From b5c8be8b1db44ded07cb1b437b9f33ebff5848c1 Mon Sep 17 00:00:00 2001 From: comfyanonymous Date: Tue, 18 Nov 2025 19:37:20 -0500 Subject: [PATCH 1/4] ComfyUI 0.3.70 --- comfyui_version.py | 2 +- pyproject.toml | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/comfyui_version.py b/comfyui_version.py index 1e554eb9f..9b77aabe9 100644 --- a/comfyui_version.py +++ b/comfyui_version.py @@ -1,3 +1,3 @@ # This file is automatically generated by the build process when version is # updated in pyproject.toml. -__version__ = "0.3.69" +__version__ = "0.3.70" diff --git a/pyproject.toml b/pyproject.toml index a14b383b3..289b7145b 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -1,6 +1,6 @@ [project] name = "ComfyUI" -version = "0.3.69" +version = "0.3.70" readme = "README.md" license = { file = "LICENSE" } requires-python = ">=3.9" From 17027f2a6a20a31e2c6f3be2b1a06f39ad3a68d9 Mon Sep 17 00:00:00 2001 From: comfyanonymous <121283862+comfyanonymous@users.noreply.github.com> Date: Tue, 18 Nov 2025 19:36:03 -0800 Subject: [PATCH 2/4] Add a way to disable the final norm in the llama based TE models. (#10794) --- comfy/text_encoders/llama.py | 18 +++++++++++++++--- 1 file changed, 15 insertions(+), 3 deletions(-) diff --git a/comfy/text_encoders/llama.py b/comfy/text_encoders/llama.py index c050759fe..feb44bbb0 100644 --- a/comfy/text_encoders/llama.py +++ b/comfy/text_encoders/llama.py @@ -32,6 +32,7 @@ class Llama2Config: q_norm = None k_norm = None rope_scale = None + final_norm: bool = True @dataclass class Qwen25_3BConfig: @@ -53,6 +54,7 @@ class Qwen25_3BConfig: q_norm = None k_norm = None rope_scale = None + final_norm: bool = True @dataclass class Qwen25_7BVLI_Config: @@ -74,6 +76,7 @@ class Qwen25_7BVLI_Config: q_norm = None k_norm = None rope_scale = None + final_norm: bool = True @dataclass class Gemma2_2B_Config: @@ -96,6 +99,7 @@ class Gemma2_2B_Config: k_norm = None sliding_attention = None rope_scale = None + final_norm: bool = True @dataclass class Gemma3_4B_Config: @@ -118,6 +122,7 @@ class Gemma3_4B_Config: k_norm = "gemma3" sliding_attention = [False, False, False, False, False, 1024] rope_scale = [1.0, 8.0] + final_norm: bool = True class RMSNorm(nn.Module): def __init__(self, dim: int, eps: float = 1e-5, add=False, device=None, dtype=None): @@ -366,7 +371,12 @@ class Llama2_(nn.Module): transformer(config, index=i, device=device, dtype=dtype, ops=ops) for i in range(config.num_hidden_layers) ]) - self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps, add=config.rms_norm_add, device=device, dtype=dtype) + + if config.final_norm: + self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps, add=config.rms_norm_add, device=device, dtype=dtype) + else: + self.norm = None + # self.lm_head = ops.Linear(config.hidden_size, config.vocab_size, bias=False, device=device, dtype=dtype) def forward(self, x, attention_mask=None, embeds=None, num_tokens=None, intermediate_output=None, final_layer_norm_intermediate=True, dtype=None, position_ids=None, embeds_info=[]): @@ -421,14 +431,16 @@ class Llama2_(nn.Module): if i == intermediate_output: intermediate = x.clone() - x = self.norm(x) + if self.norm is not None: + x = self.norm(x) + if all_intermediate is not None: all_intermediate.append(x.unsqueeze(1).clone()) if all_intermediate is not None: intermediate = torch.cat(all_intermediate, dim=1) - if intermediate is not None and final_layer_norm_intermediate: + if intermediate is not None and final_layer_norm_intermediate and self.norm is not None: intermediate = self.norm(intermediate) return x, intermediate From 65ee24c9789b93660ebe978a3186486f105298c2 Mon Sep 17 00:00:00 2001 From: Alexander Piskun <13381981+bigcat88@users.noreply.github.com> Date: Wed, 19 Nov 2025 11:25:28 +0200 Subject: [PATCH 3/4] change display name of PreviewAny node to "Preview as Text" (#10796) --- comfy_extras/nodes_preview_any.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/comfy_extras/nodes_preview_any.py b/comfy_extras/nodes_preview_any.py index e749fa6ae..139b07c93 100644 --- a/comfy_extras/nodes_preview_any.py +++ b/comfy_extras/nodes_preview_any.py @@ -39,5 +39,5 @@ NODE_CLASS_MAPPINGS = { } NODE_DISPLAY_NAME_MAPPINGS = { - "PreviewAny": "Preview Any", + "PreviewAny": "Preview as Text", } From 6a1d3a1ae131f3fff7f45a7e835eb10e9d1338ee Mon Sep 17 00:00:00 2001 From: Alexander Piskun <13381981+bigcat88@users.noreply.github.com> Date: Thu, 20 Nov 2025 00:49:01 +0200 Subject: [PATCH 4/4] convert hunyuan3d.py to V3 schema (#10664) --- comfy_api/latest/__init__.py | 4 +- comfy_api/latest/_io.py | 5 +- comfy_api/latest/_util/__init__.py | 3 + comfy_api/latest/_util/geometry_types.py | 12 + comfy_extras/nodes_hunyuan3d.py | 274 +++++++++++++---------- 5 files changed, 178 insertions(+), 120 deletions(-) create mode 100644 comfy_api/latest/_util/geometry_types.py diff --git a/comfy_api/latest/__init__.py b/comfy_api/latest/__init__.py index b7a3fa9c1..176ae36e0 100644 --- a/comfy_api/latest/__init__.py +++ b/comfy_api/latest/__init__.py @@ -7,7 +7,7 @@ from comfy_api.internal.singleton import ProxiedSingleton from comfy_api.internal.async_to_sync import create_sync_class from comfy_api.latest._input import ImageInput, AudioInput, MaskInput, LatentInput, VideoInput from comfy_api.latest._input_impl import VideoFromFile, VideoFromComponents -from comfy_api.latest._util import VideoCodec, VideoContainer, VideoComponents +from comfy_api.latest._util import VideoCodec, VideoContainer, VideoComponents, MESH, VOXEL from . import _io as io from . import _ui as ui # from comfy_api.latest._resources import _RESOURCES as resources #noqa: F401 @@ -104,6 +104,8 @@ class Types: VideoCodec = VideoCodec VideoContainer = VideoContainer VideoComponents = VideoComponents + MESH = MESH + VOXEL = VOXEL ComfyAPI = ComfyAPI_latest diff --git a/comfy_api/latest/_io.py b/comfy_api/latest/_io.py index 0b701260f..863254ce7 100644 --- a/comfy_api/latest/_io.py +++ b/comfy_api/latest/_io.py @@ -27,6 +27,7 @@ from comfy_api.internal import (_ComfyNodeInternal, _NodeOutputInternal, classpr prune_dict, shallow_clone_class) from comfy_api.latest._resources import Resources, ResourcesLocal from comfy_execution.graph_utils import ExecutionBlocker +from ._util import MESH, VOXEL # from comfy_extras.nodes_images import SVG as SVG_ # NOTE: needs to be moved before can be imported due to circular reference @@ -656,11 +657,11 @@ class LossMap(ComfyTypeIO): @comfytype(io_type="VOXEL") class Voxel(ComfyTypeIO): - Type = Any # TODO: VOXEL class is defined in comfy_extras/nodes_hunyuan3d.py; should be moved to somewhere else before referenced directly in v3 + Type = VOXEL @comfytype(io_type="MESH") class Mesh(ComfyTypeIO): - Type = Any # TODO: MESH class is defined in comfy_extras/nodes_hunyuan3d.py; should be moved to somewhere else before referenced directly in v3 + Type = MESH @comfytype(io_type="HOOKS") class Hooks(ComfyTypeIO): diff --git a/comfy_api/latest/_util/__init__.py b/comfy_api/latest/_util/__init__.py index 9019c46db..fc5431dda 100644 --- a/comfy_api/latest/_util/__init__.py +++ b/comfy_api/latest/_util/__init__.py @@ -1,8 +1,11 @@ from .video_types import VideoContainer, VideoCodec, VideoComponents +from .geometry_types import VOXEL, MESH __all__ = [ # Utility Types "VideoContainer", "VideoCodec", "VideoComponents", + "VOXEL", + "MESH", ] diff --git a/comfy_api/latest/_util/geometry_types.py b/comfy_api/latest/_util/geometry_types.py new file mode 100644 index 000000000..385122778 --- /dev/null +++ b/comfy_api/latest/_util/geometry_types.py @@ -0,0 +1,12 @@ +import torch + + +class VOXEL: + def __init__(self, data: torch.Tensor): + self.data = data + + +class MESH: + def __init__(self, vertices: torch.Tensor, faces: torch.Tensor): + self.vertices = vertices + self.faces = faces diff --git a/comfy_extras/nodes_hunyuan3d.py b/comfy_extras/nodes_hunyuan3d.py index f6e71e0a8..adca14f62 100644 --- a/comfy_extras/nodes_hunyuan3d.py +++ b/comfy_extras/nodes_hunyuan3d.py @@ -7,63 +7,79 @@ from comfy.ldm.modules.diffusionmodules.mmdit import get_1d_sincos_pos_embed_fro import folder_paths import comfy.model_management from comfy.cli_args import args +from typing_extensions import override +from comfy_api.latest import ComfyExtension, IO, Types +from comfy_api.latest._util import MESH, VOXEL # only for backward compatibility if someone import it from this file (will be removed later) # noqa -class EmptyLatentHunyuan3Dv2: + +class EmptyLatentHunyuan3Dv2(IO.ComfyNode): @classmethod - def INPUT_TYPES(s): - return { - "required": { - "resolution": ("INT", {"default": 3072, "min": 1, "max": 8192}), - "batch_size": ("INT", {"default": 1, "min": 1, "max": 4096, "tooltip": "The number of latent images in the batch."}), - } - } + def define_schema(cls): + return IO.Schema( + node_id="EmptyLatentHunyuan3Dv2", + category="latent/3d", + inputs=[ + IO.Int.Input("resolution", default=3072, min=1, max=8192), + IO.Int.Input("batch_size", default=1, min=1, max=4096, tooltip="The number of latent images in the batch."), + ], + outputs=[ + IO.Latent.Output(), + ] + ) - RETURN_TYPES = ("LATENT",) - FUNCTION = "generate" - - CATEGORY = "latent/3d" - - def generate(self, resolution, batch_size): + @classmethod + def execute(cls, resolution, batch_size) -> IO.NodeOutput: latent = torch.zeros([batch_size, 64, resolution], device=comfy.model_management.intermediate_device()) - return ({"samples": latent, "type": "hunyuan3dv2"}, ) + return IO.NodeOutput({"samples": latent, "type": "hunyuan3dv2"}) -class Hunyuan3Dv2Conditioning: + generate = execute # TODO: remove + + +class Hunyuan3Dv2Conditioning(IO.ComfyNode): @classmethod - def INPUT_TYPES(s): - return {"required": {"clip_vision_output": ("CLIP_VISION_OUTPUT",), - }} + def define_schema(cls): + return IO.Schema( + node_id="Hunyuan3Dv2Conditioning", + category="conditioning/video_models", + inputs=[ + IO.ClipVisionOutput.Input("clip_vision_output"), + ], + outputs=[ + IO.Conditioning.Output(display_name="positive"), + IO.Conditioning.Output(display_name="negative"), + ] + ) - RETURN_TYPES = ("CONDITIONING", "CONDITIONING") - RETURN_NAMES = ("positive", "negative") - - FUNCTION = "encode" - - CATEGORY = "conditioning/video_models" - - def encode(self, clip_vision_output): + @classmethod + def execute(cls, clip_vision_output) -> IO.NodeOutput: embeds = clip_vision_output.last_hidden_state positive = [[embeds, {}]] negative = [[torch.zeros_like(embeds), {}]] - return (positive, negative) + return IO.NodeOutput(positive, negative) + + encode = execute # TODO: remove -class Hunyuan3Dv2ConditioningMultiView: +class Hunyuan3Dv2ConditioningMultiView(IO.ComfyNode): @classmethod - def INPUT_TYPES(s): - return {"required": {}, - "optional": {"front": ("CLIP_VISION_OUTPUT",), - "left": ("CLIP_VISION_OUTPUT",), - "back": ("CLIP_VISION_OUTPUT",), - "right": ("CLIP_VISION_OUTPUT",), }} + def define_schema(cls): + return IO.Schema( + node_id="Hunyuan3Dv2ConditioningMultiView", + category="conditioning/video_models", + inputs=[ + IO.ClipVisionOutput.Input("front", optional=True), + IO.ClipVisionOutput.Input("left", optional=True), + IO.ClipVisionOutput.Input("back", optional=True), + IO.ClipVisionOutput.Input("right", optional=True), + ], + outputs=[ + IO.Conditioning.Output(display_name="positive"), + IO.Conditioning.Output(display_name="negative"), + ] + ) - RETURN_TYPES = ("CONDITIONING", "CONDITIONING") - RETURN_NAMES = ("positive", "negative") - - FUNCTION = "encode" - - CATEGORY = "conditioning/video_models" - - def encode(self, front=None, left=None, back=None, right=None): + @classmethod + def execute(cls, front=None, left=None, back=None, right=None) -> IO.NodeOutput: all_embeds = [front, left, back, right] out = [] pos_embeds = None @@ -76,29 +92,35 @@ class Hunyuan3Dv2ConditioningMultiView: embeds = torch.cat(out, dim=1) positive = [[embeds, {}]] negative = [[torch.zeros_like(embeds), {}]] - return (positive, negative) + return IO.NodeOutput(positive, negative) + + encode = execute # TODO: remove -class VOXEL: - def __init__(self, data): - self.data = data - -class VAEDecodeHunyuan3D: +class VAEDecodeHunyuan3D(IO.ComfyNode): @classmethod - def INPUT_TYPES(s): - return {"required": {"samples": ("LATENT", ), - "vae": ("VAE", ), - "num_chunks": ("INT", {"default": 8000, "min": 1000, "max": 500000}), - "octree_resolution": ("INT", {"default": 256, "min": 16, "max": 512}), - }} - RETURN_TYPES = ("VOXEL",) - FUNCTION = "decode" + def define_schema(cls): + return IO.Schema( + node_id="VAEDecodeHunyuan3D", + category="latent/3d", + inputs=[ + IO.Latent.Input("samples"), + IO.Vae.Input("vae"), + IO.Int.Input("num_chunks", default=8000, min=1000, max=500000), + IO.Int.Input("octree_resolution", default=256, min=16, max=512), + ], + outputs=[ + IO.Voxel.Output(), + ] + ) - CATEGORY = "latent/3d" + @classmethod + def execute(cls, vae, samples, num_chunks, octree_resolution) -> IO.NodeOutput: + voxels = Types.VOXEL(vae.decode(samples["samples"], vae_options={"num_chunks": num_chunks, "octree_resolution": octree_resolution})) + return IO.NodeOutput(voxels) + + decode = execute # TODO: remove - def decode(self, vae, samples, num_chunks, octree_resolution): - voxels = VOXEL(vae.decode(samples["samples"], vae_options={"num_chunks": num_chunks, "octree_resolution": octree_resolution})) - return (voxels, ) def voxel_to_mesh(voxels, threshold=0.5, device=None): if device is None: @@ -396,24 +418,24 @@ def voxel_to_mesh_surfnet(voxels, threshold=0.5, device=None): return final_vertices, faces -class MESH: - def __init__(self, vertices, faces): - self.vertices = vertices - self.faces = faces - -class VoxelToMeshBasic: +class VoxelToMeshBasic(IO.ComfyNode): @classmethod - def INPUT_TYPES(s): - return {"required": {"voxel": ("VOXEL", ), - "threshold": ("FLOAT", {"default": 0.6, "min": -1.0, "max": 1.0, "step": 0.01}), - }} - RETURN_TYPES = ("MESH",) - FUNCTION = "decode" + def define_schema(cls): + return IO.Schema( + node_id="VoxelToMeshBasic", + category="3d", + inputs=[ + IO.Voxel.Input("voxel"), + IO.Float.Input("threshold", default=0.6, min=-1.0, max=1.0, step=0.01), + ], + outputs=[ + IO.Mesh.Output(), + ] + ) - CATEGORY = "3d" - - def decode(self, voxel, threshold): + @classmethod + def execute(cls, voxel, threshold) -> IO.NodeOutput: vertices = [] faces = [] for x in voxel.data: @@ -421,21 +443,29 @@ class VoxelToMeshBasic: vertices.append(v) faces.append(f) - return (MESH(torch.stack(vertices), torch.stack(faces)), ) + return IO.NodeOutput(Types.MESH(torch.stack(vertices), torch.stack(faces))) -class VoxelToMesh: + decode = execute # TODO: remove + + +class VoxelToMesh(IO.ComfyNode): @classmethod - def INPUT_TYPES(s): - return {"required": {"voxel": ("VOXEL", ), - "algorithm": (["surface net", "basic"], ), - "threshold": ("FLOAT", {"default": 0.6, "min": -1.0, "max": 1.0, "step": 0.01}), - }} - RETURN_TYPES = ("MESH",) - FUNCTION = "decode" + def define_schema(cls): + return IO.Schema( + node_id="VoxelToMesh", + category="3d", + inputs=[ + IO.Voxel.Input("voxel"), + IO.Combo.Input("algorithm", options=["surface net", "basic"]), + IO.Float.Input("threshold", default=0.6, min=-1.0, max=1.0, step=0.01), + ], + outputs=[ + IO.Mesh.Output(), + ] + ) - CATEGORY = "3d" - - def decode(self, voxel, algorithm, threshold): + @classmethod + def execute(cls, voxel, algorithm, threshold) -> IO.NodeOutput: vertices = [] faces = [] @@ -449,7 +479,9 @@ class VoxelToMesh: vertices.append(v) faces.append(f) - return (MESH(torch.stack(vertices), torch.stack(faces)), ) + return IO.NodeOutput(Types.MESH(torch.stack(vertices), torch.stack(faces))) + + decode = execute # TODO: remove def save_glb(vertices, faces, filepath, metadata=None): @@ -581,31 +613,32 @@ def save_glb(vertices, faces, filepath, metadata=None): return filepath -class SaveGLB: +class SaveGLB(IO.ComfyNode): @classmethod - def INPUT_TYPES(s): - return {"required": {"mesh": ("MESH", ), - "filename_prefix": ("STRING", {"default": "mesh/ComfyUI"}), }, - "hidden": {"prompt": "PROMPT", "extra_pnginfo": "EXTRA_PNGINFO"}, } + def define_schema(cls): + return IO.Schema( + node_id="SaveGLB", + category="3d", + is_output_node=True, + inputs=[ + IO.Mesh.Input("mesh"), + IO.String.Input("filename_prefix", default="mesh/ComfyUI"), + ], + hidden=[IO.Hidden.prompt, IO.Hidden.extra_pnginfo] + ) - RETURN_TYPES = () - FUNCTION = "save" - - OUTPUT_NODE = True - - CATEGORY = "3d" - - def save(self, mesh, filename_prefix, prompt=None, extra_pnginfo=None): + @classmethod + def execute(cls, mesh, filename_prefix) -> IO.NodeOutput: full_output_folder, filename, counter, subfolder, filename_prefix = folder_paths.get_save_image_path(filename_prefix, folder_paths.get_output_directory()) results = [] metadata = {} if not args.disable_metadata: - if prompt is not None: - metadata["prompt"] = json.dumps(prompt) - if extra_pnginfo is not None: - for x in extra_pnginfo: - metadata[x] = json.dumps(extra_pnginfo[x]) + if cls.hidden.prompt is not None: + metadata["prompt"] = json.dumps(cls.hidden.prompt) + if cls.hidden.extra_pnginfo is not None: + for x in cls.hidden.extra_pnginfo: + metadata[x] = json.dumps(cls.hidden.extra_pnginfo[x]) for i in range(mesh.vertices.shape[0]): f = f"{filename}_{counter:05}_.glb" @@ -616,15 +649,22 @@ class SaveGLB: "type": "output" }) counter += 1 - return {"ui": {"3d": results}} + return IO.NodeOutput(ui={"3d": results}) -NODE_CLASS_MAPPINGS = { - "EmptyLatentHunyuan3Dv2": EmptyLatentHunyuan3Dv2, - "Hunyuan3Dv2Conditioning": Hunyuan3Dv2Conditioning, - "Hunyuan3Dv2ConditioningMultiView": Hunyuan3Dv2ConditioningMultiView, - "VAEDecodeHunyuan3D": VAEDecodeHunyuan3D, - "VoxelToMeshBasic": VoxelToMeshBasic, - "VoxelToMesh": VoxelToMesh, - "SaveGLB": SaveGLB, -} +class Hunyuan3dExtension(ComfyExtension): + @override + async def get_node_list(self) -> list[type[IO.ComfyNode]]: + return [ + EmptyLatentHunyuan3Dv2, + Hunyuan3Dv2Conditioning, + Hunyuan3Dv2ConditioningMultiView, + VAEDecodeHunyuan3D, + VoxelToMeshBasic, + VoxelToMesh, + SaveGLB, + ] + + +async def comfy_entrypoint() -> Hunyuan3dExtension: + return Hunyuan3dExtension()