None or metadata.get("keep_diffusers_format") != "true": + sd = diffusers_convert.convert_vae_state_dict(sd) if model_management.is_amd(): VAE_KL_MEM_RATIO = 2.73 @@ -538,6 +572,20 @@ class VAE: self.first_stage_model = StageC_coder() self.downscale_ratio = 32 self.latent_channels = 16 + elif "decoder.up_blocks.2.upsamplers.0.upscale_conv.weight" in sd: # seedvr2 + self.first_stage_model = comfy.ldm.seedvr.vae.VideoAutoencoderKLWrapper() + self.latent_channels = 16 + self.latent_dim = 3 + self.disable_offload = True + self.memory_used_decode = lambda shape, dtype: _seedvr2_vae_decode_memory_used(shape) + self.memory_used_encode = lambda shape, dtype: (max(shape[2], 5) * shape[3] * shape[4] * 64) * model_management.dtype_size(dtype) + self.working_dtypes = [torch.float16, torch.bfloat16, torch.float32] + self.downscale_ratio = (lambda a: max(0, math.floor((a + 3) / 4)), 8, 8) + self.downscale_index_formula = (4, 8, 8) + self.upscale_ratio = (lambda a: max(0, a * 4 - 3), 8, 8) + self.upscale_index_formula = (4, 8, 8) + self.process_input = lambda image: image * 2.0 - 1.0 + self.crop_input = False elif "decoder.conv_in.weight" in sd: if sd['decoder.conv_in.weight'].shape[1] == 64: ddconfig = {"block_out_channels": [128, 256, 512, 512, 1024, 1024], "in_channels": 3, "out_channels": 3, "num_res_blocks": 2, "ffactor_spatial": 32, "downsample_match_channel": True, "upsample_match_channel": True} @@ -665,6 +713,7 @@ class VAE: self.downscale_ratio = (lambda a: max(0, math.floor((a + 7) / 8)), 32, 32) self.downscale_index_formula = (8, 32, 32) self.working_dtypes = [torch.bfloat16, torch.float32] + elif "decoder.conv_in.conv.weight" in sd and sd['decoder.conv_in.conv.weight'].shape[1] == 32: ddconfig = {"block_out_channels": [128, 256, 512, 1024, 1024], "in_channels": 3, "out_channels": 3, "num_res_blocks": 2, "ffactor_spatial": 16, "ffactor_temporal": 4, "downsample_match_channel": True, "upsample_match_channel": True} ddconfig['z_channels'] = sd["decoder.conv_in.conv.weight"].shape[1] @@ -994,6 +1043,40 @@ class VAE: decode_fn = lambda a: self.first_stage_model.decode(a.to(self.vae_dtype).to(self.device)).to(dtype=self.vae_output_dtype()) return self.process_output(comfy.utils.tiled_scale_multidim(samples, decode_fn, tile=(tile_t, tile_x, tile_y), overlap=overlap, upscale_amount=self.upscale_ratio, out_channels=self.output_channels, index_formulas=self.upscale_index_formula, output_device=self.output_device)) + def decode_tiled_seedvr2(self, samples, tile_x=32, tile_y=32, overlap=8, tile_t=16, overlap_t=4): + sf_s = getattr(self.first_stage_model, "spatial_downsample_factor", 8) + sf_t = getattr(self.first_stage_model, "temporal_downsample_factor", 4) + if tile_t is None: + tile_t = 16 + if overlap_t is None: + overlap_t = 4 + if tile_t > 0: + temporal_size = tile_t * sf_t + temporal_overlap = max(0, overlap_t) * sf_t + else: + temporal_size = 0 + temporal_overlap = 0 + args = { + "enable_tiling": True, + "tile_size": (tile_y * sf_s, tile_x * sf_s), + "tile_overlap": (overlap * sf_s, overlap * sf_s), + "temporal_size": temporal_size, + "temporal_overlap": temporal_overlap, + } + output = self.first_stage_model.decode( + samples.to(self.vae_dtype).to(self.device), + seedvr2_tiling=args, + ) + return self.process_output(output.to(device=self.output_device, dtype=self.vae_output_dtype(), copy=True)) + + def _format_seedvr2_encoded_samples(self, samples): + if isinstance(self.first_stage_model, comfy.ldm.seedvr.vae.VideoAutoencoderKLWrapper): + if samples.ndim == 4: + samples = samples.unsqueeze(2) + samples = samples.contiguous() + samples = samples * 0.9152 + return samples + def encode_tiled_(self, pixel_samples, tile_x=512, tile_y=512, overlap = 64): steps = pixel_samples.shape[0] * comfy.utils.get_tiled_scale_steps(pixel_samples.shape[3], pixel_samples.shape[2], tile_x, tile_y, overlap) steps += pixel_samples.shape[0] * comfy.utils.get_tiled_scale_steps(pixel_samples.shape[3], pixel_samples.shape[2], tile_x // 2, tile_y * 2, overlap) @@ -1030,6 +1113,36 @@ class VAE: encode_fn = lambda a: self.first_stage_model.encode((self.process_input(a)).to(self.vae_dtype).to(self.device)).to(dtype=self.vae_output_dtype()) return comfy.utils.tiled_scale_multidim(samples, encode_fn, tile=(tile_t, tile_x, tile_y), overlap=overlap, upscale_amount=self.downscale_ratio, out_channels=self.latent_channels, downscale=True, index_formulas=self.downscale_index_formula, output_device=self.output_device) + def encode_tiled_seedvr2(self, pixel_samples, tile_x=None, tile_y=None, overlap=None, tile_t=None, overlap_t=None): + if tile_y is None: + tile_y = 512 + if tile_x is None: + tile_x = 512 + if overlap is None: + overlap_y = 64 + overlap_x = 64 + else: + overlap_y = overlap + overlap_x = overlap + if tile_t is None: + tile_t = 9999 + if overlap_t is None: + overlap_t = 0 + overlap_y = min(overlap_y, max(0, tile_y - 8)) + overlap_x = min(overlap_x, max(0, tile_x - 8)) + self.first_stage_model.device = self.device + x = self.process_input(pixel_samples).to(self.vae_dtype).to(self.device) + output = comfy.ldm.seedvr.vae.tiled_vae( + x, + self.first_stage_model, + tile_size=(tile_y, tile_x), + tile_overlap=(overlap_y, overlap_x), + temporal_size=tile_t, + temporal_overlap=overlap_t, + encode=True, + ) + return output.to(device=self.output_device, dtype=self.vae_output_dtype()) + def decode(self, samples_in, vae_options={}): self.throw_exception_if_invalid() pixel_samples = None @@ -1077,16 +1190,40 @@ class VAE: if dims == 1 or self.extra_1d_channel is not None: pixel_samples = self.decode_tiled_1d(samples_in) elif dims == 2: - pixel_samples = self.decode_tiled_(samples_in) + # SeedVR2 latents arrive in 4D collapsed form ``(B, 16*T, H, W)`` + # downstream of ``SeedVR2Conditioning`` (which performs the + # ``rearrange(b c t h w -> b (c t) h w)`` collapse). The + # generic ``decode_tiled_`` would treat the channel dim as + # spatial-only and crash on the collapsed (16, T) layout + # under ``tiled_scale``'s mask broadcast; route SeedVR2 4D + # latents to ``decode_tiled_seedvr2`` instead, whose wrapper + # dispatch handles both 4D and 5D inputs. + if isinstance(self.first_stage_model, comfy.ldm.seedvr.vae.VideoAutoencoderKLWrapper): + tile = 256 // self.spacial_compression_decode() + overlap = tile // 4 + pixel_samples = self.decode_tiled_seedvr2(samples_in, tile_x=tile, tile_y=tile, overlap=overlap) + else: + pixel_samples = self.decode_tiled_(samples_in) elif dims == 3: tile = 256 // self.spacial_compression_decode() overlap = tile // 4 - pixel_samples = self.decode_tiled_3d(samples_in, tile_x=tile, tile_y=tile, overlap=(1, overlap, overlap)) + if isinstance(self.first_stage_model, comfy.ldm.seedvr.vae.VideoAutoencoderKLWrapper): + pixel_samples = self.decode_tiled_seedvr2(samples_in, tile_x=tile, tile_y=tile, overlap=overlap) + else: + pixel_samples = self.decode_tiled_3d(samples_in, tile_x=tile, tile_y=tile, overlap=(1, overlap, overlap)) pixel_samples = pixel_samples.to(self.output_device).movedim(1,-1) return pixel_samples - def decode_tiled(self, samples, tile_x=None, tile_y=None, overlap=None, tile_t=None, overlap_t=None): + def decode_tiled( + self, + samples, + tile_x=None, + tile_y=None, + overlap=None, + tile_t=None, + overlap_t=None, + ): self.throw_exception_if_invalid() memory_used = self.memory_used_decode(samples.shape, self.vae_dtype) #TODO: calculate mem required for tile model_management.load_models_gpu([self.patcher], memory_required=memory_used, force_full_load=self.disable_offload) @@ -1100,7 +1237,20 @@ class VAE: args["overlap"] = overlap with model_management.cuda_device_context(self.device): - if dims == 1 or self.extra_1d_channel is not None: + if isinstance(self.first_stage_model, comfy.ldm.seedvr.vae.VideoAutoencoderKLWrapper) and dims in (2, 3): + seedvr2_args = {} + if tile_x is not None: + seedvr2_args["tile_x"] = tile_x + if tile_y is not None: + seedvr2_args["tile_y"] = tile_y + if overlap is not None: + seedvr2_args["overlap"] = overlap + if tile_t is not None: + seedvr2_args["tile_t"] = tile_t + if overlap_t is not None: + seedvr2_args["overlap_t"] = overlap_t + output = self.decode_tiled_seedvr2(samples, **seedvr2_args) + elif dims == 1 or self.extra_1d_channel is not None: args.pop("tile_y") output = self.decode_tiled_1d(samples, **args) elif dims == 2: @@ -1142,6 +1292,8 @@ class VAE: else: pixels_in = pixels_in.to(self.device) out = self.first_stage_model.encode(pixels_in) + if isinstance(out, tuple): + out = out[0] out = out.to(self.output_device).to(dtype=self.vae_output_dtype()) if samples is None: samples = torch.empty((pixel_samples.shape[0],) + tuple(out.shape[1:]), device=self.output_device, dtype=self.vae_output_dtype()) @@ -1161,20 +1313,23 @@ class VAE: if self.latent_dim == 3: tile = 256 overlap = tile // 4 - samples = self.encode_tiled_3d(pixel_samples, tile_x=tile, tile_y=tile, overlap=(1, overlap, overlap)) + if isinstance(self.first_stage_model, comfy.ldm.seedvr.vae.VideoAutoencoderKLWrapper): + samples = self.encode_tiled_seedvr2(pixel_samples, tile_x=tile, tile_y=tile, overlap=overlap) + else: + samples = self.encode_tiled_3d(pixel_samples, tile_x=tile, tile_y=tile, overlap=(1, overlap, overlap)) elif self.latent_dim == 1 or self.extra_1d_channel is not None: samples = self.encode_tiled_1d(pixel_samples) else: samples = self.encode_tiled_(pixel_samples) - return samples + return self._format_seedvr2_encoded_samples(samples) def encode_tiled(self, pixel_samples, tile_x=None, tile_y=None, overlap=None, tile_t=None, overlap_t=None): self.throw_exception_if_invalid() pixel_samples = self.vae_encode_crop_pixels(pixel_samples) dims = self.latent_dim pixel_samples = pixel_samples.movedim(-1, 1) - if dims == 3: + if dims == 3 and pixel_samples.ndim < 5: if not self.not_video: pixel_samples = pixel_samples.movedim(1, 0).unsqueeze(0) else: @@ -1198,22 +1353,47 @@ class VAE: elif dims == 2: samples = self.encode_tiled_(pixel_samples, **args) elif dims == 3: - if tile_t is not None: - tile_t_latent = max(2, self.downscale_ratio[0](tile_t)) + if isinstance(self.first_stage_model, comfy.ldm.seedvr.vae.VideoAutoencoderKLWrapper): + seedvr2_args = {} + if tile_x is not None: + seedvr2_args["tile_x"] = tile_x + else: + seedvr2_args["tile_x"] = 512 + if tile_y is not None: + seedvr2_args["tile_y"] = tile_y + else: + seedvr2_args["tile_y"] = 512 + if overlap is not None: + seedvr2_args["overlap"] = overlap + else: + seedvr2_args["overlap"] = 64 + if tile_t is not None: + seedvr2_args["tile_t"] = tile_t + else: + seedvr2_args["tile_t"] = 9999 + if overlap_t is not None: + seedvr2_args["overlap_t"] = overlap_t + else: + seedvr2_args["overlap_t"] = 0 + samples = self.encode_tiled_seedvr2(pixel_samples, **seedvr2_args) else: - tile_t_latent = 9999 - args["tile_t"] = self.upscale_ratio[0](tile_t_latent) + if tile_t is not None: + tile_t_latent = max(2, self.downscale_ratio[0](tile_t)) + else: + tile_t_latent = 9999 + args["tile_t"] = self.upscale_ratio[0](tile_t_latent) - if overlap_t is None: - args["overlap"] = (1, overlap, overlap) - else: - args["overlap"] = (self.upscale_ratio[0](max(1, min(tile_t_latent // 2, self.downscale_ratio[0](overlap_t)))), overlap, overlap) - maximum = pixel_samples.shape[2] - maximum = self.upscale_ratio[0](self.downscale_ratio[0](maximum)) + spatial_overlap = overlap if overlap is not None else 64 + if overlap_t is None: + args["overlap"] = (1, spatial_overlap, spatial_overlap) + else: + args["overlap"] = (self.upscale_ratio[0](max(1, min(tile_t_latent // 2, self.downscale_ratio[0](overlap_t)))), spatial_overlap, spatial_overlap) + maximum = pixel_samples.shape[2] + maximum = self.upscale_ratio[0](self.downscale_ratio[0](maximum)) - samples = self.encode_tiled_3d(pixel_samples[:,:,:maximum], **args) + samples = self.encode_tiled_3d(pixel_samples[:,:,:maximum], **args) - return samples + return self._format_seedvr2_encoded_samples(samples) def get_sd(self): return self.first_stage_model.state_dict() @@ -1735,6 +1915,17 @@ def load_checkpoint(config_path=None, ckpt_path=None, output_vae=True, output_cl return (model, clip, vae) + +def _set_model_config_inference_dtype(model_config, dtype, manual_cast_dtype, device): + set_dtype = model_config.set_inference_dtype + parameters = inspect.signature(set_dtype).parameters + supports_device = "device" in parameters or any(p.kind == inspect.Parameter.VAR_KEYWORD for p in parameters.values()) + if supports_device: + set_dtype(dtype, manual_cast_dtype, device=device) + else: + set_dtype(dtype, manual_cast_dtype) + + def load_checkpoint_guess_config(ckpt_path, output_vae=True, output_clip=True, output_clipvision=False, embedding_directory=None, output_model=True, model_options={}, te_model_options={}, disable_dynamic=False): sd, metadata = comfy.utils.load_torch_file(ckpt_path, return_metadata=True) out = load_state_dict_guess_config(sd, output_vae, output_clip, output_clipvision, embedding_directory, output_model, model_options, te_model_options=te_model_options, metadata=metadata, disable_dynamic=disable_dynamic) @@ -1842,7 +2033,7 @@ def load_state_dict_guess_config(sd, output_vae=True, output_clip=True, output_c manual_cast_dtype = model_management.unet_manual_cast(None, load_device, model_config.supported_inference_dtypes) else: manual_cast_dtype = model_management.unet_manual_cast(unet_dtype, load_device, model_config.supported_inference_dtypes) - model_config.set_inference_dtype(unet_dtype, manual_cast_dtype) + _set_model_config_inference_dtype(model_config, unet_dtype, manual_cast_dtype, load_device) if model_config.clip_vision_prefix is not None: if output_clipvision: @@ -1983,7 +2174,7 @@ def load_diffusion_model_state_dict(sd, model_options={}, metadata=None, disable manual_cast_dtype = model_management.unet_manual_cast(None, load_device, model_config.supported_inference_dtypes) else: manual_cast_dtype = model_management.unet_manual_cast(unet_dtype, load_device, model_config.supported_inference_dtypes) - model_config.set_inference_dtype(unet_dtype, manual_cast_dtype) + _set_model_config_inference_dtype(model_config, unet_dtype, manual_cast_dtype, load_device) if custom_operations is not None: model_config.custom_operations = custom_operations diff --git a/comfy/supported_models.py b/comfy/supported_models.py index 00941da53..30134e6d1 100644 --- a/comfy/supported_models.py +++ b/comfy/supported_models.py @@ -1647,6 +1647,35 @@ class Chroma(supported_models_base.BASE): t5_detect = comfy.text_encoders.sd3_clip.t5_xxl_detect(state_dict, "{}t5xxl.transformer.".format(pref)) return supported_models_base.ClipTarget(comfy.text_encoders.pixart_t5.PixArtTokenizer, comfy.text_encoders.pixart_t5.pixart_te(**t5_detect)) +class SeedVR2(supported_models_base.BASE): + unet_config = { + "image_model": "seedvr2" + } + latent_format = comfy.latent_formats.SeedVR2 + + vae_key_prefix = ["vae."] + text_encoder_key_prefix = ["text_encoders."] + supported_inference_dtypes = [torch.bfloat16, torch.float16, torch.float32] + sampling_settings = { + "shift": 1.0, + } + + def set_inference_dtype(self, dtype, manual_cast_dtype, device=None): + if ( + dtype == torch.float16 + and manual_cast_dtype is None + and comfy.model_management.should_use_bf16(device) + ): + manual_cast_dtype = torch.bfloat16 + super().set_inference_dtype(dtype, manual_cast_dtype, device=device) + + def get_model(self, state_dict, prefix="", device=None): + out = model_base.SeedVR2(self, device=device) + return out + + def clip_target(self, state_dict={}): + return None + class ChromaRadiance(Chroma): unet_config = { "image_model": "chroma_radiance", @@ -1966,7 +1995,6 @@ class LongCatImage(supported_models_base.BASE): hunyuan_detect = comfy.text_encoders.hunyuan_video.llama_detect(state_dict, "{}qwen25_7b.transformer.".format(pref)) return supported_models_base.ClipTarget(comfy.text_encoders.longcat_image.LongCatImageTokenizer, comfy.text_encoders.longcat_image.te(**hunyuan_detect)) - class RT_DETR_v4(supported_models_base.BASE): unet_config = { "image_model": "RT_DETR_v4", @@ -2203,6 +2231,7 @@ models = [ HiDream, HiDreamO1, Chroma, + SeedVR2, ChromaRadiance, ACEStep, ACEStep15, diff --git a/comfy/supported_models_base.py b/comfy/supported_models_base.py index 0e7a829ba..572f9984e 100644 --- a/comfy/supported_models_base.py +++ b/comfy/supported_models_base.py @@ -115,7 +115,7 @@ class BASE: replace_prefix = {"": self.vae_key_prefix[0]} return utils.state_dict_prefix_replace(state_dict, replace_prefix) - def set_inference_dtype(self, dtype, manual_cast_dtype): + def set_inference_dtype(self, dtype, manual_cast_dtype, device=None): self.unet_config['dtype'] = dtype self.manual_cast_dtype = manual_cast_dtype diff --git a/comfy_extras/nodes_seedvr.py b/comfy_extras/nodes_seedvr.py new file mode 100644 index 000000000..6bc2de17f --- /dev/null +++ b/comfy_extras/nodes_seedvr.py @@ -0,0 +1,1164 @@ +from typing_extensions import override +from comfy_api.latest import ComfyExtension, io +import torch +import math +import logging +from einops import rearrange + +import gc +import comfy.model_management +import comfy.sample +import comfy.samplers +from comfy.ldm.seedvr.vae import ( + adain_color_transfer, + lab_color_transfer, + wavelet_color_transfer, +) + +from torchvision.transforms import functional as TVF +from torchvision.transforms import Lambda +from torchvision.transforms.functional import InterpolationMode + + +_SEEDVR2_INVALID_MODEL_MSG_PREFIX = ( + "SeedVR2Conditioning: model object does not match expected SeedVR2 structure" +) +LAB_SCALE_MULTIPLIER = 13 +WAVELET_SCALE_MULTIPLIER = 10 +ADAIN_SCALE_MULTIPLIER = 6 +COLOR_CORRECTION_MEMORY_HEADROOM = 0.75 + +# Private sentinel for getattr default: distinguishes "attribute missing" +# from "attribute present but None" so the failure message is accurate. +_ATTR_MISSING = object() + + +def _seedvr2_auto_chunk_attempts(t_latent, t_pixel, frames_per_chunk): + """Return stricter 4n+1 frame chunk sizes for auto OOM retries.""" + attempts = [frames_per_chunk] + current_chunk_latent = ( + t_latent if t_pixel <= frames_per_chunk + else (frames_per_chunk - 1) // 4 + 1 + ) + current_chunk_count = max(1, math.ceil(t_latent / current_chunk_latent)) + seen = {frames_per_chunk} + + for target_chunks in range(max(2, current_chunk_count + 1), t_latent + 1): + chunk_latent = max(1, math.ceil(t_latent / target_chunks)) + candidate = 4 * (chunk_latent - 1) + 1 + if candidate in seen: + continue + if candidate >= attempts[-1]: + continue + attempts.append(candidate) + seen.add(candidate) + + return attempts + + +def _resolve_seedvr2_diffusion_model(model): + """Resolve the inner SeedVR2 diffusion-model module from a ComfyUI model + patcher object. Fails loud with a ``RuntimeError`` whose message begins + with ``_SEEDVR2_INVALID_MODEL_MSG_PREFIX`` when the expected wrapper + shape (``model.model.diffusion_model``) is absent. + + Distinguishes four failure modes via the ``_ATTR_MISSING`` sentinel: + ``model.model`` missing, ``model.model is None``, + ``model.model.diffusion_model`` missing, ``model.model.diffusion_model + is None``. Each mode produces an accurate error message rather than + conflating "attribute missing" with "attribute is None". + """ + inner = getattr(model, "model", _ATTR_MISSING) + if inner is _ATTR_MISSING: + raise RuntimeError( + f"{_SEEDVR2_INVALID_MODEL_MSG_PREFIX}: input has no 'model' attribute " + f"(got type {type(model).__name__})." + ) + if inner is None: + raise RuntimeError( + f"{_SEEDVR2_INVALID_MODEL_MSG_PREFIX}: input.model is None " + f"(input type {type(model).__name__})." + ) + diffusion_model = getattr(inner, "diffusion_model", _ATTR_MISSING) + if diffusion_model is _ATTR_MISSING: + raise RuntimeError( + f"{_SEEDVR2_INVALID_MODEL_MSG_PREFIX}: 'model.model' has no " + f"'diffusion_model' attribute (got type {type(inner).__name__})." + ) + if diffusion_model is None: + raise RuntimeError( + f"{_SEEDVR2_INVALID_MODEL_MSG_PREFIX}: 'model.model.diffusion_model' " + f"is None (model.model type {type(inner).__name__})." + ) + return diffusion_model + + +def _apply_rope_freqs_float32_cast(diffusion_model): + """Cast every nested module's ``rope.freqs`` parameter data to ``float32`` + when it is not already in float32. Idempotency is per-tensor by dtype + check, NOT a per-instance sentinel attribute — a sentinel would survive + Comfy's dynamic model unload/reload cycle while ``rope.freqs`` itself + is restored from the archived dtype, leaving RoPE running in fp16/bf16 + on subsequent calls. The dtype check makes the cast self-correcting + against weight-restore lifecycle events. Iteration cost is one walk of + the diffusion-model module tree per ``execute()`` call (microseconds). + """ + for module in diffusion_model.modules(): + if hasattr(module, 'rope') and hasattr(module.rope, 'freqs'): + if module.rope.freqs.data.dtype != torch.float32: + module.rope.freqs.data = module.rope.freqs.data.to(torch.float32) + + +def clear_vae_memory(vae_model): + for module in vae_model.modules(): + if hasattr(module, "memory"): + module.memory = None + gc.collect() + comfy.model_management.soft_empty_cache() + +def expand_dims(tensor, ndim): + shape = tensor.shape + (1,) * (ndim - tensor.ndim) + return tensor.reshape(shape) + +def get_conditions(latent, latent_blur): + t, h, w, c = latent.shape + cond = torch.ones([t, h, w, c + 1], device=latent.device, dtype=latent.dtype) + cond[:, ..., :-1] = latent_blur[:] + cond[:, ..., -1:] = 1.0 + return cond + +def timestep_transform(timesteps, latents_shapes): + vt = 4 + vs = 8 + frames = (latents_shapes[:, 0] - 1) * vt + 1 + heights = latents_shapes[:, 1] * vs + widths = latents_shapes[:, 2] * vs + + # Compute shift factor. + def get_lin_function(x1, y1, x2, y2): + m = (y2 - y1) / (x2 - x1) + b = y1 - m * x1 + return lambda x: m * x + b + + img_shift_fn = get_lin_function(x1=256 * 256, y1=1.0, x2=1024 * 1024, y2=3.2) + vid_shift_fn = get_lin_function(x1=256 * 256 * 37, y1=1.0, x2=1280 * 720 * 145, y2=5.0) + shift = torch.where( + frames > 1, + vid_shift_fn(heights * widths * frames), + img_shift_fn(heights * widths), + ).to(timesteps.device) + + # Shift timesteps. + T = 1000.0 + timesteps = timesteps / T + timesteps = shift * timesteps / (1 + (shift - 1) * timesteps) + timesteps = timesteps * T + return timesteps + +def inter(x_0, x_T, t): + t = expand_dims(t, x_0.ndim) + T = 1000.0 + B = lambda t: t / T + A = lambda t: 1 - (t / T) + return A(t) * x_0 + B(t) * x_T +def area_resize(image, max_area): + + height, width = image.shape[-2:] + scale = math.sqrt(max_area / (height * width)) + + resized_height, resized_width = round(height * scale), round(width * scale) + + return TVF.resize( + image, + size=(resized_height, resized_width), + interpolation=InterpolationMode.BICUBIC, + ) + +def div_pad(image, factor): + + height_factor, width_factor = factor + height, width = image.shape[-2:] + + pad_height = (height_factor - (height % height_factor)) % height_factor + pad_width = (width_factor - (width % width_factor)) % width_factor + + if pad_height == 0 and pad_width == 0: + return image + + if isinstance(image, torch.Tensor): + padding = (0, pad_width, 0, pad_height) + image = torch.nn.functional.pad(image, padding, mode='constant', value=0.0) + + return image + +def cut_videos(videos): + t = videos.size(1) + if t == 1: + return videos + if t <= 4 : + padding = [videos[:, -1].unsqueeze(1)] * (4 - t + 1) + padding = torch.cat(padding, dim=1) + videos = torch.cat([videos, padding], dim=1) + return videos + if (t - 1) % (4) == 0: + return videos + else: + padding = [videos[:, -1].unsqueeze(1)] * ( + 4 - ((t - 1) % (4)) + ) + padding = torch.cat(padding, dim=1) + videos = torch.cat([videos, padding], dim=1) + assert (videos.size(1) - 1) % (4) == 0 + return videos + +def side_resize(image, size): + antialias = not (isinstance(image, torch.Tensor) and image.device.type == 'mps') + resized = TVF.resize(image, size, InterpolationMode.BICUBIC, antialias=antialias) + return resized + + +def _seedvr2_input_shorter_edge(images, node_name): + if images.dim() == 4: + return min(images.shape[1], images.shape[2]) + if images.dim() == 5: + return min(images.shape[2], images.shape[3]) + raise ValueError( + f"{node_name}: expected 4-D or 5-D IMAGE tensor, " + f"got shape {tuple(images.shape)}" + ) + + +def _seedvr2_resize_and_pad(images, upscaled_shorter_edge, node_name): + if upscaled_shorter_edge < 2: + raise ValueError( + f"{node_name}: resolved upscaled_shorter_edge must be at least 2 pixels; " + f"got {upscaled_shorter_edge}." + ) + original_image = images + if images.dim() == 4: + # Comfy video components arrive as a 4-D IMAGE frame sequence: + # (frames, H, W, C). SeedVR2 consumes that as one video. + images = images.unsqueeze(0) + elif images.dim() != 5: + raise ValueError( + f"{node_name}: expected 4-D or 5-D IMAGE tensor, " + f"got shape {tuple(images.shape)}" + ) + images = images.permute(0, 1, 4, 2, 3) + + b, t, c, h, w = images.shape + images = images.reshape(b * t, c, h, w) + + clip = Lambda(lambda x: torch.clamp(x, 0.0, 1.0)) + images = side_resize(images, upscaled_shorter_edge) + + images = clip(images) + images = div_pad(images, (16, 16)) + _, _, new_h, new_w = images.shape + + images = images.reshape(b, t, c, new_h, new_w) + images = cut_videos(images) + images_bthwc = rearrange(images, "b t c h w -> b t h w c") + + return io.NodeOutput(images_bthwc, original_image, upscaled_shorter_edge) + + +class SeedVR2Resize(io.ComfyNode): + @classmethod + def define_schema(cls): + return io.Schema( + node_id="SeedVR2Resize", + category="image/video", + inputs=[ + io.Image.Input("images"), + io.Float.Input("multiplier", default=4.0, min=0.01), + ], + outputs=[ + io.Image.Output("input_pixels"), + io.Image.Output("original_image"), + io.Int.Output("upscaled_shorter_edge"), + ] + ) + + @classmethod + def execute(cls, images, multiplier=4.0): + if multiplier <= 0: + raise ValueError( + f"SeedVR2Resize: multiplier must be > 0; got {multiplier}." + ) + shorter_edge = _seedvr2_input_shorter_edge(images, "SeedVR2Resize") + upscaled_shorter_edge = int(round(shorter_edge * multiplier)) + if upscaled_shorter_edge < 2: + raise ValueError( + "SeedVR2Resize: multiplier resolved upscaled_shorter_edge " + f"to {upscaled_shorter_edge}; use a multiplier that resolves " + "to at least 2 pixels." + ) + return _seedvr2_resize_and_pad( + images, upscaled_shorter_edge, "SeedVR2Resize", + ) + + +class SeedVR2ResizeAdvanced(io.ComfyNode): + @classmethod + def define_schema(cls): + return io.Schema( + node_id="SeedVR2ResizeAdvanced", + category="image/video", + inputs=[ + io.Image.Input("images"), + io.Int.Input("shorter_edge", default=1280, min=2), + ], + outputs=[ + io.Image.Output("input_pixels"), + io.Image.Output("original_image"), + io.Int.Output("upscaled_shorter_edge"), + ] + ) + + @classmethod + def execute(cls, images, shorter_edge): + return _seedvr2_resize_and_pad( + images, shorter_edge, "SeedVR2ResizeAdvanced", + ) + + +class SeedVR2PostProcessing(io.ComfyNode): + @classmethod + def define_schema(cls): + return io.Schema( + node_id="SeedVR2PostProcessing", + category="image/video", + inputs=[ + io.Image.Input("decoded"), + io.Image.Input("original_image"), + io.Int.Input("upscaled_shorter_edge", min=2, force_input=True), + io.Combo.Input("color_correction_method", options=["lab", "wavelet", "adain", "none"], default="lab"), + ], + outputs=[io.Image.Output()], + ) + + @classmethod + def execute(cls, decoded, original_image, upscaled_shorter_edge, color_correction_method): + cls._validate_upscaled_shorter_edge(upscaled_shorter_edge) + decoded_5d, decoded_was_4d = cls._as_bthwc(decoded) + original_5d, _ = cls._as_bthwc(original_image) + decoded_5d = cls._restore_reference_batch_time(decoded_5d, original_5d) + + b = min(decoded_5d.shape[0], original_5d.shape[0]) + t = min(decoded_5d.shape[1], original_5d.shape[1]) + reference_h, reference_w = cls._resized_shorter_edge_dims( + original_5d.shape[2], original_5d.shape[3], upscaled_shorter_edge, + ) + + decoded_5d = decoded_5d[:b, :t, :, :, :] + target_h = min(decoded_5d.shape[2], reference_h) + target_w = min(decoded_5d.shape[3], reference_w) + decoded_5d = decoded_5d[:, :, :target_h, :target_w, :] + if color_correction_method in ("lab", "wavelet", "adain"): + reference_5d = cls._resize_original_reference(original_image, upscaled_shorter_edge) + reference_5d = reference_5d[:b, :t, :, :, :] + reference_5d = cls._resize_reference(reference_5d, target_h, target_w) + output_device = decoded_5d.device + decoded_raw = cls._to_seedvr2_raw(decoded_5d) + reference_raw = cls._to_seedvr2_raw(reference_5d) + decoded_flat = rearrange(decoded_raw, "b t h w c -> (b t) c h w") + reference_flat = rearrange(reference_raw, "b t h w c -> (b t) c h w") + output = cls._color_transfer_chunked( + decoded_flat, reference_flat, output_device, color_correction_method, + ) + output = rearrange(output, "(b t) c h w -> b t h w c", b=b, t=t) + output = output.add(1.0).div(2.0).clamp(0.0, 1.0) + elif color_correction_method == "none": + output = decoded_5d + else: + raise ValueError(f"SeedVR2PostProcessing: unknown color_correction_method {color_correction_method!r}") + + h2 = output.shape[-3] - (output.shape[-3] % 2) + w2 = output.shape[-2] - (output.shape[-2] % 2) + output = output[:, :, :h2, :w2, :] + if decoded_was_4d: + output = output.reshape(-1, output.shape[-3], output.shape[-2], output.shape[-1]) + return io.NodeOutput(output) + + @staticmethod + def _as_bthwc(images): + if images.ndim == 4: + return images.unsqueeze(0), True + if images.ndim == 5: + return images, False + raise ValueError( + f"SeedVR2PostProcessing: expected 4-D or 5-D IMAGE tensor, got shape {tuple(images.shape)}" + ) + + @staticmethod + def _restore_reference_batch_time(decoded, reference): + if decoded.shape[0] != 1: + return decoded + ref_b, ref_t = reference.shape[:2] + if ref_b < 1 or decoded.shape[1] % ref_b != 0: + return decoded + decoded_t = decoded.shape[1] // ref_b + if decoded_t < ref_t: + return decoded + return decoded.reshape(ref_b, decoded_t, decoded.shape[2], decoded.shape[3], decoded.shape[4]) + + @staticmethod + def _to_seedvr2_raw(images): + return images.mul(2.0).sub(1.0) + + @staticmethod + def _validate_upscaled_shorter_edge(upscaled_shorter_edge): + if not isinstance(upscaled_shorter_edge, int) or upscaled_shorter_edge < 2: + raise ValueError( + "SeedVR2PostProcessing: upscaled_shorter_edge must be an integer " + f"of at least 2 pixels; got {upscaled_shorter_edge!r}." + ) + + @staticmethod + def _resized_shorter_edge_dims(height, width, upscaled_shorter_edge): + if height <= width: + return upscaled_shorter_edge, int(upscaled_shorter_edge * width / height) + return int(upscaled_shorter_edge * height / width), upscaled_shorter_edge + + @classmethod + def _resize_original_reference(cls, original, upscaled_shorter_edge): + original_5d, _ = cls._as_bthwc(original) + b, t = original_5d.shape[:2] + original_flat = rearrange(original_5d, "b t h w c -> (b t) c h w") + resized_flat = side_resize(original_flat, upscaled_shorter_edge).clamp(0.0, 1.0) + return rearrange(resized_flat, "(b t) c h w -> b t h w c", b=b, t=t) + + @staticmethod + def _color_transfer_on_vae_device(decoded_flat, reference_flat, output_device, transfer_fn): + color_device = comfy.model_management.vae_device() + decoded_flat = decoded_flat.to(device=color_device) + reference_flat = reference_flat.to(device=color_device) + output = transfer_fn(decoded_flat, reference_flat) + return output.to(device=output_device) + + @staticmethod + def _lab_color_transfer_on_vae_device(decoded_flat, reference_flat, output_device): + color_device = comfy.model_management.vae_device() + result = None + for start in range(decoded_flat.shape[0]): + decoded_frame = decoded_flat[start:start + 1].to(device=color_device).clone() + reference_frame = reference_flat[start:start + 1].to(device=color_device).clone() + output = lab_color_transfer(decoded_frame, reference_frame).to(device=output_device) + if result is None: + result = torch.empty( + (decoded_flat.shape[0],) + tuple(output.shape[1:]), + device=output_device, + dtype=output.dtype, + ) + result[start:start + 1].copy_(output) + if result is None: + raise ValueError("SeedVR2PostProcessing: LAB color correction requires at least one frame.") + return result + + @classmethod + def _color_transfer_chunked(cls, decoded_flat, reference_flat, output_device, color_correction_method): + chunk_size = cls._estimate_color_correction_chunk_size(decoded_flat, color_correction_method) + while True: + next_chunk_size = None + try: + return cls._run_color_transfer_chunks( + decoded_flat, reference_flat, output_device, color_correction_method, chunk_size, + ) + except Exception as e: + comfy.model_management.raise_non_oom(e) + if chunk_size <= 1: + raise RuntimeError( + "SeedVR2PostProcessing: color correction OOM at one frame; " + f"color_correction_method={color_correction_method}, shape={tuple(decoded_flat.shape)}." + ) from e + next_chunk_size = max(1, chunk_size // 2) + + comfy.model_management.soft_empty_cache() + chunk_size = next_chunk_size + + @classmethod + def _run_color_transfer_chunks(cls, decoded_flat, reference_flat, output_device, color_correction_method, chunk_size): + result = None + for start in range(0, decoded_flat.shape[0], chunk_size): + end = min(start + chunk_size, decoded_flat.shape[0]) + decoded_chunk = decoded_flat[start:end] + reference_chunk = reference_flat[start:end] + if color_correction_method == "lab": + output = cls._lab_color_transfer_on_vae_device(decoded_chunk, reference_chunk, output_device) + elif color_correction_method == "wavelet": + output = cls._color_transfer_on_vae_device( + decoded_chunk, reference_chunk, output_device, wavelet_color_transfer, + ) + else: + output = cls._color_transfer_on_vae_device( + decoded_chunk, reference_chunk, output_device, adain_color_transfer, + ) + if result is None: + result = torch.empty( + (decoded_flat.shape[0],) + tuple(output.shape[1:]), + device=output_device, + dtype=output.dtype, + ) + result[start:end].copy_(output) + if result is None: + raise ValueError("SeedVR2PostProcessing: color correction requires at least one frame.") + return result + + @classmethod + def _estimate_color_correction_chunk_size(cls, decoded_flat, color_correction_method): + multiplier = cls._color_correction_memory_multiplier(color_correction_method) + frames = decoded_flat.shape[0] + _, channels, height, width = decoded_flat.shape + dtype_bytes = max(decoded_flat.element_size(), 4) + bytes_per_frame = height * width * channels * dtype_bytes * multiplier + if bytes_per_frame <= 0: + return frames + color_device = comfy.model_management.vae_device() + free_memory = comfy.model_management.get_free_memory(color_device) + chunk_size = int((free_memory * COLOR_CORRECTION_MEMORY_HEADROOM) // bytes_per_frame) + return max(1, min(frames, chunk_size)) + + @staticmethod + def _color_correction_memory_multiplier(color_correction_method): + if color_correction_method == "lab": + return LAB_SCALE_MULTIPLIER + if color_correction_method == "wavelet": + return WAVELET_SCALE_MULTIPLIER + if color_correction_method == "adain": + return ADAIN_SCALE_MULTIPLIER + raise ValueError(f"SeedVR2PostProcessing: unknown color_correction_method {color_correction_method!r}") + + @staticmethod + def _resize_reference(reference, height, width): + if reference.shape[2] == height and reference.shape[3] == width: + return reference + b, t = reference.shape[:2] + reference_flat = rearrange(reference, "b t h w c -> (b t) c h w") + resized = TVF.resize( + reference_flat, + size=(height, width), + interpolation=InterpolationMode.BICUBIC, + antialias=not (isinstance(reference_flat, torch.Tensor) and reference_flat.device.type == "mps"), + ) + return rearrange(resized, "(b t) c h w -> b t h w c", b=b, t=t) + + +class SeedVR2Conditioning(io.ComfyNode): + @classmethod + def define_schema(cls): + return io.Schema( + node_id="SeedVR2Conditioning", + category="image/video", + inputs=[ + io.Model.Input("model"), + io.Latent.Input("vae_conditioning", display_name="LATENT"), + ], + outputs=[ + io.Model.Output(display_name = "model"), + io.Conditioning.Output(display_name = "positive"), + io.Conditioning.Output(display_name = "negative"), + io.Latent.Output(display_name = "latent"), + ], + ) + + @classmethod + def execute(cls, model, vae_conditioning) -> io.NodeOutput: + + vae_conditioning = vae_conditioning["samples"] + if vae_conditioning.ndim != 5: + raise ValueError( + "SeedVR2Conditioning expects a 5-D VAE latent in Comfy " + f"channel-first layout; got shape {tuple(vae_conditioning.shape)}." + ) + if vae_conditioning.shape[-1] == _SEEDVR2_LATENT_CHANNELS and vae_conditioning.shape[1] != _SEEDVR2_LATENT_CHANNELS: + raise ValueError( + "SeedVR2Conditioning expects SeedVR2 VAE latents in Comfy " + f"channel-first layout (B, {_SEEDVR2_LATENT_CHANNELS}, T, H, W); " + f"got channel-last shape {tuple(vae_conditioning.shape)}." + ) + vae_conditioning = vae_conditioning.movedim(1, -1).contiguous() + model_patcher = model + model = _resolve_seedvr2_diffusion_model(model_patcher) + pos_cond = model.positive_conditioning + neg_cond = model.negative_conditioning + + # Fail-loud guard against silently-wrong output when a numz-format + # DiT-only ``.safetensors`` (no ``positive_conditioning`` / + # ``negative_conditioning`` keys) is loaded via ``UNETLoader``. + # ``NaDiT.__init__`` zero-fills the buffers via ``torch.zeros`` (see + # ``comfy/ldm/seedvr/model.py``); ``load_state_dict(strict=False)`` + # leaves them at zero when the keys are absent. Detect that state + # here rather than at ``BaseModel.extra_conds`` (per sampling step, + # wasteful) or at the resolver helper (mixes structural shape with + # semantic content). Both buffers must be checked together — partial + # bake regressions could populate one but not the other. + if ( + pos_cond.float().abs().sum().item() == 0 + and neg_cond.float().abs().sum().item() == 0 + ): + raise RuntimeError( + f"{_SEEDVR2_INVALID_MODEL_MSG_PREFIX}: positive_conditioning " + f"and negative_conditioning buffers are zero-valued — model " + f"file appears to be a numz-format DiT-only export missing " + f"the SeedVR2 conditioning tensors. " + f"Re-bake the file with ``positive_conditioning`` (58, 5120) " + f"and ``negative_conditioning`` (64, 5120) keys at top level, " + f"or load via CheckpointLoaderSimple from a bundled " + f"checkpoint." + ) + + _apply_rope_freqs_float32_cast(model) + + condition = torch.stack([get_conditions(c, c) for c in vae_conditioning]) + condition = condition.movedim(-1, 1) + latent = vae_conditioning.movedim(-1, 1) + + latent = rearrange(latent, "b c t h w -> b (c t) h w") + condition = rearrange(condition, "b c t h w -> b (c t) h w") + + negative = [[neg_cond.unsqueeze(0), {"condition": condition}]] + positive = [[pos_cond.unsqueeze(0), {"condition": condition}]] + + return io.NodeOutput(model_patcher, positive, negative, {"samples": latent}) + +# SeedVR2 latent / conditioning channel constants. The SeedVR2 conditioning +# stage collapses ``(B, C, T, H, W) -> (B, C*T, H, W)`` for both the latent +# (C=16) and the per-frame condition tensor (C=17 = 16 latent + 1 mask), as +# required by ``NaDiT.forward`` which un-collapses via +# ``view(B, 16, -1, H, W)`` and ``view(B, 17, -1, H, W)`` respectively. +_SEEDVR2_LATENT_CHANNELS = 16 +_SEEDVR2_CONDITION_CHANNELS = 17 + + +def _slice_collapsed_4d_along_t(tensor_4d: torch.Tensor, t_start: int, + t_end: int, channels: int) -> torch.Tensor: + """Slice a SeedVR2-style collapsed 4D tensor ``(B, channels*T, H, W)`` + along the latent T axis, returning ``(B, channels*(t_end - t_start), H, W)``. + + Reshape -> slice -> ``.contiguous()`` -> re-collapse. ``reshape`` is + used for the un-collapse so non-contiguous incoming tensors from + cropping or slicing nodes are accepted. The + ``.contiguous()`` is mandatory: T-axis slicing of a 5D tensor produces a + non-contiguous view, and the subsequent re-collapse requires contiguous + storage. + """ + B, CT, H, W = tensor_4d.shape + if CT % channels != 0: + raise ValueError( + f"_slice_collapsed_4d_along_t: collapsed channel dim {CT} is not " + f"divisible by channels={channels}; tensor shape {tuple(tensor_4d.shape)}." + ) + T = CT // channels + if not (0 <= t_start < t_end <= T): + raise ValueError( + f"_slice_collapsed_4d_along_t: slice [{t_start}:{t_end}] out of " + f"range for T={T}." + ) + new_T = t_end - t_start + sliced = tensor_4d.reshape(B, channels, T, H, W)[:, :, t_start:t_end, :, :].contiguous() + return sliced.reshape(B, channels * new_T, H, W) + + +def _slice_seedvr2_cond_along_t(cond_list, t_start: int, t_end: int): + """Build a new SeedVR2 conditioning list with the per-frame ``condition`` + tensor sliced along the latent T axis. + + SeedVR2 conditioning entries have the shape + ``[text_cond_tensor, options_dict]`` where ``options_dict["condition"]`` + is a 4D collapsed ``(B, 17*T, H, W)`` tensor; the text tensor itself has + no temporal axis and is passed through unchanged. Other keys in the + options dict (controlnets, etc.) are also passed through unchanged. If + an entry has no ``"condition"`` key, the entry is forwarded verbatim. + + A new list of ``[text_cond, new_options_dict]`` pairs is returned; the + original ``cond_list`` and its options dicts are not mutated. + """ + new_list = [] + for entry in cond_list: + text_cond, options = entry[0], entry[1] + if "condition" not in options: + new_list.append(entry) + continue + new_options = options.copy() + new_options["condition"] = _slice_collapsed_4d_along_t( + new_options["condition"], t_start, t_end, + _SEEDVR2_CONDITION_CHANNELS, + ) + new_list.append([text_cond, new_options]) + return new_list + + +def _slice_seedvr2_noise_mask_along_t(noise_mask: torch.Tensor, + samples_4d: torch.Tensor, + t_start: int, + t_end: int): + """Slice collapsed SeedVR2 masks and preserve standard masks. + + ``SetLatentNoiseMask`` produces ``(B, 1, H, W)`` masks that KSampler + expands to the latent shape. Only masks already expanded to the full + collapsed ``(B, 16*T, H, W)`` shape need temporal slicing here. + """ + if noise_mask.ndim == samples_4d.ndim and noise_mask.shape[1] == samples_4d.shape[1]: + return _slice_collapsed_4d_along_t( + noise_mask, t_start, t_end, _SEEDVR2_LATENT_CHANNELS, + ) + return noise_mask + + +def _concat_chunks_along_t(chunks_4d, channels: int) -> torch.Tensor: + """Concatenate a list of SeedVR2-style collapsed 4D tensors + ``(B, channels*T_i, H, W)`` along the latent T axis. Each chunk is + un-collapsed to 5D, concatenated on ``dim=2``, then re-collapsed to 4D. + """ + if len(chunks_4d) == 0: + raise ValueError("_concat_chunks_along_t: empty chunk list.") + fives = [] + for ch in chunks_4d: + B, CT, H, W = ch.shape + if CT % channels != 0: + raise ValueError( + f"_concat_chunks_along_t: chunk shape {tuple(ch.shape)} " + f"channel dim {CT} not divisible by channels={channels}." + ) + T = CT // channels + fives.append(ch.reshape(B, channels, T, H, W)) + cat = torch.cat(fives, dim=2).contiguous() + B, C, T_total, H, W = cat.shape + return cat.reshape(B, C * T_total, H, W) + + +def _hann_blend_weights_1d(overlap: int, device, dtype) -> torch.Tensor: + """Build a 1D crossfade weight tensor of length ``overlap`` for the + *previous* chunk's contribution; the current chunk's weight is + ``1 - w_prev``. + + Mirrors the numz ``blend_overlapping_frames`` shape + (AInVFX/numz fork ``src/core/generation_utils.py``, + ``blend_overlapping_frames``): a Hann window with a ``[1/3, 2/3]`` + dead-band when ``overlap >= 3``, and a plain linear ramp when + ``overlap < 3`` (the dead-band would collapse the transition for + very small overlap counts). The numz reference operates on + pixel-space tensors ``[overlap, H, W, C]``; this 1D form is + reshaped by the caller to broadcast across the latent's + ``(B, C, T_overlap, H, W)`` axes. + """ + if overlap < 1: + raise ValueError( + f"_hann_blend_weights_1d: overlap must be >= 1; got {overlap}." + ) + if overlap >= 3: + t = torch.linspace(0.0, 1.0, steps=overlap, device=device, dtype=dtype) + blend_start = 1.0 / 3.0 + blend_end = 2.0 / 3.0 + u = ((t - blend_start) / (blend_end - blend_start)).clamp(0.0, 1.0) + return 0.5 + 0.5 * torch.cos(torch.pi * u) + return torch.linspace(1.0, 0.0, steps=overlap, device=device, dtype=dtype) + + +def _blend_overlap_region(prev_tail_5d: torch.Tensor, + cur_head_5d: torch.Tensor) -> torch.Tensor: + """Blend two 5D ``(B, C, T_overlap, H, W)`` tensors of equal shape + using a 1D Hann/linear ramp along the T axis. ``prev_tail_5d`` + receives the descending weight; ``cur_head_5d`` receives + ``1 - w_prev``. + + The caller is responsible for ensuring both inputs have identical + shape and dtype/device. + """ + if prev_tail_5d.shape != cur_head_5d.shape: + raise ValueError( + f"_blend_overlap_region: shape mismatch " + f"prev {tuple(prev_tail_5d.shape)} vs " + f"cur {tuple(cur_head_5d.shape)}." + ) + overlap = int(prev_tail_5d.shape[2]) + w_prev_1d = _hann_blend_weights_1d( + overlap, prev_tail_5d.device, prev_tail_5d.dtype, + ) + # Reshape to (1, 1, overlap, 1, 1) for broadcast across B, C, H, W. + w_prev = w_prev_1d.view(1, 1, overlap, 1, 1) + w_cur = 1.0 - w_prev + return prev_tail_5d * w_prev + cur_head_5d * w_cur + + +def _concat_chunks_with_overlap_blend(chunk_specs, channels: int, + overlap_latent: int) -> torch.Tensor: + """Concatenate temporally-overlapping chunks back into a single + collapsed 4D tensor, blending overlap regions with a Hann/linear + crossfade. + + ``chunk_specs`` is a list of ``(t_start, t_end, chunk_4d)`` tuples + in source-latent T coordinates. ``overlap_latent == 0`` is a fast + path that delegates to plain concatenation (and produces output + bit-identical to ``_concat_chunks_along_t`` of the same chunks). + + The blend at each pair of adjacent chunks acts on the actual + overlap region width ``min(prev_end - cur_start, current chunk + length)``, which may be smaller than ``overlap_latent`` when the + final chunk is a runt shorter than the configured overlap. + """ + if len(chunk_specs) == 0: + raise ValueError("_concat_chunks_with_overlap_blend: empty chunk list.") + if overlap_latent < 0: + raise ValueError( + f"_concat_chunks_with_overlap_blend: overlap_latent must be " + f">= 0; got {overlap_latent}." + ) + + # Validate channel divisibility once and capture per-chunk T. + chunk_5d = [] + for t_start, t_end, ch in chunk_specs: + B, CT, H, W = ch.shape + if CT % channels != 0: + raise ValueError( + f"_concat_chunks_with_overlap_blend: chunk shape " + f"{tuple(ch.shape)} channel dim {CT} not divisible " + f"by channels={channels}." + ) + T = CT // channels + if t_end - t_start != T: + raise ValueError( + f"_concat_chunks_with_overlap_blend: chunk T={T} mismatches " + f"declared range [{t_start}:{t_end}]." + ) + chunk_5d.append((t_start, t_end, ch.reshape(B, channels, T, H, W))) + + if overlap_latent == 0: + # Fast path: pure concat in the caller-provided chunk order. + return _concat_chunks_along_t( + [c.reshape(c.shape[0], channels * c.shape[2], c.shape[3], c.shape[4]) + for _, _, c in chunk_5d], + channels, + ) + + T_total = max(t_end for _, t_end, _ in chunk_5d) + first_5d = chunk_5d[0][2] + B = first_5d.shape[0] + H = first_5d.shape[3] + W = first_5d.shape[4] + result = torch.empty( + (B, channels, T_total, H, W), + device=first_5d.device, dtype=first_5d.dtype, + ) + filled_until = 0 + for i, (cs, ce, ct_5d) in enumerate(chunk_5d): + chunk_T = int(ct_5d.shape[2]) + if i == 0: + result[:, :, cs:ce, :, :] = ct_5d + filled_until = ce + continue + # Overlap region width is bounded by both the previous fill + # frontier and the current chunk's actual length (for runt + # final chunks shorter than the configured overlap). + overlap_len = min(filled_until - cs, chunk_T) + if overlap_len > 0: + prev_tail = result[:, :, cs:cs + overlap_len, :, :].contiguous() + cur_head = ct_5d[:, :, :overlap_len, :, :].contiguous() + blended = _blend_overlap_region(prev_tail, cur_head) + result[:, :, cs:cs + overlap_len, :, :] = blended + tail_start = cs + overlap_len + tail_end = ce + if tail_end > tail_start: + result[:, :, tail_start:tail_end, :, :] = ( + ct_5d[:, :, overlap_len:, :, :] + ) + else: + # Disjoint chunks (overlap_latent set but this pair did not + # actually overlap, e.g. step_latent equal to chunk_latent + # in a degenerate config). Treat as concat. + result[:, :, cs:ce, :, :] = ct_5d + filled_until = ce + + return result.contiguous().reshape(B, channels * T_total, H, W) + + +def _run_standard_sample(model, seed: int, steps: int, cfg: float, + sampler_name: str, scheduler: str, + positive, negative, latent_image: dict, + denoise: float) -> dict: + """Single-shot delegation that mirrors the standard ``common_ksampler`` + flow (``nodes.py:common_ksampler``): generate noise from seed, run + ``comfy.sample.sample``, return a latent dict. Used by the + ProgressiveSampler short-circuit when the full sequence fits in one + chunk so chunking introduces no overhead for small videos. + """ + samples_in = latent_image["samples"] + samples_in = comfy.sample.fix_empty_latent_channels( + model, samples_in, latent_image.get("downscale_ratio_spacial", None), + ) + batch_inds = latent_image.get("batch_index", None) + noise = comfy.sample.prepare_noise(samples_in, seed, batch_inds) + noise_mask = latent_image.get("noise_mask", None) + samples = comfy.sample.sample( + model, noise, steps, cfg, sampler_name, scheduler, + positive, negative, samples_in, + denoise=denoise, noise_mask=noise_mask, seed=seed, + ) + out = latent_image.copy() + out.pop("downscale_ratio_spacial", None) + out["samples"] = samples + return out + + +class SeedVR2ProgressiveSampler(io.ComfyNode): + """Sequential temporal chunking sampler for SeedVR2 native. + + Drop-in replacement for ``KSampler`` in SeedVR2 native workflows that + OOM on long sequences. The latent enters the sampler in SeedVR2's + collapsed form ``(B, 16*T, H, W)`` (collapsed by ``SeedVR2Conditioning`` + at ``rearrange(b c t h w -> b (c t) h w)``); this node slices that + tensor along the temporal axis, runs the configured inner sampler + sequentially per chunk against the standard ``comfy.sample.sample`` + entry point, and concatenates per-chunk outputs back into a single + ``(B, 16*T_total, H, W)`` latent. + + ``frames_per_chunk`` is expressed in pixel-frame units to match the + SeedVR2 4n+1 constraint enforced upstream by ``cut_videos`` and the + VAE's ``temporal_downsample_factor=4``. A pixel chunk size ``F`` + maps to ``(F - 1) // 4 + 1`` latent-frame chunks. + + Determinism contract: a single noise tensor is generated once from + the user seed and sliced per chunk (rather than re-seeding each + chunk), so a workflow that fits in a single chunk produces output + identical to a workflow that fits in N chunks at the same seed, + modulo the inherent T-axis chunk-boundary independence of the model. + """ + + @classmethod + def define_schema(cls): + return io.Schema( + node_id="SeedVR2ProgressiveSampler", + category="sampling", + inputs=[ + io.Model.Input("model"), + io.Int.Input("seed", default=0, min=0, + max=0xffffffffffffffff, + control_after_generate=True), + io.Int.Input("steps", default=20, min=1, max=10000), + io.Float.Input("cfg", default=1.0, min=0.0, max=100.0, + step=0.1, round=0.01), + io.Combo.Input("sampler_name", + options=comfy.samplers.SAMPLER_NAMES), + io.Combo.Input("scheduler", + options=comfy.samplers.SCHEDULER_NAMES), + io.Conditioning.Input("positive"), + io.Conditioning.Input("negative"), + io.Latent.Input("latent_image"), + io.Float.Input("denoise", default=1.0, min=0.0, max=1.0, + step=0.01), + io.Int.Input("frames_per_chunk", default=21, min=1, + max=16384, step=4), + io.Int.Input("temporal_overlap", default=0, min=0, + max=16384, + tooltip="Latent-frame overlap between " + "adjacent chunks; blended with a " + "Hann window (linear for overlap " + "< 3). 0 = no blend, pure concat. " + "Values >= the chunk's latent-frame " + "length use the maximum valid " + "overlap; 1 latent frame corresponds " + "to ~4 pixel frames."), + io.Combo.Input("chunking_mode", + options=["manual", "auto"], + default="manual", + tooltip="manual = use frames_per_chunk " + "exactly; auto = retry only real OOM " + "failures with progressively smaller " + "temporal chunks."), + ], + outputs=[io.Latent.Output()], + ) + + @classmethod + def execute(cls, model, seed, steps, cfg, sampler_name, scheduler, + positive, negative, latent_image, denoise, + frames_per_chunk, temporal_overlap, + chunking_mode="manual") -> io.NodeOutput: + # 4n+1 validation in pixel-frame domain. The SeedVR2 native pipeline + # requires pixel-frame counts of the form 4n+1 (1, 5, 9, 13, ...), + # imposed at ``cut_videos`` upstream and propagated through the VAE's + # temporal_downsample_factor=4. Reject violations explicitly before + # any model invocation; a silent rounding would mis-align chunk + # boundaries with the 4n+1 lattice. + if frames_per_chunk < 1 or (frames_per_chunk - 1) % 4 != 0: + raise ValueError( + f"SeedVR2ProgressiveSampler: frames_per_chunk must be a " + f"4n+1 pixel-frame count (1, 5, 9, 13, 17, 21, ...); " + f"got {frames_per_chunk}." + ) + + samples_4d = latent_image["samples"] + samples_4d = comfy.sample.fix_empty_latent_channels( + model, samples_4d, + latent_image.get("downscale_ratio_spacial", None), + ) + if samples_4d.ndim != 4: + raise ValueError( + f"SeedVR2ProgressiveSampler: expected 4D collapsed latent " + f"(B, 16*T, H, W); got shape {tuple(samples_4d.shape)}." + ) + B, CT, H, W = samples_4d.shape + if CT % _SEEDVR2_LATENT_CHANNELS != 0: + raise ValueError( + f"SeedVR2ProgressiveSampler: collapsed channel dim {CT} is " + f"not divisible by SeedVR2 latent channels " + f"{_SEEDVR2_LATENT_CHANNELS}; latent does not appear to be " + f"SeedVR2-shaped." + ) + T_latent = CT // _SEEDVR2_LATENT_CHANNELS + T_pixel = 4 * (T_latent - 1) + 1 + + if chunking_mode not in ("manual", "auto"): + raise ValueError( + f"SeedVR2ProgressiveSampler: chunking_mode must be " + f"'manual' or 'auto'; got {chunking_mode!r}." + ) + + if chunking_mode == "auto": + attempts = _seedvr2_auto_chunk_attempts( + T_latent, T_pixel, frames_per_chunk, + ) + for i, attempt_frames_per_chunk in enumerate(attempts): + retry = False + try: + return cls.execute( + model=model, seed=seed, steps=steps, cfg=cfg, + sampler_name=sampler_name, scheduler=scheduler, + positive=positive, negative=negative, + latent_image=latent_image, denoise=denoise, + frames_per_chunk=attempt_frames_per_chunk, + temporal_overlap=temporal_overlap, + chunking_mode="manual", + ) + except Exception as e: + comfy.model_management.raise_non_oom(e) + if i == len(attempts) - 1: + raise RuntimeError( + "SeedVR2ProgressiveSampler: exhausted auto " + "chunking attempts after OOM. Tried " + f"frames_per_chunk values {attempts}." + ) from e + retry = True + + if retry: + logging.warning( + "SeedVR2ProgressiveSampler auto chunking OOM at " + "frames_per_chunk=%s; retrying with " + "frames_per_chunk=%s.", + attempt_frames_per_chunk, attempts[i + 1], + ) + comfy.model_management.soft_empty_cache() + + # Short-circuit: total fits in one chunk -> standard path with no + # chunking overhead. Output of this branch is byte-identical to the + # built-in KSampler given the same (model, seed, steps, cfg, + # sampler_name, scheduler, positive, negative, latent_image, + # denoise) tuple. + if T_pixel <= frames_per_chunk: + return io.NodeOutput(_run_standard_sample( + model, seed, steps, cfg, sampler_name, scheduler, + positive, negative, latent_image, denoise, + )) + + # Map pixel chunk -> latent chunk. Each chunk's latent length is + # at most ``chunk_latent``; the final chunk may be a runt that + # is automatically 4n+1-aligned in the pixel domain by the + # T_pixel = 4*(T_latent-1) + 1 mapping (every positive integer + # T_latent corresponds to a valid 4n+1 pixel count). + chunk_latent = (frames_per_chunk - 1) // 4 + 1 + + # ``temporal_overlap`` is exposed in latent-frame units, but users + # do not know the derived latent chunk length. Treat oversized + # values as "maximum valid overlap" while preserving a strictly + # positive chunk-loop stride. + if temporal_overlap < 0: + raise ValueError( + f"SeedVR2ProgressiveSampler: temporal_overlap must be >= 0; " + f"got {temporal_overlap}." + ) + temporal_overlap = min(temporal_overlap, chunk_latent - 1) + step_latent = chunk_latent - temporal_overlap + + # Generate full noise once from the user seed, then slice along T + # per chunk. Using one global noise tensor (rather than re-seeding + # per chunk) preserves seed-determinism across chunk-count + # variations: the same (seed, total T_latent) always produces the + # same noise samples regardless of how the work is partitioned. + batch_inds = latent_image.get("batch_index", None) + noise_full = comfy.sample.prepare_noise(samples_4d, seed, batch_inds) + + noise_mask = latent_image.get("noise_mask", None) + + # Build the flat list of chunk ranges first so the chunking + # geometry is fully known before any sample call. + chunk_ranges = [] + for chunk_start in range(0, T_latent, step_latent): + chunk_end = min(chunk_start + chunk_latent, T_latent) + if chunk_start >= chunk_end: + # The final iteration of a stride that lands exactly on + # T_latent produces a zero-length chunk; skip it. + break + chunk_ranges.append((chunk_start, chunk_end)) + if chunk_end >= T_latent: + break + + def _sample_one_chunk(chunk_start, chunk_end): + samples_chunk = _slice_collapsed_4d_along_t( + samples_4d, chunk_start, chunk_end, + _SEEDVR2_LATENT_CHANNELS, + ) + noise_chunk = _slice_collapsed_4d_along_t( + noise_full, chunk_start, chunk_end, + _SEEDVR2_LATENT_CHANNELS, + ) + positive_chunk = _slice_seedvr2_cond_along_t( + positive, chunk_start, chunk_end, + ) + negative_chunk = _slice_seedvr2_cond_along_t( + negative, chunk_start, chunk_end, + ) + + # Per-chunk noise_mask handling: standard masks are passed + # through for KSampler expansion; pre-expanded collapsed + # masks are sliced. + chunk_noise_mask = None + if noise_mask is not None: + chunk_noise_mask = _slice_seedvr2_noise_mask_along_t( + noise_mask, samples_4d, chunk_start, chunk_end, + ) + + return comfy.sample.sample( + model, noise_chunk, steps, cfg, sampler_name, scheduler, + positive_chunk, negative_chunk, samples_chunk, + denoise=denoise, noise_mask=chunk_noise_mask, seed=seed, + ) + + chunk_specs = [] + for chunk_start, chunk_end in chunk_ranges: + chunk_samples = _sample_one_chunk(chunk_start, chunk_end) + chunk_specs.append((chunk_start, chunk_end, chunk_samples)) + + final = _concat_chunks_with_overlap_blend( + chunk_specs, _SEEDVR2_LATENT_CHANNELS, temporal_overlap, + ) + + out = latent_image.copy() + out.pop("downscale_ratio_spacial", None) + out["samples"] = final + return io.NodeOutput(out) + + +class SeedVRExtension(ComfyExtension): + @override + async def get_node_list(self) -> list[type[io.ComfyNode]]: + return [ + SeedVR2Conditioning, + SeedVR2Resize, + SeedVR2ResizeAdvanced, + SeedVR2PostProcessing, + SeedVR2ProgressiveSampler, + ] + +async def comfy_entrypoint() -> SeedVRExtension: + return SeedVRExtension() diff --git a/nodes.py b/nodes.py index 528bf316f..a5d1b16cd 100644 --- a/nodes.py +++ b/nodes.py @@ -47,14 +47,18 @@ import node_helpers if args.enable_manager: import comfyui_manager + def before_node_execution(): comfy.model_management.throw_exception_if_processing_interrupted() + def interrupt_processing(value=True): comfy.model_management.interrupt_current_processing(value) + MAX_RESOLUTION=16384 + class CLIPTextEncode(ComfyNodeABC): @classmethod def INPUT_TYPES(s) -> InputTypeDict: @@ -323,8 +327,8 @@ class VAEDecodeTiled: return {"required": {"samples": ("LATENT", ), "vae": ("VAE", ), "tile_size": ("INT", {"default": 512, "min": 64, "max": 4096, "step": 32, "advanced": True}), "overlap": ("INT", {"default": 64, "min": 0, "max": 4096, "step": 32, "advanced": True}), - "temporal_size": ("INT", {"default": 64, "min": 8, "max": 4096, "step": 4, "tooltip": "Only used for video VAEs: Amount of frames to decode at a time.", "advanced": True}), - "temporal_overlap": ("INT", {"default": 8, "min": 4, "max": 4096, "step": 4, "tooltip": "Only used for video VAEs: Amount of frames to overlap.", "advanced": True}), + "temporal_size": ("INT", {"default": 64, "min": 0, "max": 4096, "step": 4, "tooltip": "Only used for video VAEs: Amount of frames to decode at a time. SeedVR2 allows 0 to disable temporal slicing.", "advanced": True}), + "temporal_overlap": ("INT", {"default": 8, "min": 0, "max": 4096, "step": 4, "tooltip": "Only used for video VAEs: Amount of frames to overlap.", "advanced": True}), }} RETURN_TYPES = ("IMAGE",) FUNCTION = "decode" @@ -334,18 +338,32 @@ class VAEDecodeTiled: def decode(self, vae, samples, tile_size, overlap=64, temporal_size=64, temporal_overlap=8): if tile_size < overlap * 4: overlap = tile_size // 4 - if temporal_size < temporal_overlap * 2: - temporal_overlap = temporal_overlap // 2 temporal_compression = vae.temporal_compression_decode() if temporal_compression is not None: - temporal_size = max(2, temporal_size // temporal_compression) - temporal_overlap = max(1, min(temporal_size // 2, temporal_overlap // temporal_compression)) + if temporal_size <= 0: + temporal_size = 0 + temporal_overlap = 0 + else: + requested_temporal_overlap = temporal_overlap + if temporal_size < temporal_overlap * 2: + temporal_overlap = temporal_overlap // 2 + temporal_size = max(2, temporal_size // temporal_compression) + temporal_overlap = min(temporal_size // 2, temporal_overlap // temporal_compression) + if requested_temporal_overlap > 0: + temporal_overlap = max(1, temporal_overlap) else: temporal_size = None temporal_overlap = None compression = vae.spacial_compression_decode() - images = vae.decode_tiled(samples["samples"], tile_x=tile_size // compression, tile_y=tile_size // compression, overlap=overlap // compression, tile_t=temporal_size, overlap_t=temporal_overlap) + images = vae.decode_tiled( + samples["samples"], + tile_x=tile_size // compression, + tile_y=tile_size // compression, + overlap=overlap // compression, + tile_t=temporal_size, + overlap_t=temporal_overlap, + ) if len(images.shape) == 5: #Combine batches images = images.reshape(-1, images.shape[-3], images.shape[-2], images.shape[-1]) return (images, ) @@ -362,7 +380,7 @@ class VAEEncode: def encode(self, vae, pixels): t = vae.encode(pixels) - return ({"samples":t}, ) + return ({"samples": t}, ) class VAEEncodeTiled: @classmethod @@ -370,8 +388,8 @@ class VAEEncodeTiled: return {"required": {"pixels": ("IMAGE", ), "vae": ("VAE", ), "tile_size": ("INT", {"default": 512, "min": 64, "max": 4096, "step": 64, "advanced": True}), "overlap": ("INT", {"default": 64, "min": 0, "max": 4096, "step": 32, "advanced": True}), - "temporal_size": ("INT", {"default": 64, "min": 8, "max": 4096, "step": 4, "tooltip": "Only used for video VAEs: Amount of frames to encode at a time.", "advanced": True}), - "temporal_overlap": ("INT", {"default": 8, "min": 4, "max": 4096, "step": 4, "tooltip": "Only used for video VAEs: Amount of frames to overlap.", "advanced": True}), + "temporal_size": ("INT", {"default": 64, "min": 0, "max": 4096, "step": 4, "tooltip": "Only used for video VAEs: Amount of frames to encode at a time. SeedVR2 allows 0 to disable temporal slicing.", "advanced": True}), + "temporal_overlap": ("INT", {"default": 8, "min": 0, "max": 4096, "step": 4, "tooltip": "Only used for video VAEs: Amount of frames to overlap.", "advanced": True}), }} RETURN_TYPES = ("LATENT",) FUNCTION = "encode" @@ -379,6 +397,9 @@ class VAEEncodeTiled: CATEGORY = "experimental" def encode(self, vae, pixels, tile_size, overlap, temporal_size=64, temporal_overlap=8): + if temporal_size <= 0: + temporal_size = 0 + temporal_overlap = 0 t = vae.encode_tiled(pixels, tile_x=tile_size, tile_y=tile_size, overlap=overlap, tile_t=temporal_size, overlap_t=temporal_overlap) return ({"samples": t}, ) @@ -2417,6 +2438,7 @@ async def init_builtin_extra_nodes(): "nodes_camera_trajectory.py", "nodes_edit_model.py", "nodes_tcfg.py", + "nodes_seedvr.py", "nodes_context_windows.py", "nodes_qwen.py", "nodes_chroma_radiance.py", diff --git a/tests-unit/comfy_extras_test/test_seedvr2_conditioning.py b/tests-unit/comfy_extras_test/test_seedvr2_conditioning.py new file mode 100644 index 000000000..ea26e1e37 --- /dev/null +++ b/tests-unit/comfy_extras_test/test_seedvr2_conditioning.py @@ -0,0 +1,213 @@ +"""Consolidated SeedVR2 conditioning and refactor regression tests. + +Merges the prior test_seedvr2_refactor_nodes.py and +test_seedvr_conditioning_hardening.py modules. Refactor tests use the +top-level comfy_extras.nodes_seedvr import; conditioning-hardening tests +use _import_nodes_seedvr_isolated() for sys.modules isolation when +mocking comfy.model_management. +""" + +import importlib +import sys +from unittest.mock import MagicMock + +import pytest +import torch +import torch.nn as nn + +from comfy.cli_args import args as cli_args + +if not torch.cuda.is_available(): + cli_args.cpu = True + + +_SENTINEL = object() +_TARGETS = ( + ("comfy.model_management", "comfy"), + ("comfy_extras.nodes_seedvr", "comfy_extras"), +) + + +def _import_nodes_seedvr_isolated(): + """Import comfy_extras.nodes_seedvr with comfy.model_management mocked.""" + priors = [] + for mod_name, parent_name in _TARGETS: + prior_mod = sys.modules.get(mod_name, _SENTINEL) + parent = sys.modules.get(parent_name) + attr = mod_name.split(".")[-1] + prior_attr = ( + getattr(parent, attr, _SENTINEL) if parent is not None else _SENTINEL + ) + priors.append((mod_name, parent_name, attr, prior_mod, prior_attr)) + + mock_mm = MagicMock() + for fn in ( + "xformers_enabled", "xformers_enabled_vae", + "pytorch_attention_enabled", "pytorch_attention_enabled_vae", + "sage_attention_enabled", "flash_attention_enabled", + "is_intel_xpu", + ): + getattr(mock_mm, fn).return_value = False + tv = torch.version.__version__.split(".") + mock_mm.torch_version_numeric = (int(tv[0]), int(tv[1])) + mock_mm.WINDOWS = False + sys.modules["comfy.model_management"] = mock_mm + if sys.modules.get("comfy") is None: + import comfy as _comfy_pkg # noqa: F401 + comfy_pkg = sys.modules.get("comfy") + if comfy_pkg is not None: + setattr(comfy_pkg, "model_management", mock_mm) + nodes_seedvr = sys.modules.get("comfy_extras.nodes_seedvr") or ( + importlib.import_module("comfy_extras.nodes_seedvr") + ) + + def _restore(): + for mod_name, parent_name, attr, prior_mod, prior_attr in priors: + if prior_mod is _SENTINEL: + sys.modules.pop(mod_name, None) + else: + sys.modules[mod_name] = prior_mod + parent = sys.modules.get(parent_name) + if parent is None: + continue + if prior_attr is _SENTINEL: + if hasattr(parent, attr): + delattr(parent, attr) + else: + setattr(parent, attr, prior_attr) + + return nodes_seedvr, _restore + + +class _Rope(nn.Module): + """Minimal RoPE stub exposing a `freqs` parameter.""" + def __init__(self): + super().__init__() + self.freqs = nn.Parameter(torch.zeros(4)) + + +class _Block(nn.Module): + """Minimal transformer block stub holding a `_Rope`.""" + def __init__(self): + super().__init__() + self.rope = _Rope() + + +class _DiffusionModel(nn.Module): + """Stub diffusion model with N blocks and pos/neg conditioning buffers.""" + def __init__(self, n_blocks=3, zero_conditioning=False, conditioning_dtype=torch.float32): + super().__init__() + self.blocks = nn.ModuleList([_Block() for _ in range(n_blocks)]) + pos = torch.zeros if zero_conditioning else torch.ones + self.register_buffer("positive_conditioning", pos((2, 4), dtype=conditioning_dtype)) + self.register_buffer("negative_conditioning", torch.zeros((3, 4), dtype=conditioning_dtype)) + + +class _ModelInner: + """Inner model wrapper exposing `.diffusion_model`.""" + def __init__(self, diffusion_model): + self.diffusion_model = diffusion_model + + +class _ModelPatcher: + """ModelPatcher stub exposing `.model._ModelInner`.""" + def __init__(self, diffusion_model): + self.model = _ModelInner(diffusion_model) + + +def test_seedvr2_conditioning_schema_exposes_model_passthrough_output(): + nodes_seedvr, restore = _import_nodes_seedvr_isolated() + try: + schema = nodes_seedvr.SeedVR2Conditioning.define_schema() + assert [input_item.id for input_item in schema.inputs] == [ + "model", + "vae_conditioning", + ] + assert schema.inputs[1].display_name == "LATENT" + assert [output.display_name for output in schema.outputs] == [ + "model", + "positive", + "negative", + "latent", + ] + finally: + restore() + + +def test_seedvr2_conditioning_returns_packed_input_latent_deterministically(): + nodes_seedvr, restore = _import_nodes_seedvr_isolated() + try: + diffusion_model = _DiffusionModel() + patcher = _ModelPatcher(diffusion_model) + samples = torch.arange(1, 25, dtype=torch.float32).reshape(1, 2, 3, 2, 2) + vae_conditioning = {"samples": samples} + + _, first_positive, first_negative, first_latent = ( + nodes_seedvr.SeedVR2Conditioning.execute( + patcher, + vae_conditioning, + ) + ) + _, second_positive, second_negative, second_latent = ( + nodes_seedvr.SeedVR2Conditioning.execute( + patcher, + vae_conditioning, + ) + ) + + expected_latent = samples.reshape(1, 6, 2, 2) + channel_last = samples.movedim(1, -1).contiguous() + expected_condition = torch.cat( + [ + channel_last, + torch.ones((*channel_last.shape[:-1], 1)), + ], + dim=-1, + ).movedim(-1, 1).reshape(1, 9, 2, 2) + + assert torch.equal(first_latent["samples"], expected_latent) + assert torch.equal(second_latent["samples"], expected_latent) + assert torch.equal( + first_positive[0][1]["condition"], + expected_condition, + ) + assert torch.equal( + second_positive[0][1]["condition"], + expected_condition, + ) + assert torch.equal( + first_negative[0][1]["condition"], + expected_condition, + ) + assert torch.equal( + second_negative[0][1]["condition"], + expected_condition, + ) + finally: + restore() + + +def test_seedvr2_conditioning_fails_loud_on_zero_buffers(): + nodes_seedvr, restore = _import_nodes_seedvr_isolated() + try: + diffusion_model = _DiffusionModel(zero_conditioning=True) + patcher = _ModelPatcher(diffusion_model) + vae_conditioning = {"samples": torch.zeros((1, 2, 1, 1, 1))} + + with pytest.raises(RuntimeError) as excinfo: + nodes_seedvr.SeedVR2Conditioning.execute( + patcher, vae_conditioning, + ) + + message = str(excinfo.value) + assert message.startswith( + nodes_seedvr._SEEDVR2_INVALID_MODEL_MSG_PREFIX + ), ( + "Fail-loud message must use the standard " + "_SEEDVR2_INVALID_MODEL_MSG_PREFIX so callers/log scrapers " + f"can match it. Got: {message!r}" + ) + assert "positive_conditioning" in message + assert "negative_conditioning" in message + finally: + restore() diff --git a/tests-unit/comfy_extras_test/test_seedvr2_nodes.py b/tests-unit/comfy_extras_test/test_seedvr2_nodes.py new file mode 100644 index 000000000..d5e7213a4 --- /dev/null +++ b/tests-unit/comfy_extras_test/test_seedvr2_nodes.py @@ -0,0 +1,70 @@ +import importlib +import inspect +import sys +from unittest.mock import MagicMock, patch + +import torch + +from comfy.cli_args import args as cli_args + +if not torch.cuda.is_available(): + cli_args.cpu = True + +import comfy_extras.nodes_seedvr as nodes_seedvr # noqa: E402 + + +def test_resize_simple_multiplier_resolves_upscaled_shorter_edge(): + images = torch.zeros(1, 3, 16, 20, 3) + + output = nodes_seedvr.SeedVR2Resize.execute(images, 4.0) + + input_pixels, original_image, upscaled_shorter_edge = output.result + assert tuple(input_pixels.shape) == (1, 5, 64, 80, 3) + assert input_pixels.min().item() == 0.0 + assert input_pixels.max().item() == 0.0 + assert original_image is images + assert upscaled_shorter_edge == 64 + + +def test_seedvr_node_signature_matches_schema(): + mock_mm = MagicMock() + mock_mm.xformers_enabled.return_value = False + mock_mm.xformers_enabled_vae.return_value = False + mock_mm.sage_attention_enabled.return_value = False + mock_mm.flash_attention_enabled.return_value = False + + sentinel = object() + prior_cpu = cli_args.cpu + cli_args.cpu = True + prior_module = sys.modules.get("comfy_extras.nodes_seedvr", sentinel) + comfy_pkg = sys.modules.get("comfy") + prior_mm_attr = getattr(comfy_pkg, "model_management", sentinel) if comfy_pkg else sentinel + + with patch.dict(sys.modules, {"comfy.model_management": mock_mm}): + if comfy_pkg is not None: + setattr(comfy_pkg, "model_management", mock_mm) + sys.modules.pop("comfy_extras.nodes_seedvr", None) + try: + nodes_seedvr = importlib.import_module("comfy_extras.nodes_seedvr") + for node_cls in (nodes_seedvr.SeedVR2Resize, nodes_seedvr.SeedVR2ResizeAdvanced): + schema_ids = [i.id for i in node_cls.define_schema().inputs] + exec_params = [ + p for p in inspect.signature(node_cls.execute).parameters.keys() + if p != "cls" + ] + assert schema_ids == exec_params, ( + f"{node_cls.__name__} schema/execute drift: " + f"schema_ids={schema_ids}, exec_params={exec_params}" + ) + finally: + cli_args.cpu = prior_cpu + if prior_module is sentinel: + sys.modules.pop("comfy_extras.nodes_seedvr", None) + else: + sys.modules["comfy_extras.nodes_seedvr"] = prior_module + if comfy_pkg is not None: + if prior_mm_attr is sentinel: + if hasattr(comfy_pkg, "model_management"): + delattr(comfy_pkg, "model_management") + else: + setattr(comfy_pkg, "model_management", prior_mm_attr) diff --git a/tests-unit/comfy_extras_test/test_seedvr2_post_processing.py b/tests-unit/comfy_extras_test/test_seedvr2_post_processing.py new file mode 100644 index 000000000..9d41f8657 --- /dev/null +++ b/tests-unit/comfy_extras_test/test_seedvr2_post_processing.py @@ -0,0 +1,60 @@ +from unittest.mock import patch + +import torch + +from comfy.cli_args import args as cli_args + +if not torch.cuda.is_available(): + cli_args.cpu = True + +from comfy_extras import nodes_seedvr # noqa: E402 + + +def _schema_ids(items): + return [item.id for item in items] + + +def test_seedvr2_post_processing_schema(): + schema = nodes_seedvr.SeedVR2PostProcessing.define_schema() + + assert _schema_ids(schema.inputs) == ["decoded", "original_image", "upscaled_shorter_edge", "color_correction_method"] + assert schema.inputs[2].default is None + assert schema.inputs[2].min == 2 + assert schema.inputs[2].force_input is True + assert schema.inputs[3].options == ["lab", "wavelet", "adain", "none"] + assert schema.inputs[3].default == "lab" + assert schema.outputs[0].get_io_type() == "IMAGE" + + +def test_seedvr2_post_processing_oom_error_uses_color_correction_method(monkeypatch): + decoded = torch.full((1, 3, 4, 4), 0.25) + reference = torch.full((1, 3, 4, 4), 0.75) + + def _lab(content, style): + raise torch.cuda.OutOfMemoryError("CUDA out of memory") + + monkeypatch.setattr(nodes_seedvr.comfy.model_management, "vae_device", lambda: torch.device("cpu")) + monkeypatch.setattr(nodes_seedvr.comfy.model_management, "get_free_memory", lambda device: 1_000_000) + monkeypatch.setattr(nodes_seedvr.comfy.model_management, "soft_empty_cache", lambda: None) + + with patch.object(nodes_seedvr, "lab_color_transfer", _lab): + try: + nodes_seedvr.SeedVR2PostProcessing._color_transfer_chunked( + decoded, reference, torch.device("cpu"), "lab", + ) + except RuntimeError as exc: + assert "color_correction_method=lab" in str(exc) + assert " method=lab" not in str(exc) + else: + raise AssertionError("expected RuntimeError for one-frame LAB OOM") + + +def test_seedvr2_post_processing_unknown_color_correction_method_raises(): + decoded = torch.zeros(1, 2, 4, 4, 3) + original = torch.zeros(1, 2, 4, 4, 3) + try: + nodes_seedvr.SeedVR2PostProcessing.execute(decoded, original, 4, "bogus") + except ValueError as exc: + assert "color_correction_method" in str(exc) + else: + raise AssertionError("expected ValueError for unknown color_correction_method") diff --git a/tests-unit/comfy_test/model_detection_test.py b/tests-unit/comfy_test/model_detection_test.py index 4e9350602..c63f69a0d 100644 --- a/tests-unit/comfy_test/model_detection_test.py +++ b/tests-unit/comfy_test/model_detection_test.py @@ -73,6 +73,24 @@ def _make_flux_schnell_comfyui_sd(): return sd +def _make_seedvr2_7b_separate_mm_sd(): + return { + "blocks.35.mlp.vid.proj_in.weight": torch.empty(1, 3072), + } + + +def _make_seedvr2_7b_shared_mm_sd(): + return { + "blocks.35.mlp.all.proj_in_gate.weight": torch.empty(1, 1), + } + + +def _make_seedvr2_3b_shared_mm_sd(): + return { + "blocks.31.mlp.all.proj_in_gate.weight": torch.empty(1, 1), + } + + class TestModelDetection: """Verify that first-match model detection selects the correct model based on list ordering and unet_config specificity.""" @@ -125,6 +143,48 @@ class TestModelDetection: assert model_config is not None assert type(model_config).__name__ == "FluxSchnell" + def test_seedvr2_7b_separate_mm_detection_config(self): + sd = _make_seedvr2_7b_separate_mm_sd() + unet_config = detect_unet_config(sd, "") + + assert unet_config is not None + assert unet_config["image_model"] == "seedvr2" + assert unet_config["vid_dim"] == 3072 + assert unet_config["heads"] == 24 + assert unet_config["num_layers"] == 36 + assert unet_config["mm_layers"] == 36 + assert unet_config["mlp_type"] == "normal" + assert unet_config["qk_rope"] is True + assert unet_config["rope_type"] == "rope3d" + assert unet_config["rope_dim"] == 64 + + def test_seedvr2_7b_shared_mm_detection_config(self): + sd = _make_seedvr2_7b_shared_mm_sd() + unet_config = detect_unet_config(sd, "") + + assert unet_config is not None + assert unet_config["image_model"] == "seedvr2" + assert unet_config["vid_dim"] == 3072 + assert unet_config["heads"] == 24 + assert unet_config["num_layers"] == 36 + assert unet_config["mm_layers"] == 10 + assert unet_config["mlp_type"] == "swiglu" + assert unet_config["qk_rope"] is True + assert unet_config["rope_type"] == "rope3d" + assert unet_config["rope_dim"] == 64 + + def test_seedvr2_3b_shared_mm_detection_config(self): + sd = _make_seedvr2_3b_shared_mm_sd() + unet_config = detect_unet_config(sd, "") + + assert unet_config is not None + assert unet_config["image_model"] == "seedvr2" + assert unet_config["vid_dim"] == 2560 + assert unet_config["heads"] == 20 + assert unet_config["num_layers"] == 32 + assert unet_config["mlp_type"] == "swiglu" + assert unet_config["qk_rope"] is None + def test_unet_config_and_required_keys_combination_is_unique(self): """Each model in the registry must have a unique combination of ``unet_config`` and ``required_keys``. If two models share the same diff --git a/tests-unit/comfy_test/seedvr_vae_forward_test.py b/tests-unit/comfy_test/seedvr_vae_forward_test.py new file mode 100644 index 000000000..f9dbd6890 --- /dev/null +++ b/tests-unit/comfy_test/seedvr_vae_forward_test.py @@ -0,0 +1,90 @@ +"""Regression: ``comfy.ldm.seedvr.vae.VideoAutoencoderKL.forward`` must +honor the actual tensor/tuple return contract of ``encode()`` and +``decode_()`` and must NOT dereference diffusers-style ``.latent_dist`` +or ``.sample`` attributes on those returns. + +The pre-fix body raised ``AttributeError: 'Tensor' object has no +attribute 'latent_dist'`` for ``mode in {"encode", "all"}`` and +``AttributeError: 'VideoAutoencoderKL' object has no attribute 'decode'`` +for ``mode == "decode"`` (the class only defines ``decode_`` with a +trailing underscore). The post-fix body unwraps the optional one-element +tuple shape that ``return_dict=False`` produces and returns the tensor +directly. + +Tests construct a stub subclass of ``VideoAutoencoderKL`` that bypasses +the heavy ``__init__`` via ``torch.nn.Module.__init__(self)`` and +overrides ``encode``/``decode_`` with known tensors so the contract can +be probed without loading any real VAE weights. +""" + +import torch +import torch.nn as nn + +from comfy.cli_args import args as cli_args + +if not torch.cuda.is_available(): + cli_args.cpu = True + +from comfy.ldm.seedvr.vae import VideoAutoencoderKL # noqa: E402 + + +_LATENT_SHAPE = (1, 16, 2, 2, 2) +_DECODED_SHAPE = (1, 3, 5, 16, 16) +_INPUT_ENCODE_SHAPE = (1, 3, 5, 16, 16) +_INPUT_DECODE_SHAPE = (1, 16, 2, 2, 2) + + +class _StubVAE(VideoAutoencoderKL): + def __init__(self): + nn.Module.__init__(self) + self._encode_out = torch.zeros(*_LATENT_SHAPE) + self._decode_out = torch.zeros(*_DECODED_SHAPE) + + def encode(self, x, return_dict=True): + return self._encode_out + + def decode_(self, z, return_dict=True): + return self._decode_out + + +def test_forward_encode_returns_tensor(): + vae = _StubVAE() + x = torch.zeros(*_INPUT_ENCODE_SHAPE) + result = vae.forward(x, mode="encode") + assert type(result) is torch.Tensor + assert result.shape == torch.Size(_LATENT_SHAPE) + + +def test_forward_decode_returns_tensor(): + vae = _StubVAE() + z = torch.zeros(*_INPUT_DECODE_SHAPE) + result = vae.forward(z, mode="decode") + assert type(result) is torch.Tensor + assert result.shape == torch.Size(_DECODED_SHAPE) + + +class _TupleReturningStubVAE(VideoAutoencoderKL): + """Stub variant whose ``encode``/``decode_`` return the + ``(tensor,)`` one-element tuple shape ``return_dict=False`` produces + in the parent class. Exercises the unwrap branch of + ``VideoAutoencoderKL.forward``. + """ + + def __init__(self): + nn.Module.__init__(self) + self._encode_tensor = torch.zeros(*_LATENT_SHAPE) + self._decode_tensor = torch.zeros(*_DECODED_SHAPE) + + def encode(self, x, return_dict=True): + return (self._encode_tensor,) + + def decode_(self, z, return_dict=True): + return (self._decode_tensor,) + + +def test_forward_all_unwraps_one_tuple_at_each_step(): + vae = _TupleReturningStubVAE() + x = torch.zeros(*_INPUT_ENCODE_SHAPE) + result = vae.forward(x, mode="all") + assert type(result) is torch.Tensor + assert result.shape == torch.Size(_DECODED_SHAPE) diff --git a/tests-unit/comfy_test/test_seedvr2_dtype.py b/tests-unit/comfy_test/test_seedvr2_dtype.py new file mode 100644 index 000000000..e5d79a306 --- /dev/null +++ b/tests-unit/comfy_test/test_seedvr2_dtype.py @@ -0,0 +1,47 @@ +import torch + +from comfy.cli_args import args as cli_args + +if not torch.cuda.is_available(): + cli_args.cpu = True + +import comfy.sd +import comfy.supported_models +import comfy.ldm.seedvr.model as seedvr_model + + +def test_seedvr2_fp16_manual_cast_only_for_bf16_device(monkeypatch): + bf16_device = object() + fp16_device = object() + + monkeypatch.setattr( + comfy.supported_models.comfy.model_management, + "should_use_bf16", + lambda device=None: device is bf16_device, + ) + + bf16_config = comfy.supported_models.SeedVR2({"image_model": "seedvr2"}) + bf16_config.set_inference_dtype(torch.float16, None, device=bf16_device) + assert bf16_config.manual_cast_dtype is torch.bfloat16 + + fp16_config = comfy.supported_models.SeedVR2({"image_model": "seedvr2"}) + fp16_config.set_inference_dtype(torch.float16, None, device=fp16_device) + assert fp16_config.manual_cast_dtype is None + + +def test_seedvr2_text_conditioning_accepts_cfg1_single_branch(): + context = torch.arange(6, dtype=torch.float32).reshape(1, 3, 2) + + txt, txt_shape = seedvr_model.NaDiT._resolve_text_conditioning(object(), context, [0]) + + torch.testing.assert_close(txt, context.squeeze(0)) + torch.testing.assert_close(txt_shape, torch.tensor([[3]], device=context.device)) + + +def test_seedvr2_vae_decode_memory_covers_full_frame_lab_transfer(): + estimate = comfy.sd._seedvr2_vae_decode_memory_used((1, 16, 26, 120, 160)) + old_estimate = 16 * 120 * 160 * (4 * 8 * 8) * 2 + + assert estimate == 101 * 960 * 1280 * 160 + assert estimate > 15 * 1024 ** 3 + assert estimate > old_estimate * 100 diff --git a/tests-unit/comfy_test/test_seedvr2_internals.py b/tests-unit/comfy_test/test_seedvr2_internals.py new file mode 100644 index 000000000..60ce0c5b4 --- /dev/null +++ b/tests-unit/comfy_test/test_seedvr2_internals.py @@ -0,0 +1,389 @@ +"""Consolidated SeedVR2 internals regression tests. + +Sources (all merged verbatim, helper names disambiguated where colliding): + + * RoPE rewrite — NaMMRotaryEmbedding3d.forward must match the legacy + apply_rotary_emb wrapper oracle at fp32. + * GroupNorm limit gate — causal_norm_wrapper at vae.py:509 must compare + memory_occupy against get_norm_limit(), not float('inf'). + * var_attention backend registry. + * var_attention_pytorch SeedVR2-named guard — present-API shape contract + with AST-level pinning of the guard ordering. + +Pre-import CPU-only guard is required because comfy.ldm.seedvr.model and +comfy.ldm.modules.attention transitively pull in comfy.model_management, +which probes torch.cuda.current_device() at import time unless args.cpu is +set first. +""" + +from __future__ import annotations + +import ast +import inspect +import logging +import textwrap +import warnings +from unittest.mock import patch + +import pytest +import torch + +from comfy.cli_args import args + +if not torch.cuda.is_available(): + args.cpu = True + +import comfy.ldm.seedvr.model as seedvr_model # noqa: E402 +import comfy.ldm.seedvr.vae as vae_mod # noqa: E402 +import comfy.ldm.modules.attention as attention # noqa: E402 +import comfy.ops as comfy_ops # noqa: E402 +from comfy.ldm.seedvr.model import ( # noqa: E402 + Cache, + NaMMRotaryEmbedding3d, +) +from comfy.ldm.seedvr.vae import ( # noqa: E402 + causal_norm_wrapper, + set_norm_limit, +) +from comfy.ldm.modules.attention import var_attention_pytorch # noqa: E402 + + +# --------------------------------------------------------------------------- +# RoPE rewrite tests (test_seedvr_rope_rewrite.py) +# --------------------------------------------------------------------------- + +# Test rig dimensions. dim=192 → per-axis rope dim = 64 (even, lucidrains +# requirement). vid_shape=(2,4,4) → L_vid = 32. txt_shape=(8,) → L_txt = 8. +_DIM = 192 +_HEADS = 4 +_VID_T, _VID_H, _VID_W = 2, 4, 4 +_TXT_L = 8 +_L_VID = _VID_T * _VID_H * _VID_W +_SEED = 0 + + +def _make_inputs(dtype=torch.float32, device="cpu"): + """Construct the 6 forward inputs + cache. Deterministic via local + Generator so global RNG state is not mutated. + """ + g = torch.Generator(device=device).manual_seed(_SEED) + vid_q = torch.randn(_L_VID, _HEADS, _DIM, dtype=dtype, device=device, generator=g) + vid_k = torch.randn(_L_VID, _HEADS, _DIM, dtype=dtype, device=device, generator=g) + txt_q = torch.randn(_TXT_L, _HEADS, _DIM, dtype=dtype, device=device, generator=g) + txt_k = torch.randn(_TXT_L, _HEADS, _DIM, dtype=dtype, device=device, generator=g) + vid_shape = torch.tensor([[_VID_T, _VID_H, _VID_W]], dtype=torch.long, device=device) + txt_shape = torch.tensor([[_TXT_L]], dtype=torch.long, device=device) + cache = Cache(disable=True) + return vid_q, vid_k, vid_shape, txt_q, txt_k, txt_shape, cache + + +def _legacy_get_freqs(rope: NaMMRotaryEmbedding3d, vid_shape, txt_shape): + """Reproduce the pre-rewrite ``get_freqs`` body verbatim against + ``self.get_axial_freqs`` (parent ``RotaryEmbeddingBase`` method, + unchanged by the rewrite). + """ + max_temporal = 0 + max_height = 0 + max_width = 0 + max_txt_len = 0 + for (f, h, w), l in zip(vid_shape.tolist(), txt_shape[:, 0].tolist()): + max_temporal = max(max_temporal, l + f) + max_height = max(max_height, h) + max_width = max(max_width, w) + max_txt_len = max(max_txt_len, l) + with torch.amp.autocast(device_type="cuda", enabled=False): + vid_freqs_full = rope.get_axial_freqs( + min(max_temporal + 16, 1024), + min(max_height + 4, 128), + min(max_width + 4, 128), + ).float() + txt_freqs_full = rope.get_axial_freqs(min(max_txt_len + 16, 1024)) + vid_freq_list, txt_freq_list = [], [] + for (f, h, w), l in zip(vid_shape.tolist(), txt_shape[:, 0].tolist()): + vid_freq = vid_freqs_full[l : l + f, :h, :w].reshape(-1, vid_freqs_full.size(-1)) + txt_freq = txt_freqs_full[:l].repeat(1, 3).reshape(-1, vid_freqs_full.size(-1)) + vid_freq_list.append(vid_freq) + txt_freq_list.append(txt_freq) + return torch.cat(vid_freq_list, dim=0), torch.cat(txt_freq_list, dim=0) + + +def _legacy_forward(rope: NaMMRotaryEmbedding3d, vid_q, vid_k, vid_shape, + txt_q, txt_k, txt_shape): + """Compute expected forward output via the unchanged + ``apply_rotary_emb`` wrapper fed with legacy-shape freqs. This is the + oracle. The wrapper itself is out of scope for the rewrite (Shape B). + """ + vid_freqs, txt_freqs = _legacy_get_freqs(rope, vid_shape, txt_shape) + vid_freqs = vid_freqs.to(vid_q.device) + txt_freqs = txt_freqs.to(txt_q.device) + + from einops import rearrange + + vid_q = rearrange(vid_q, "L h d -> h L d") + vid_k = rearrange(vid_k, "L h d -> h L d") + vid_q_out = seedvr_model.apply_rotary_emb(vid_freqs, vid_q.float()).to(vid_q.dtype) + vid_k_out = seedvr_model.apply_rotary_emb(vid_freqs, vid_k.float()).to(vid_k.dtype) + vid_q_out = rearrange(vid_q_out, "h L d -> L h d") + vid_k_out = rearrange(vid_k_out, "h L d -> L h d") + + txt_q = rearrange(txt_q, "L h d -> h L d") + txt_k = rearrange(txt_k, "L h d -> h L d") + txt_q_out = seedvr_model.apply_rotary_emb(txt_freqs, txt_q.float()).to(txt_q.dtype) + txt_k_out = seedvr_model.apply_rotary_emb(txt_freqs, txt_k.float()).to(txt_k.dtype) + txt_q_out = rearrange(txt_q_out, "h L d -> L h d") + txt_k_out = rearrange(txt_k_out, "h L d -> L h d") + return vid_q_out, vid_k_out, txt_q_out, txt_k_out + + +def test_namm_forward_output_tensor_equal_against_legacy_oracle(): + rope = NaMMRotaryEmbedding3d(dim=_DIM) + vid_q, vid_k, vid_shape, txt_q, txt_k, txt_shape, cache = _make_inputs() + + expected_vid_q, expected_vid_k, expected_txt_q, expected_txt_k = _legacy_forward( + rope, + vid_q.clone(), vid_k.clone(), vid_shape, + txt_q.clone(), txt_k.clone(), txt_shape, + ) + + actual_vid_q, actual_vid_k, actual_txt_q, actual_txt_k = rope.forward( + vid_q.clone(), vid_k.clone(), vid_shape, + txt_q.clone(), txt_k.clone(), txt_shape, cache, + ) + + torch.testing.assert_close(actual_vid_q, expected_vid_q, rtol=0, atol=0, + msg="vid_q output diverges from wrapper oracle") + torch.testing.assert_close(actual_vid_k, expected_vid_k, rtol=0, atol=0, + msg="vid_k output diverges from wrapper oracle") + torch.testing.assert_close(actual_txt_q, expected_txt_q, rtol=0, atol=0, + msg="txt_q output diverges from wrapper oracle") + torch.testing.assert_close(actual_txt_k, expected_txt_k, rtol=0, atol=0, + msg="txt_k output diverges from wrapper oracle") + + +# --------------------------------------------------------------------------- +# GroupNorm limit tests (test_seedvr_groupnorm_limit.py) +# --------------------------------------------------------------------------- + +_NUM_CHANNELS = 8 +_NUM_GROUPS = 4 +_TENSOR_SHAPE = (1, 8, 2, 4, 4) + +_GROUPNORM_SUBCLASSES = [ + pytest.param(comfy_ops.disable_weight_init.GroupNorm, id="disable_weight_init"), + pytest.param(comfy_ops.manual_cast.GroupNorm, id="manual_cast"), +] + + +@pytest.mark.parametrize("groupnorm_cls", _GROUPNORM_SUBCLASSES) +def test_seedvr_groupnorm_low_limit_uses_chunked_groupnorm_path(groupnorm_cls): + real_group_norm = vae_mod.F.group_norm + set_norm_limit(1e-9) + try: + gn = groupnorm_cls(num_channels=_NUM_CHANNELS, num_groups=_NUM_GROUPS) + gn.eval() + + forward_hook_calls = [] + + def _hook(module, inputs, output): + forward_hook_calls.append(tuple(inputs[0].shape)) + + spy_calls = [] + + def _group_norm_spy(input_tensor, num_groups_arg, *args, **kwargs): + spy_calls.append({"num_groups": int(num_groups_arg)}) + return real_group_norm(input_tensor, num_groups_arg, *args, **kwargs) + + handle = gn.register_forward_hook(_hook) + try: + with patch.object(vae_mod.F, "group_norm", side_effect=_group_norm_spy): + out_tensor = causal_norm_wrapper(gn, torch.randn(*_TENSOR_SHAPE)) + finally: + handle.remove() + + full_calls = len(forward_hook_calls) + chunked_calls = sum(1 for entry in spy_calls if entry["num_groups"] < _NUM_GROUPS) + + assert tuple(int(s) for s in out_tensor.shape) == _TENSOR_SHAPE + assert full_calls == 0, ( + f"low-limit GroupNorm gate must NOT take the full-forward path; got full_calls={full_calls}" + ) + assert chunked_calls > 0, ( + f"low-limit GroupNorm gate must take the chunked path; got chunked_calls={chunked_calls}" + ) + finally: + set_norm_limit(None) + + +# --------------------------------------------------------------------------- +# var_attention backend tests (test_seedvr_var_attention_backends.py) +# --------------------------------------------------------------------------- + +def test_var_attention_registry_contains_always_available_entries(): + assert attention.REGISTERED_ATTENTION_FUNCTIONS["var_attention_pytorch"] is attention.var_attention_pytorch + assert attention.REGISTERED_ATTENTION_FUNCTIONS["var_attention_sub_quad"] is attention.var_attention_sub_quad + assert attention.REGISTERED_ATTENTION_FUNCTIONS["var_attention_split"] is attention.var_attention_split + + +def test_seedvr2_7b_swin_attention_forward_uses_optimized_var_attention(monkeypatch): + dim = 8 + heads = 2 + head_dim = 4 + attn = seedvr_model.NaSwinAttention( + vid_dim=dim, + txt_dim=dim, + heads=heads, + head_dim=head_dim, + qk_bias=False, + qk_norm=seedvr_model.CustomRMSNorm, + qk_norm_eps=1e-6, + rope_type=None, + rope_dim=head_dim, + shared_weights=False, + window=(2, 1, 1), + window_method="720pwin_by_size_bysize", + version=True, + device="cpu", + dtype=torch.float32, + operations=comfy_ops.disable_weight_init, + ) + generator = torch.Generator(device="cpu").manual_seed(11) + vid = torch.randn(8, dim, generator=generator) + txt = torch.randn(3, dim, generator=generator) + vid_shape = torch.tensor([[2, 2, 2]], dtype=torch.long) + txt_shape = torch.tensor([[3]], dtype=torch.long) + calls = [] + + def fake_optimized_var_attention(**kwargs): + calls.append(kwargs) + return kwargs["q"] + + monkeypatch.setattr(seedvr_model, "optimized_var_attention", fake_optimized_var_attention) + + vid_out, txt_out = attn(vid, txt, vid_shape, txt_shape, seedvr_model.Cache(disable=True)) + + assert tuple(vid_out.shape) == (8, dim) + assert tuple(txt_out.shape) == (3, dim) + assert len(calls) == 1 + call = calls[0] + assert tuple(call["q"].shape) == (14, heads, head_dim) + assert tuple(call["k"].shape) == (14, heads, head_dim) + assert tuple(call["v"].shape) == (14, heads, head_dim) + assert call["heads"] == heads + assert call["skip_reshape"] is True + assert call["skip_output_reshape"] is True + torch.testing.assert_close( + call["cu_seqlens_q"], + torch.tensor([0, 7, 14], dtype=torch.int32), + rtol=0, + atol=0, + ) + torch.testing.assert_close( + call["cu_seqlens_k"], + torch.tensor([0, 7, 14], dtype=torch.int32), + rtol=0, + atol=0, + ) + + +# --------------------------------------------------------------------------- +# var_attention_pytorch SeedVR2 guard tests +# (test_var_attention_pytorch_seedvr2_guard.py) +# --------------------------------------------------------------------------- + +def _pytorch_guard_inputs(): + heads, head_dim, total_tokens = 2, 8, 6 + embed_dim = heads * head_dim + q = torch.randn(total_tokens, embed_dim) + k = torch.randn(total_tokens, embed_dim) + v = torch.randn(total_tokens, embed_dim) + cu = torch.tensor([0, 3, 6], dtype=torch.int32) + return q, k, v, heads, cu, cu, total_tokens, embed_dim + + +def _assert_guard_source_pin(): + src = textwrap.dedent(inspect.getsource(var_attention_pytorch)) + tree = ast.parse(src) + raise_lines = [] + nested_lines = [] + for node in ast.walk(tree): + if isinstance(node, ast.Raise) and isinstance(node.exc, ast.Call): + func = node.exc.func + if isinstance(func, ast.Name) and func.id == "RuntimeError": + raise_lines.append(node.lineno) + if isinstance(node, ast.Attribute) and node.attr == "nested_tensor_from_jagged": + nested_lines.append(node.lineno) + assert raise_lines, ( + "var_attention_pytorch has no `raise RuntimeError(...)` AST node; " + f"the SeedVR2-named guard is missing.\n--- source ---\n{src}" + ) + assert nested_lines, ( + "var_attention_pytorch source has no `nested_tensor_from_jagged` " + f"attribute access; cannot pin guard ordering.\n" + f"--- source ---\n{src}" + ) + first_raise = min(raise_lines) + first_nested = min(nested_lines) + assert first_raise < first_nested, ( + f"`raise RuntimeError(...)` first appears at line {first_raise}, " + f"but `torch.nested.nested_tensor_from_jagged` is referenced first " + f"at line {first_nested}; the guard must precede the lookup.\n" + f"--- source ---\n{src}" + ) + + +def test_missing_api_raises_seedvr2_runtime_error(monkeypatch): + monkeypatch.delattr(torch.nested, "nested_tensor_from_jagged", raising=False) + q, k, v, heads, cu_q, cu_k, _, _ = _pytorch_guard_inputs() + + with pytest.raises(RuntimeError, match=r"SeedVR2.*nested_tensor_from_jagged"): + var_attention_pytorch(q, k, v, heads, cu_q, cu_k) + + _assert_guard_source_pin() + + +def test_missing_namespace_raises_seedvr2_runtime_error(monkeypatch): + monkeypatch.delattr(torch, "nested", raising=False) + q, k, v, heads, cu_q, cu_k, _, _ = _pytorch_guard_inputs() + + with pytest.raises(RuntimeError, match=r"SeedVR2.*nested_tensor_from_jagged"): + var_attention_pytorch(q, k, v, heads, cu_q, cu_k) + + _assert_guard_source_pin() + + +def test_present_api_returns_expected_shape(): + q, k, v, heads, cu_q, cu_k, total_tokens, embed_dim = _pytorch_guard_inputs() + + torch_fx_logger = logging.getLogger("torch.fx._symbolic_trace") + old_torch_fx_level = torch_fx_logger.level + torch_fx_logger.setLevel(logging.ERROR) + try: + with warnings.catch_warnings(): + warnings.filterwarnings( + "ignore", + message="The PyTorch API of nested tensors is in prototype stage.*", + category=UserWarning, + ) + out = var_attention_pytorch(q, k, v, heads, cu_q, cu_k) + finally: + torch_fx_logger.setLevel(old_torch_fx_level) + + assert tuple(out.shape) == (total_tokens, embed_dim), ( + f"expected ({total_tokens}, {embed_dim}); got {tuple(out.shape)}" + ) + + _assert_guard_source_pin() + + +def test_malformed_offsets_propagates_torch_runtime_error(): + q, k, v, heads, _, _, _, _ = _pytorch_guard_inputs() + cu_q_bad = torch.tensor([0, 3, 7], dtype=torch.int32) + cu_k_ok = torch.tensor([0, 3, 6], dtype=torch.int32) + + with pytest.raises(RuntimeError) as exc_info: + var_attention_pytorch(q, k, v, heads, cu_q_bad, cu_k_ok) + + msg = str(exc_info.value) + assert "SeedVR2" not in msg + + _assert_guard_source_pin() diff --git a/tests-unit/comfy_test/test_seedvr2_model.py b/tests-unit/comfy_test/test_seedvr2_model.py new file mode 100644 index 000000000..b81ff2d71 --- /dev/null +++ b/tests-unit/comfy_test/test_seedvr2_model.py @@ -0,0 +1,308 @@ +"""Consolidated SeedVR2 model/graph/forward regression tests. + +Merged from: +- seedvr_model_test.py +- test_seedvr_7b_final_block_text_path.py +- test_seedvr_forward_no_device_cast.py +- test_seedvr_latent_format.py +- test_seedvr2_vae_graph_boundaries.py +""" + +from __future__ import annotations + +from unittest.mock import MagicMock + +import torch +from torch import nn + +from comfy.cli_args import args + +if not torch.cuda.is_available(): + args.cpu = True + +import comfy # noqa: E402 +import comfy.latent_formats # noqa: E402 +import comfy.ldm.seedvr.model # noqa: E402 +import comfy.ldm.seedvr.model as seedvr_model # noqa: E402 +import comfy.ldm.seedvr.vae as seedvr_vae_mod # noqa: E402 +import comfy.model_management # noqa: E402 +import comfy.sample # noqa: E402 +import comfy.sd as sd_mod # noqa: E402 +import nodes as nodes_mod # noqa: E402 +from comfy.ldm.seedvr.model import NaDiT # noqa: E402 + + +# --------------------------------------------------------------------------- +# Helpers from seedvr_model_test.py +# --------------------------------------------------------------------------- + + +def _make_standin(positive_conditioning): + class _StandIn(torch.nn.Module): + def __init__(self): + super().__init__() + self.register_buffer( + "positive_conditioning", positive_conditioning + ) + + _resolve_text_conditioning = NaDiT._resolve_text_conditioning + + return _StandIn() + + +# --------------------------------------------------------------------------- +# Helpers from test_seedvr_7b_final_block_text_path.py +# --------------------------------------------------------------------------- + + +class _StubModule(nn.Module): + def __init__(self, *args, **kwargs): + super().__init__() + + +def _capture_last_layer_flags(monkeypatch, vid_dim: int, txt_in_dim: int) -> list[bool]: + flags = [] + + class _Block(_StubModule): + def __init__(self, *args, **kwargs): + flags.append(kwargs["is_last_layer"]) + super().__init__() + + monkeypatch.setattr(seedvr_model, "NaPatchIn", _StubModule) + monkeypatch.setattr(seedvr_model, "NaPatchOut", _StubModule) + monkeypatch.setattr(seedvr_model, "TimeEmbedding", _StubModule) + monkeypatch.setattr(seedvr_model, "NaMMSRTransformerBlock", _Block) + + seedvr_model.NaDiT( + norm_eps=1e-5, + qk_rope=None, + num_layers=4, + mlp_type="normal", + vid_dim=vid_dim, + txt_in_dim=txt_in_dim, + heads=24, + mm_layers=3, + ) + + return flags + + +# --------------------------------------------------------------------------- +# Helpers from test_seedvr_latent_format.py +# --------------------------------------------------------------------------- + + +class _Model: + def __init__(self, latent_format): + self._latent_format = latent_format + + def get_model_object(self, name): + assert name == "latent_format" + return self._latent_format + + +# --------------------------------------------------------------------------- +# Helpers from test_seedvr2_vae_graph_boundaries.py +# --------------------------------------------------------------------------- + + +class _Patcher: + def get_free_memory(self, device): + return 1024 * 1024 * 1024 + + +class _EncodeWrapper(seedvr_vae_mod.VideoAutoencoderKLWrapper): + def __init__(self, encoded): + nn.Module.__init__(self) + self.encoded = encoded + self.spatial_downsample_factor = 8 + self.temporal_downsample_factor = 4 + self.seen = [] + + def encode(self, x): + self.seen.append(tuple(x.shape)) + return self.encoded.to(device=x.device, dtype=x.dtype) + + +class _DecodeWrapper(seedvr_vae_mod.VideoAutoencoderKLWrapper): + def __init__(self): + nn.Module.__init__(self) + self.spatial_downsample_factor = 8 + self.temporal_downsample_factor = 4 + self.calls = [] + + def decode(self, z, seedvr2_tiling=None): + self.calls.append({"shape": tuple(z.shape), "seedvr2_tiling": seedvr2_tiling}) + if z.ndim == 4: + b, tc, h, w = z.shape + t = tc // 16 + else: + b, _, t, h, w = z.shape + return torch.zeros(b, 3, t, h * 8, w * 8, dtype=z.dtype, device=z.device) + + +def _make_vae(wrapper): + vae = sd_mod.VAE.__new__(sd_mod.VAE) + vae.first_stage_model = wrapper + vae.device = torch.device("cpu") + vae.output_device = torch.device("cpu") + vae.vae_dtype = torch.float32 + vae.latent_channels = 16 + vae.latent_dim = 3 + vae.downscale_ratio = (lambda a: max(0, (a + 3) // 4), 8, 8) + vae.upscale_ratio = (lambda a: max(0, a * 4 - 3), 8, 8) + vae.output_channels = 3 + vae.disable_offload = True + vae.extra_1d_channel = None + vae.crop_input = False + vae.not_video = False + vae.patcher = _Patcher() + vae.process_input = lambda image: image + vae.process_output = lambda image: image.add(1.0).div(2.0).clamp(0.0, 1.0) + vae.vae_output_dtype = lambda: torch.float32 + vae.memory_used_encode = lambda shape, dtype: 1 + vae.memory_used_decode = lambda shape, dtype: 1 + vae.throw_exception_if_invalid = lambda: None + vae.vae_encode_crop_pixels = lambda pixels: pixels + vae.spacial_compression_decode = lambda: 8 + vae.temporal_compression_decode = lambda: 4 + return vae + + +# --------------------------------------------------------------------------- +# Tests from seedvr_model_test.py +# --------------------------------------------------------------------------- + + +def test_missing_context_falls_back_to_positive_buffer(): + """AC: ``context is None`` falls back to the registered + ``positive_conditioning`` buffer and runs to completion — no + silent zero substitution, no raised exception. + """ + pos_buffer = torch.full((58, 5120), 7.0) + standin = _make_standin(pos_buffer) + txt, txt_shape = standin._resolve_text_conditioning(None) + assert txt.shape == (58, 5120) + assert (txt == 7.0).all(), ( + "fallback path must use the positive_conditioning buffer " + "verbatim, not a zero tensor" + ) + assert txt_shape.shape == (1, 1) + assert txt_shape[0, 0].item() == 58 + + +# --------------------------------------------------------------------------- +# Tests from test_seedvr_7b_final_block_text_path.py +# --------------------------------------------------------------------------- + + +def test_seedvr2_7b_keeps_final_block_text_path(monkeypatch): + assert _capture_last_layer_flags(monkeypatch, vid_dim=3072, txt_in_dim=3072) == [ + False, + False, + False, + False, + ] + + +def test_seedvr2_7b_rope3d_matches_wrapper_oracle(): + rope = seedvr_model.get_na_rope("rope3d", dim=64) + generator = torch.Generator(device="cpu").manual_seed(0) + q = torch.randn(4, 2, 128, generator=generator) + k = torch.randn(4, 2, 128, generator=generator) + shape = torch.tensor([[1, 2, 2]], dtype=torch.long) + freqs = rope.get_axial_freqs(1, 2, 2).reshape(4, -1) + + expected_q = seedvr_model.apply_rotary_emb( + freqs, + q.permute(1, 0, 2).float(), + ).to(q.dtype).permute(1, 0, 2) + expected_k = seedvr_model.apply_rotary_emb( + freqs, + k.permute(1, 0, 2).float(), + ).to(k.dtype).permute(1, 0, 2) + + actual_q, actual_k = rope(q.clone(), k.clone(), shape, seedvr_model.Cache(disable=True)) + + torch.testing.assert_close(actual_q, expected_q, rtol=0, atol=0) + torch.testing.assert_close(actual_k, expected_k, rtol=0, atol=0) + + +# --------------------------------------------------------------------------- +# Tests from test_seedvr_latent_format.py +# --------------------------------------------------------------------------- + + +def test_seedvr2_latent_format_uses_16_channels_without_3d_empty_latent_expansion(): + latent_format = comfy.latent_formats.SeedVR2() + latent_image = torch.zeros(1, 1, 4, 5) + + fixed = comfy.sample.fix_empty_latent_channels(_Model(latent_format), latent_image) + + assert latent_format.latent_channels == 16 + assert latent_format.latent_dimensions == 2 + assert fixed.shape == (1, 16, 4, 5) + + +# --------------------------------------------------------------------------- +# Tests from test_seedvr2_vae_graph_boundaries.py +# --------------------------------------------------------------------------- + + +def test_seedvr2_encode_and_encode_tiled_preserve_native_latent_contract(monkeypatch): + monkeypatch.setattr(sd_mod.model_management, "load_models_gpu", lambda *a, **k: None) + + encoded = torch.full((1, 16, 2, 4, 5), 2.0) + vae = _make_vae(_EncodeWrapper(encoded)) + pixels = torch.zeros(1, 5, 32, 40, 3) + + node_output = nodes_mod.VAEEncode().encode(vae, pixels)[0] + node_latent = node_output["samples"] + assert set(node_output) == {"samples"} + assert tuple(node_latent.shape) == (1, 16, 2, 4, 5) + assert node_latent.dtype == torch.float32 + assert node_latent.stride()[-1] == 1 + assert torch.equal(node_latent, torch.full_like(node_latent, 2.0 * 0.9152)) + + tiled = torch.full((1, 16, 2, 4, 5), 3.0) + monkeypatch.setattr(seedvr_vae_mod, "tiled_vae", MagicMock(return_value=tiled)) + tiled_output = nodes_mod.VAEEncodeTiled().encode( + vae, + pixels, + tile_size=512, + overlap=64, + temporal_size=16, + temporal_overlap=4, + )[0] + tiled_latent = tiled_output["samples"] + assert set(tiled_output) == {"samples"} + assert tuple(tiled_latent.shape) == (1, 16, 2, 4, 5) + assert tiled_latent.dtype == torch.float32 + assert torch.equal(tiled_latent, torch.full_like(tiled_latent, 3.0 * 0.9152)) + + +def test_vaedecode_tiled_visible_inputs_are_seedvr2_decode_tiling_authority(monkeypatch): + monkeypatch.setattr(sd_mod.model_management, "load_models_gpu", lambda *a, **k: None) + vae = _make_vae(_DecodeWrapper()) + + nodes_mod.VAEDecodeTiled().decode( + vae, + {"samples": torch.zeros(1, 16, 2, 4, 5)}, + tile_size=512, + overlap=64, + temporal_size=16, + temporal_overlap=4, + ) + + assert vae.first_stage_model.calls == [ + { + "shape": (1, 16, 2, 4, 5), + "seedvr2_tiling": { + "enable_tiling": True, + "tile_size": (512, 512), + "tile_overlap": (64, 64), + "temporal_size": 16, + "temporal_overlap": 4, + }, + } + ] diff --git a/tests-unit/comfy_test/test_seedvr2_vae_decode.py b/tests-unit/comfy_test/test_seedvr2_vae_decode.py new file mode 100644 index 000000000..ea9f978f3 --- /dev/null +++ b/tests-unit/comfy_test/test_seedvr2_vae_decode.py @@ -0,0 +1,91 @@ +from unittest.mock import patch + +import pytest +import torch +import torch.nn as nn + +from comfy.cli_args import args as cli_args + +if not torch.cuda.is_available(): + cli_args.cpu = True + +import comfy.ldm.seedvr.vae as vae_mod # noqa: E402 +from comfy_extras import nodes_seedvr # noqa: E402 + + +def _make_wrapper() -> vae_mod.VideoAutoencoderKLWrapper: + wrapper = vae_mod.VideoAutoencoderKLWrapper.__new__( + vae_mod.VideoAutoencoderKLWrapper + ) + nn.Module.__init__(wrapper) + return wrapper + + +def _fingerprint_decode_(self, z, return_dict=True): + b = int(z.shape[0]) + t = int(z.shape[2]) + h = int(z.shape[3]) + w = int(z.shape[4]) + out = torch.empty(b, 3, t, h * 8, w * 8) + for batch_idx in range(b): + out[batch_idx].fill_(float(batch_idx + 1)) + return out + + +def _decode_with_patches(wrapper, z): + with patch.object(vae_mod.VideoAutoencoderKL, "decode_", _fingerprint_decode_): + return wrapper.decode(z) + + +def test_decode_b2_t3_multi_frame_batch_unchanged(): + wrapper = _make_wrapper() + + out = _decode_with_patches(wrapper, torch.zeros(2, 16 * 3, 2, 2)) + + assert tuple(out.shape) == (2, 3, 3, 16, 16) + + +class _Wrapper(vae_mod.VideoAutoencoderKLWrapper): + def __init__(self): + nn.Module.__init__(self) + self.calls = [] + + def parameters(self): + return iter([torch.nn.Parameter(torch.zeros(()))]) + +def _decode_stub(self, latent): + self.calls.append(tuple(latent.shape)) + return torch.zeros(latent.shape[0], 3, latent.shape[2], latent.shape[3] * 8, latent.shape[4] * 8) + + +def test_seedvr2_wrapper_decode_accepts_5d_channel_first_latents_without_preprocessor_state(): + wrapper = _Wrapper() + + with patch.object(vae_mod.VideoAutoencoderKL, "decode_", _decode_stub): + out = wrapper.decode(torch.zeros(1, 16, 2, 4, 5)) + + assert tuple(out.shape) == (1, 3, 2, 32, 40) + assert wrapper.calls == [(1, 16, 2, 4, 5)] + + +def test_seedvr2_wrapper_decode_rejects_wrong_rank_latents(): + wrapper = _Wrapper() + + with pytest.raises(RuntimeError, match=r"latent input must be 4-D collapsed .* or 5-D"): + wrapper.decode(torch.zeros(1, 16, 4)) + + +def _t_padded(t_in: int) -> int: + if t_in == 1: + return 1 + if t_in <= 4: + return 5 + if (t_in - 1) % 4 == 0: + return t_in + return t_in + (4 - ((t_in - 1) % 4)) + + +@pytest.mark.parametrize("t_in", [1, 5, 9]) +def test_t_padded_matches_cut_videos(t_in): + dummy = torch.zeros(1, t_in, 1, 1, 1) + assert nodes_seedvr.cut_videos(dummy).shape[1] == _t_padded(t_in) diff --git a/tests-unit/comfy_test/test_seedvr2_vae_tiled.py b/tests-unit/comfy_test/test_seedvr2_vae_tiled.py new file mode 100644 index 000000000..442480149 --- /dev/null +++ b/tests-unit/comfy_test/test_seedvr2_vae_tiled.py @@ -0,0 +1,350 @@ +from contextlib import ExitStack +from unittest.mock import MagicMock, patch + +import torch +import torch.nn as nn + +from comfy.cli_args import args as cli_args + +if not torch.cuda.is_available(): + cli_args.cpu = True + +import comfy.ldm.seedvr.vae as vae_mod # noqa: E402 +import comfy.ldm.seedvr.vae as seedvr_vae_mod # noqa: E402 +import comfy.sd as sd_mod # noqa: E402 +from comfy.ldm.seedvr.vae import MemoryState, tiled_vae # noqa: E402 + + +# --------------------------------------------------------------------------- +# From test_seedvr_vae_tiled_decode_latent_min_size_override.py +# --------------------------------------------------------------------------- + + +def test_runtime_decode_zero_temporal_size_disables_slicing_for_call(): + from comfy.ldm.seedvr.vae import MemoryState, VideoAutoencoderKL, tiled_vae + + class StubVAEModel(torch.nn.Module): + def __init__(self): + super().__init__() + self.slicing_latent_min_size = 2 + self.spatial_downsample_factor = 8 + self.temporal_downsample_factor = 4 + self.device = torch.device("cpu") + self.use_slicing = True + self._dummy = torch.nn.Parameter(torch.zeros(1, dtype=torch.float32)) + self.decode_min_sizes = [] + self.memory_states = [] + + def decode_(self, t_chunk): + self.decode_min_sizes.append(self.slicing_latent_min_size) + return VideoAutoencoderKL.slicing_decode(self, t_chunk) + + def _decode(self, z, memory_state=MemoryState.DISABLED): + self.memory_states.append(memory_state) + b, c, d, h, w = z.shape + return torch.zeros((b, 3, d, h * 8, w * 8), dtype=z.dtype) + + vae = StubVAEModel() + z = torch.zeros((1, 16, 5, 8, 8), dtype=torch.float32) + + tiled_vae( + z, + vae, + tile_size=(64, 64), + tile_overlap=(0, 0), + temporal_size=0, + temporal_overlap=0, + encode=False, + ) + + assert vae.decode_min_sizes == [5] + assert vae.memory_states == [MemoryState.DISABLED] + assert vae.slicing_latent_min_size == 2 + + +# --------------------------------------------------------------------------- +# From test_seedvr_vae_tiled_encode_runt_slice_override.py +# --------------------------------------------------------------------------- + + +def test_zero_temporal_size_preserves_min_size_when_encode_raises(): + from comfy.ldm.seedvr.vae import tiled_vae + + class RaisingVAEModel(torch.nn.Module): + def __init__(self): + super().__init__() + self.slicing_sample_min_size = 4 + self.spatial_downsample_factor = 8 + self.temporal_downsample_factor = 4 + self.device = torch.device("cpu") + self._dummy = torch.nn.Parameter(torch.zeros(1, dtype=torch.float32)) + + def encode(self, t_chunk): + raise RuntimeError("simulated encode failure") + + vae = RaisingVAEModel() + x = torch.zeros((1, 3, 12, 64, 64), dtype=torch.float32) + + raised = False + try: + tiled_vae( + x, + vae, + tile_size=(64, 64), + tile_overlap=(0, 0), + temporal_size=0, + temporal_overlap=0, + encode=True, + ) + except RuntimeError as exc: + if "simulated encode failure" not in str(exc): + raise + raised = True + + assert raised + assert vae.slicing_sample_min_size == 4 + + +# --------------------------------------------------------------------------- +# From test_seedvr_vae_tiled_temporal_slicing.py +# --------------------------------------------------------------------------- + + +class _SlicingDecodeVAE(nn.Module): + def __init__(self, slicing_latent_min_size): + super().__init__() + self.slicing_latent_min_size = slicing_latent_min_size + self.spatial_downsample_factor = 8 + self.temporal_downsample_factor = 4 + self.device = torch.device("cpu") + self.use_slicing = True + self._dummy = nn.Parameter(torch.zeros(1, dtype=torch.float32)) + self.decode_min_sizes = [] + self.memory_states = [] + + def decode_(self, z): + self.decode_min_sizes.append(self.slicing_latent_min_size) + return vae_mod.VideoAutoencoderKL.slicing_decode(self, z) + + def _decode(self, z, memory_state=MemoryState.DISABLED): + self.memory_states.append(memory_state) + x = z[:, :1].repeat( + 1, + 3, + 1, + self.spatial_downsample_factor, + self.spatial_downsample_factor, + ) + return x + + +def test_decode_tiled_vae_maps_temporal_args_to_latent_slicing_min_size(): + vae = _SlicingDecodeVAE(slicing_latent_min_size=2) + z = torch.arange(1 * 16 * 5 * 8 * 8, dtype=torch.float32).reshape(1, 16, 5, 8, 8) + + tiled_vae( + z, + vae, + tile_size=(64, 64), + tile_overlap=(0, 0), + temporal_size=12, + temporal_overlap=4, + encode=False, + ) + + assert vae.decode_min_sizes == [2] + assert vae.memory_states == [MemoryState.INITIALIZING, MemoryState.ACTIVE] + assert vae.slicing_latent_min_size == 2 + + wrapper = vae_mod.VideoAutoencoderKLWrapper.__new__( + vae_mod.VideoAutoencoderKLWrapper + ) + nn.Module.__init__(wrapper) + seedvr2_tiling = { + "enable_tiling": True, + "tile_size": (64, 64), + "tile_overlap": (0, 0), + "temporal_size": 8, + "temporal_overlap": 7, + } + + captured = {} + + def _fake_tiled_vae(latent, model, **kwargs): + captured.update(kwargs) + return torch.zeros(1, 3, 1, 16, 16) + + with ( + patch.object(vae_mod, "tiled_vae", side_effect=_fake_tiled_vae), + patch.object(vae_mod, "lab_color_transfer", side_effect=lambda content, style: content), + ): + wrapper.decode(torch.zeros(1, 16, 2, 2), seedvr2_tiling=seedvr2_tiling) + + assert captured["temporal_overlap"] == 7 + + +# --------------------------------------------------------------------------- +# From test_vae_decode_tiled_dispatcher_seedvr2_4d.py +# --------------------------------------------------------------------------- + + +def _force_oom(*a, **k): + raise torch.cuda.OutOfMemoryError("forced OOM for dispatcher test") + + +def _make_vae(first_stage_model, latent_channels, latent_dim): + vae = sd_mod.VAE.__new__(sd_mod.VAE) + vae.first_stage_model = first_stage_model + vae.patcher = MagicMock() + vae.patcher.get_free_memory = MagicMock(return_value=8 * 1024 * 1024 * 1024) + vae.device = vae.output_device = torch.device("cpu") + vae.vae_dtype = torch.float32 + vae.disable_offload = True + vae.extra_1d_channel = None + vae.upscale_ratio = vae.downscale_ratio = 8 + vae.upscale_index_formula = vae.downscale_index_formula = None + vae.output_channels = 3 + vae.latent_channels = latent_channels + vae.latent_dim = latent_dim + vae.vae_output_dtype = lambda: torch.float32 + vae.spacial_compression_decode = lambda: 8 + vae.process_input = lambda x: x + vae.process_output = lambda x: x + vae.throw_exception_if_invalid = lambda: None + vae.memory_used_decode = lambda *a, **k: 1 + return vae + + +def _dispatch(vae, samples, seedvr2_call, generic_call, patch_wrapper_decode): + mm = sd_mod.model_management + with ExitStack() as stack: + stack.enter_context(patch.object(mm, "raise_non_oom", lambda e: None)) + stack.enter_context(patch.object(mm, "load_models_gpu", lambda *a, **k: None)) + stack.enter_context(patch.object(mm, "soft_empty_cache", lambda: None)) + stack.enter_context(patch.object(sd_mod.VAE, "decode_tiled_seedvr2", seedvr2_call)) + stack.enter_context(patch.object(sd_mod.VAE, "decode_tiled_", generic_call)) + if patch_wrapper_decode: + stack.enter_context(patch.object( + seedvr_vae_mod.VideoAutoencoderKLWrapper, "decode", + side_effect=_force_oom)) + vae.decode(samples) + + +def test_4d_seedvr2_latent_routes_to_decode_tiled_seedvr2(): + wrapper = seedvr_vae_mod.VideoAutoencoderKLWrapper.__new__( + seedvr_vae_mod.VideoAutoencoderKLWrapper) + vae = _make_vae(wrapper, latent_channels=16, latent_dim=3) + seedvr2_call = MagicMock(return_value=torch.zeros(1, 3, 9, 64, 64)) + generic_call = MagicMock(return_value=torch.zeros(1, 3, 64, 64)) + _dispatch(vae, torch.zeros(1, 16 * 3, 8, 8), seedvr2_call, generic_call, True) + assert seedvr2_call.call_count == 1 + assert generic_call.call_count == 0 + + +def test_4d_non_seedvr2_latent_still_routes_to_generic_decode_tiled(): + first_stage = MagicMock() + first_stage.decode = MagicMock(side_effect=_force_oom) + vae = _make_vae(first_stage, latent_channels=4, latent_dim=2) + seedvr2_call = MagicMock(return_value=torch.zeros(1, 3, 9, 64, 64)) + generic_call = MagicMock(return_value=torch.zeros(1, 3, 64, 64)) + _dispatch(vae, torch.zeros(1, 4, 8, 8), seedvr2_call, generic_call, False) + assert generic_call.call_count == 1 + assert seedvr2_call.call_count == 0 + + +# --------------------------------------------------------------------------- +# From test_vae_encode_tiled_fallback_dispatcher_seedvr2.py +# --------------------------------------------------------------------------- + + +def _populate_common_vae_attrs_fallback(vae): + vae.patcher = MagicMock() + vae.patcher.get_free_memory = MagicMock(return_value=8 * 1024 * 1024 * 1024) + vae.device = torch.device("cpu") + vae.output_device = torch.device("cpu") + vae.vae_dtype = torch.float32 + vae.disable_offload = True + vae.extra_1d_channel = None + vae.upscale_ratio = 8 + vae.upscale_index_formula = None + vae.output_channels = 3 + vae.latent_channels = 16 + vae.latent_dim = 3 + vae.downscale_ratio = 8 + vae.downscale_index_formula = None + vae.not_video = False + vae.crop_input = False + vae.pad_channel_value = None + + vae.vae_output_dtype = lambda: torch.float32 + vae.spacial_compression_encode = lambda: 8 + vae.process_input = lambda x: x + vae.process_output = lambda x: x + vae.throw_exception_if_invalid = lambda: None + vae.memory_used_encode = lambda *a, **k: 1 + + +def _make_seedvr2_vae_fallback(): + vae = sd_mod.VAE.__new__(sd_mod.VAE) + wrapper = seedvr_vae_mod.VideoAutoencoderKLWrapper.__new__( + seedvr_vae_mod.VideoAutoencoderKLWrapper + ) + vae.first_stage_model = wrapper + _populate_common_vae_attrs_fallback(vae) + return vae + + +def _make_non_seedvr2_vae_fallback(): + vae = sd_mod.VAE.__new__(sd_mod.VAE) + vae.first_stage_model = MagicMock() + _populate_common_vae_attrs_fallback(vae) + return vae + + +def _force_regular_encode_oom(*args, **kwargs): + raise torch.cuda.OutOfMemoryError("forced OOM for dispatcher test") + + +def test_seedvr2_3d_routes_to_encode_tiled_seedvr2_on_oom(): + vae = _make_seedvr2_vae_fallback() + pixel_samples = torch.zeros((1, 8, 64, 64, 3)) + + seedvr2_call = MagicMock(return_value=torch.zeros(1, 16, 2, 8, 8)) + generic_call = MagicMock(return_value=torch.zeros(1, 16, 2, 8, 8)) + + with patch.object(sd_mod.model_management, "raise_non_oom", + lambda e: None), \ + patch.object(sd_mod.model_management, "load_models_gpu", + lambda *a, **k: None), \ + patch.object(sd_mod.model_management, "soft_empty_cache", + lambda: None), \ + patch.object(seedvr_vae_mod.VideoAutoencoderKLWrapper, "encode", + side_effect=_force_regular_encode_oom), \ + patch.object(sd_mod.VAE, "encode_tiled_seedvr2", seedvr2_call, + create=True), \ + patch.object(sd_mod.VAE, "encode_tiled_3d", generic_call): + vae.encode(pixel_samples) + + assert seedvr2_call.call_count == 1, ( + f"Expected encode_tiled_seedvr2 to be called once for a SeedVR2 3D " + f"input under OOM fallback; got {seedvr2_call.call_count} calls." + ) + assert generic_call.call_count == 0, ( + f"encode_tiled_3d must NOT be called for a SeedVR2 input; got " + f"{generic_call.call_count} calls." + ) + + +def test_non_seedvr2_encode_tiled_3d_default_overlap_is_concrete(): + vae = _make_non_seedvr2_vae_fallback() + vae.downscale_ratio = (lambda a: max(1, a // 4), 8, 8) + vae.upscale_ratio = (lambda a: a * 4, 8, 8) + generic_call = MagicMock(return_value=torch.zeros(1, 16, 2, 8, 8)) + pixel_samples = torch.zeros((1, 8, 64, 64, 3)) + + with patch.object(sd_mod.model_management, "load_models_gpu", + lambda *a, **k: None), \ + patch.object(sd_mod.VAE, "encode_tiled_3d", generic_call): + vae.encode_tiled(pixel_samples) + + assert generic_call.call_args.kwargs["overlap"] == (1, 64, 64) diff --git a/tests-unit/comfy_test/test_seedvr_progressive_sampler.py b/tests-unit/comfy_test/test_seedvr_progressive_sampler.py new file mode 100644 index 000000000..5e5969921 --- /dev/null +++ b/tests-unit/comfy_test/test_seedvr_progressive_sampler.py @@ -0,0 +1,126 @@ +"""Unit tests for ``comfy_extras.nodes_seedvr.SeedVR2ProgressiveSampler``.""" + +from unittest.mock import patch + +import pytest +import torch + +from comfy.cli_args import args as cli_args + +if not torch.cuda.is_available(): + cli_args.cpu = True + +import comfy.sample # noqa: E402 +import comfy_extras.nodes_seedvr as nodes_seedvr_mod # noqa: E402 +from comfy_extras.nodes_seedvr import SeedVR2ProgressiveSampler # noqa: E402 + +_LAT_C = 16 +_COND_C = 17 + + +def _make_inputs(B: int = 1, T: int = 5, H: int = 8, W: int = 8): + """Build minimal SeedVR2-shaped sampling inputs.""" + samples_5d = torch.arange( + B * _LAT_C * T * H * W, dtype=torch.float32 + ).reshape(B, _LAT_C, T, H, W) + samples = samples_5d.reshape(B, _LAT_C * T, H, W).contiguous() + + cond_5d = torch.arange( + B * _COND_C * T * H * W, dtype=torch.float32 + ).reshape(B, _COND_C, T, H, W) + 10000.0 + cond = cond_5d.reshape(B, _COND_C * T, H, W).contiguous() + + text_pos = torch.zeros(1, 4, 32) + text_neg = torch.zeros(1, 4, 32) + positive = [[text_pos, {"condition": cond.clone()}]] + negative = [[text_neg, {"condition": cond.clone()}]] + latent_image = {"samples": samples} + return latent_image, positive, negative, samples_5d, cond_5d + + +def _identity_fix_empty(model, latent_image, downscale_ratio_spacial=None): + return latent_image + + +def _fingerprinted_prepare_noise(latent_image, seed, batch_inds=None): + """Return a tensor whose values encode ``(seed, position)``.""" + base = torch.arange( + latent_image.numel(), dtype=torch.float32 + ).reshape(latent_image.shape) + return base + float(seed) * 1e6 + + +def test_progressive_sampler_schema_exposes_manual_default_auto_chunking(): + schema = SeedVR2ProgressiveSampler.define_schema() + inputs = {item.id: item for item in schema.inputs} + + assert inputs["chunking_mode"].options == ["manual", "auto"] + assert inputs["chunking_mode"].default == "manual" + + +def test_auto_chunking_walks_two_three_four_chunk_ladder(): + """Auto mode must walk 2-, 3-, then 4-chunk geometries on OOM.""" + latent, pos, neg, _, _ = _make_inputs(T=17) + calls = [] + + def _oom_until_four_chunks(model, noise, steps, cfg, sampler_name, + scheduler, positive, negative, + latent_image, denoise=1.0, + noise_mask=None, seed=None): + calls.append(tuple(latent_image.shape)) + if latent_image.shape[1] > _LAT_C * 5: + raise torch.cuda.OutOfMemoryError("chunk too large") + return latent_image.clone() + + with patch.object(comfy.sample, "sample", + side_effect=_oom_until_four_chunks), \ + patch.object(comfy.sample, "fix_empty_latent_channels", + side_effect=_identity_fix_empty), \ + patch.object(comfy.sample, "prepare_noise", + side_effect=_fingerprinted_prepare_noise), \ + patch.object(nodes_seedvr_mod.comfy.model_management, + "soft_empty_cache") as soft_empty: + out = SeedVR2ProgressiveSampler.execute( + model=None, seed=0, steps=2, cfg=1.0, + sampler_name="euler", scheduler="simple", + positive=pos, negative=neg, latent_image=latent, + denoise=1.0, frames_per_chunk=65, temporal_overlap=0, + chunking_mode="auto", + ) + + assert calls[:4] == [ + (1, _LAT_C * 17, 8, 8), + (1, _LAT_C * 9, 8, 8), + (1, _LAT_C * 6, 8, 8), + (1, _LAT_C * 5, 8, 8), + ] + assert torch.equal(out.result[0]["samples"], latent["samples"]) + assert soft_empty.call_count == 3 + + +@pytest.mark.parametrize("bad_chunk", [0, -1, 2]) +def test_t3_invalid_frames_per_chunk_raises_value_error(bad_chunk): + """``frames_per_chunk`` violating 4n+1 (or <1) must raise ``ValueError`` before any model invocation.""" + latent, pos, neg, _, _ = _make_inputs(T=5) + + sampler_called = {"n": 0} + + def _should_not_be_called(*args, **kwargs): + sampler_called["n"] += 1 + return torch.zeros(1) + + with patch.object(comfy.sample, "sample", + side_effect=_should_not_be_called), \ + patch.object(comfy.sample, "fix_empty_latent_channels", + side_effect=_identity_fix_empty), \ + patch.object(comfy.sample, "prepare_noise", + side_effect=_fingerprinted_prepare_noise): + with pytest.raises(ValueError) as excinfo: + SeedVR2ProgressiveSampler.execute( + model=None, seed=0, steps=2, cfg=1.0, + sampler_name="euler", scheduler="simple", + positive=pos, negative=neg, latent_image=latent, + denoise=1.0, frames_per_chunk=bad_chunk, temporal_overlap=0, + ) + assert str(bad_chunk) in str(excinfo.value) + assert sampler_called["n"] == 0