Merge branch 'comfyanonymous:master' into master

CODEOWNERS

@@ -1,3 +1,4 @@
 # Admins
 * @comfyanonymous
 * @kosinkadink
+* @guill

comfy/cli_args.py

@@ -121,6 +121,12 @@ upcast.add_argument("--force-upcast-attention", action="store_true", help="Force
 upcast.add_argument("--dont-upcast-attention", action="store_true", help="Disable all upcasting of attention. Should be unnecessary except for debugging.")
 
 
+parser.add_argument("--enable-manager", action="store_true", help="Enable the ComfyUI-Manager feature.")
+manager_group = parser.add_mutually_exclusive_group()
+manager_group.add_argument("--disable-manager-ui", action="store_true", help="Disables only the ComfyUI-Manager UI and endpoints. Scheduled installations and similar background tasks will still operate.")
+manager_group.add_argument("--enable-manager-legacy-ui", action="store_true", help="Enables the legacy UI of ComfyUI-Manager")
+
+
 vram_group = parser.add_mutually_exclusive_group()
 vram_group.add_argument("--gpu-only", action="store_true", help="Store and run everything (text encoders/CLIP models, etc... on the GPU).")
 vram_group.add_argument("--highvram", action="store_true", help="By default models will be unloaded to CPU memory after being used. This option keeps them in GPU memory.")
@@ -168,6 +174,7 @@ parser.add_argument("--multi-user", action="store_true", help="Enables per-user
 parser.add_argument("--verbose", default='INFO', const='DEBUG', nargs="?", choices=['DEBUG', 'INFO', 'WARNING', 'ERROR', 'CRITICAL'], help='Set the logging level')
 parser.add_argument("--log-stdout", action="store_true", help="Send normal process output to stdout instead of stderr (default).")
 
+
 # The default built-in provider hosted under web/
 DEFAULT_VERSION_STRING = "comfyanonymous/ComfyUI@latest"
 

comfy/ldm/chroma/model.py

@@ -40,7 +40,8 @@ class ChromaParams:
     out_dim: int
     hidden_dim: int
     n_layers: int
-
+    txt_ids_dims: list
+    vec_in_dim: int
 
 
 

comfy/ldm/flux/layers.py

@@ -57,6 +57,35 @@ class MLPEmbedder(nn.Module):
     def forward(self, x: Tensor) -> Tensor:
         return self.out_layer(self.silu(self.in_layer(x)))
 
+class YakMLP(nn.Module):
+    def __init__(self, hidden_size: int, intermediate_size: int, dtype=None, device=None, operations=None):
+        super().__init__()
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.gate_proj = operations.Linear(self.hidden_size, self.intermediate_size, bias=True, dtype=dtype, device=device)
+        self.up_proj = operations.Linear(self.hidden_size, self.intermediate_size, bias=True, dtype=dtype, device=device)
+        self.down_proj = operations.Linear(self.intermediate_size, self.hidden_size, bias=True, dtype=dtype, device=device)
+        self.act_fn = nn.SiLU()
+
+    def forward(self, x: Tensor) -> Tensor:
+        down_proj = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
+        return down_proj
+
+def build_mlp(hidden_size, mlp_hidden_dim, mlp_silu_act=False, yak_mlp=False, dtype=None, device=None, operations=None):
+    if yak_mlp:
+        return YakMLP(hidden_size, mlp_hidden_dim, dtype=dtype, device=device, operations=operations)
+    if mlp_silu_act:
+        return nn.Sequential(
+            operations.Linear(hidden_size, mlp_hidden_dim * 2, bias=False, dtype=dtype, device=device),
+            SiLUActivation(),
+            operations.Linear(mlp_hidden_dim, hidden_size, bias=False, dtype=dtype, device=device),
+        )
+    else:
+        return nn.Sequential(
+            operations.Linear(hidden_size, mlp_hidden_dim, bias=True, dtype=dtype, device=device),
+            nn.GELU(approximate="tanh"),
+            operations.Linear(mlp_hidden_dim, hidden_size, bias=True, dtype=dtype, device=device),
+        )
 
 class RMSNorm(torch.nn.Module):
     def __init__(self, dim: int, dtype=None, device=None, operations=None):
@@ -140,7 +169,7 @@ class SiLUActivation(nn.Module):
 
 
 class DoubleStreamBlock(nn.Module):
-    def __init__(self, hidden_size: int, num_heads: int, mlp_ratio: float, qkv_bias: bool = False, flipped_img_txt=False, modulation=True, mlp_silu_act=False, proj_bias=True, dtype=None, device=None, operations=None):
+    def __init__(self, hidden_size: int, num_heads: int, mlp_ratio: float, qkv_bias: bool = False, flipped_img_txt=False, modulation=True, mlp_silu_act=False, proj_bias=True, yak_mlp=False, dtype=None, device=None, operations=None):
         super().__init__()
 
         mlp_hidden_dim = int(hidden_size * mlp_ratio)
@@ -156,18 +185,7 @@ class DoubleStreamBlock(nn.Module):
 
         self.img_norm2 = operations.LayerNorm(hidden_size, elementwise_affine=False, eps=1e-6, dtype=dtype, device=device)
 
-        if mlp_silu_act:
-            self.img_mlp = nn.Sequential(
-                operations.Linear(hidden_size, mlp_hidden_dim * 2, bias=False, dtype=dtype, device=device),
-                SiLUActivation(),
-                operations.Linear(mlp_hidden_dim, hidden_size, bias=False, dtype=dtype, device=device),
-            )
-        else:
-            self.img_mlp = nn.Sequential(
-                operations.Linear(hidden_size, mlp_hidden_dim, bias=True, dtype=dtype, device=device),
-                nn.GELU(approximate="tanh"),
-                operations.Linear(mlp_hidden_dim, hidden_size, bias=True, dtype=dtype, device=device),
-            )
+        self.img_mlp = build_mlp(hidden_size, mlp_hidden_dim, mlp_silu_act=mlp_silu_act, yak_mlp=yak_mlp, dtype=dtype, device=device, operations=operations)
 
         if self.modulation:
             self.txt_mod = Modulation(hidden_size, double=True, dtype=dtype, device=device, operations=operations)
@@ -177,18 +195,7 @@ class DoubleStreamBlock(nn.Module):
 
         self.txt_norm2 = operations.LayerNorm(hidden_size, elementwise_affine=False, eps=1e-6, dtype=dtype, device=device)
 
-        if mlp_silu_act:
-            self.txt_mlp = nn.Sequential(
-                operations.Linear(hidden_size, mlp_hidden_dim * 2, bias=False, dtype=dtype, device=device),
-                SiLUActivation(),
-                operations.Linear(mlp_hidden_dim, hidden_size, bias=False, dtype=dtype, device=device),
-            )
-        else:
-            self.txt_mlp = nn.Sequential(
-                operations.Linear(hidden_size, mlp_hidden_dim, bias=True, dtype=dtype, device=device),
-                nn.GELU(approximate="tanh"),
-                operations.Linear(mlp_hidden_dim, hidden_size, bias=True, dtype=dtype, device=device),
-            )
+        self.txt_mlp = build_mlp(hidden_size, mlp_hidden_dim, mlp_silu_act=mlp_silu_act, yak_mlp=yak_mlp, dtype=dtype, device=device, operations=operations)
 
         self.flipped_img_txt = flipped_img_txt
 
@@ -275,6 +282,7 @@ class SingleStreamBlock(nn.Module):
         modulation=True,
         mlp_silu_act=False,
         bias=True,
+        yak_mlp=False,
         dtype=None,
         device=None,
         operations=None
@@ -288,12 +296,17 @@ class SingleStreamBlock(nn.Module):
         self.mlp_hidden_dim = int(hidden_size * mlp_ratio)
 
         self.mlp_hidden_dim_first = self.mlp_hidden_dim
+        self.yak_mlp = yak_mlp
         if mlp_silu_act:
             self.mlp_hidden_dim_first = int(hidden_size * mlp_ratio * 2)
             self.mlp_act = SiLUActivation()
         else:
             self.mlp_act = nn.GELU(approximate="tanh")
 
+        if self.yak_mlp:
+            self.mlp_hidden_dim_first *= 2
+            self.mlp_act = nn.SiLU()
+
         # qkv and mlp_in
         self.linear1 = operations.Linear(hidden_size, hidden_size * 3 + self.mlp_hidden_dim_first, bias=bias, dtype=dtype, device=device)
         # proj and mlp_out
@@ -325,7 +338,10 @@ class SingleStreamBlock(nn.Module):
         attn = attention(q, k, v, pe=pe, mask=attn_mask, transformer_options=transformer_options)
         del q, k, v
         # compute activation in mlp stream, cat again and run second linear layer
-        mlp = self.mlp_act(mlp)
+        if self.yak_mlp:
+            mlp = self.mlp_act(mlp[..., self.mlp_hidden_dim_first // 2:]) * mlp[..., :self.mlp_hidden_dim_first // 2]
+        else:
+            mlp = self.mlp_act(mlp)
         output = self.linear2(torch.cat((attn, mlp), 2))
         x += apply_mod(output, mod.gate, None, modulation_dims)
         if x.dtype == torch.float16:

comfy/ldm/flux/model.py

@@ -15,7 +15,8 @@ from .layers import (
     MLPEmbedder,
     SingleStreamBlock,
     timestep_embedding,
-    Modulation
+    Modulation,
+    RMSNorm
 )
 
 @dataclass
@@ -34,11 +35,14 @@ class FluxParams:
     patch_size: int
     qkv_bias: bool
     guidance_embed: bool
+    txt_ids_dims: list
     global_modulation: bool = False
     mlp_silu_act: bool = False
     ops_bias: bool = True
     default_ref_method: str = "offset"
     ref_index_scale: float = 1.0
+    yak_mlp: bool = False
+    txt_norm: bool = False
 
 
 class Flux(nn.Module):
@@ -76,6 +80,11 @@ class Flux(nn.Module):
         )
         self.txt_in = operations.Linear(params.context_in_dim, self.hidden_size, bias=params.ops_bias, dtype=dtype, device=device)
 
+        if params.txt_norm:
+            self.txt_norm = RMSNorm(params.context_in_dim, dtype=dtype, device=device, operations=operations)
+        else:
+            self.txt_norm = None
+
         self.double_blocks = nn.ModuleList(
             [
                 DoubleStreamBlock(
@@ -86,6 +95,7 @@ class Flux(nn.Module):
                     modulation=params.global_modulation is False,
                     mlp_silu_act=params.mlp_silu_act,
                     proj_bias=params.ops_bias,
+                    yak_mlp=params.yak_mlp,
                     dtype=dtype, device=device, operations=operations
                 )
                 for _ in range(params.depth)
@@ -94,7 +104,7 @@ class Flux(nn.Module):
 
         self.single_blocks = nn.ModuleList(
             [
-                SingleStreamBlock(self.hidden_size, self.num_heads, mlp_ratio=params.mlp_ratio, modulation=params.global_modulation is False, mlp_silu_act=params.mlp_silu_act, bias=params.ops_bias, dtype=dtype, device=device, operations=operations)
+                SingleStreamBlock(self.hidden_size, self.num_heads, mlp_ratio=params.mlp_ratio, modulation=params.global_modulation is False, mlp_silu_act=params.mlp_silu_act, bias=params.ops_bias, yak_mlp=params.yak_mlp, dtype=dtype, device=device, operations=operations)
                 for _ in range(params.depth_single_blocks)
             ]
         )
@@ -150,6 +160,8 @@ class Flux(nn.Module):
                 y = torch.zeros((img.shape[0], self.params.vec_in_dim), device=img.device, dtype=img.dtype)
             vec = vec + self.vector_in(y[:, :self.params.vec_in_dim])
 
+        if self.txt_norm is not None:
+            txt = self.txt_norm(txt)
         txt = self.txt_in(txt)
 
         vec_orig = vec
@@ -332,8 +344,9 @@ class Flux(nn.Module):
 
         txt_ids = torch.zeros((bs, context.shape[1], len(self.params.axes_dim)), device=x.device, dtype=torch.float32)
 
-        if len(self.params.axes_dim) == 4: # Flux 2
-            txt_ids[:, :, 3] = torch.linspace(0, context.shape[1] - 1, steps=context.shape[1], device=x.device, dtype=torch.float32)
+        if len(self.params.txt_ids_dims) > 0:
+            for i in self.params.txt_ids_dims:
+                txt_ids[:, :, i] = torch.linspace(0, context.shape[1] - 1, steps=context.shape[1], device=x.device, dtype=torch.float32)
 
         out = self.forward_orig(img, img_ids, context, txt_ids, timestep, y, guidance, control, transformer_options, attn_mask=kwargs.get("attention_mask", None))
         out = out[:, :img_tokens]

comfy/model_detection.py

@@ -208,12 +208,12 @@ def detect_unet_config(state_dict, key_prefix, metadata=None):
             dit_config["theta"] = 2000
             dit_config["out_channels"] = 128
             dit_config["global_modulation"] = True
-            dit_config["vec_in_dim"] = None
             dit_config["mlp_silu_act"] = True
             dit_config["qkv_bias"] = False
             dit_config["ops_bias"] = False
             dit_config["default_ref_method"] = "index"
             dit_config["ref_index_scale"] = 10.0
+            dit_config["txt_ids_dims"] = [3]
             patch_size = 1
         else:
             dit_config["image_model"] = "flux"
@@ -223,6 +223,7 @@ def detect_unet_config(state_dict, key_prefix, metadata=None):
             dit_config["theta"] = 10000
             dit_config["out_channels"] = 16
             dit_config["qkv_bias"] = True
+            dit_config["txt_ids_dims"] = []
             patch_size = 2
 
         dit_config["in_channels"] = 16
@@ -245,6 +246,8 @@ def detect_unet_config(state_dict, key_prefix, metadata=None):
         vec_in_key = '{}vector_in.in_layer.weight'.format(key_prefix)
         if vec_in_key in state_dict_keys:
             dit_config["vec_in_dim"] = state_dict[vec_in_key].shape[1]
+        else:
+            dit_config["vec_in_dim"] = None
 
         dit_config["depth"] = count_blocks(state_dict_keys, '{}double_blocks.'.format(key_prefix) + '{}.')
         dit_config["depth_single_blocks"] = count_blocks(state_dict_keys, '{}single_blocks.'.format(key_prefix) + '{}.')
@@ -270,6 +273,11 @@ def detect_unet_config(state_dict, key_prefix, metadata=None):
                 dit_config["nerf_embedder_dtype"] = torch.float32
         else:
             dit_config["guidance_embed"] = "{}guidance_in.in_layer.weight".format(key_prefix) in state_dict_keys
+            dit_config["yak_mlp"] = '{}double_blocks.0.img_mlp.gate_proj.weight'.format(key_prefix) in state_dict_keys
+            dit_config["txt_norm"] = "{}txt_norm.scale".format(key_prefix) in state_dict_keys
+            if dit_config["yak_mlp"] and dit_config["txt_norm"]:  # Ovis model
+                dit_config["txt_ids_dims"] = [1, 2]
+
         return dit_config
 
     if '{}t5_yproj.weight'.format(key_prefix) in state_dict_keys: #Genmo mochi preview

comfy/sd.py

@@ -53,6 +53,7 @@ import comfy.text_encoders.omnigen2
 import comfy.text_encoders.qwen_image
 import comfy.text_encoders.hunyuan_image
 import comfy.text_encoders.z_image
+import comfy.text_encoders.ovis
 
 import comfy.model_patcher
 import comfy.lora
@@ -956,6 +957,7 @@ class CLIPType(Enum):
     QWEN_IMAGE = 18
     HUNYUAN_IMAGE = 19
     HUNYUAN_VIDEO_15 = 20
+    OVIS = 21
 
 
 def load_clip(ckpt_paths, embedding_directory=None, clip_type=CLIPType.STABLE_DIFFUSION, model_options={}):
@@ -987,6 +989,7 @@ class TEModel(Enum):
     MISTRAL3_24B = 14
     MISTRAL3_24B_PRUNED_FLUX2 = 15
     QWEN3_4B = 16
+    QWEN3_2B = 17
 
 
 def detect_te_model(sd):
@@ -1020,9 +1023,12 @@ def detect_te_model(sd):
         if weight.shape[0] == 512:
             return TEModel.QWEN25_7B
     if "model.layers.0.post_attention_layernorm.weight" in sd:
-        if 'model.layers.0.self_attn.q_norm.weight' in sd:
-            return TEModel.QWEN3_4B
         weight = sd['model.layers.0.post_attention_layernorm.weight']
+        if 'model.layers.0.self_attn.q_norm.weight' in sd:
+            if weight.shape[0] == 2560:
+                return TEModel.QWEN3_4B
+            elif weight.shape[0] == 2048:
+                return TEModel.QWEN3_2B
         if weight.shape[0] == 5120:
             if "model.layers.39.post_attention_layernorm.weight" in sd:
                 return TEModel.MISTRAL3_24B
@@ -1150,6 +1156,9 @@ def load_text_encoder_state_dicts(state_dicts=[], embedding_directory=None, clip
         elif te_model == TEModel.QWEN3_4B:
             clip_target.clip = comfy.text_encoders.z_image.te(**llama_detect(clip_data))
             clip_target.tokenizer = comfy.text_encoders.z_image.ZImageTokenizer
+        elif te_model == TEModel.QWEN3_2B:
+            clip_target.clip = comfy.text_encoders.ovis.te(**llama_detect(clip_data))
+            clip_target.tokenizer = comfy.text_encoders.ovis.OvisTokenizer
         else:
             # clip_l
             if clip_type == CLIPType.SD3:

comfy/text_encoders/llama.py

@@ -100,6 +100,28 @@ class Qwen3_4BConfig:
     rope_scale = None
     final_norm: bool = True
 
+@dataclass
+class Ovis25_2BConfig:
+    vocab_size: int = 151936
+    hidden_size: int = 2048
+    intermediate_size: int = 6144
+    num_hidden_layers: int = 28
+    num_attention_heads: int = 16
+    num_key_value_heads: int = 8
+    max_position_embeddings: int = 40960
+    rms_norm_eps: float = 1e-6
+    rope_theta: float = 1000000.0
+    transformer_type: str = "llama"
+    head_dim = 128
+    rms_norm_add = False
+    mlp_activation = "silu"
+    qkv_bias = False
+    rope_dims = None
+    q_norm = "gemma3"
+    k_norm = "gemma3"
+    rope_scale = None
+    final_norm: bool = True
+
 @dataclass
 class Qwen25_7BVLI_Config:
     vocab_size: int = 152064
@@ -542,6 +564,15 @@ class Qwen3_4B(BaseLlama, torch.nn.Module):
         self.model = Llama2_(config, device=device, dtype=dtype, ops=operations)
         self.dtype = dtype
 
+class Ovis25_2B(BaseLlama, torch.nn.Module):
+    def __init__(self, config_dict, dtype, device, operations):
+        super().__init__()
+        config = Ovis25_2BConfig(**config_dict)
+        self.num_layers = config.num_hidden_layers
+
+        self.model = Llama2_(config, device=device, dtype=dtype, ops=operations)
+        self.dtype = dtype
+
 class Qwen25_7BVLI(BaseLlama, torch.nn.Module):
     def __init__(self, config_dict, dtype, device, operations):
         super().__init__()

comfy/text_encoders/ovis.py

@@ -0,0 +1,69 @@
+from transformers import Qwen2Tokenizer
+import comfy.text_encoders.llama
+from comfy import sd1_clip
+import os
+import torch
+import numbers
+
+class Qwen3Tokenizer(sd1_clip.SDTokenizer):
+    def __init__(self, embedding_directory=None, tokenizer_data={}):
+        tokenizer_path = os.path.join(os.path.dirname(os.path.realpath(__file__)), "qwen25_tokenizer")
+        super().__init__(tokenizer_path, pad_with_end=False, embedding_size=2048, embedding_key='qwen3_2b', tokenizer_class=Qwen2Tokenizer, has_start_token=False, has_end_token=False, pad_to_max_length=False, max_length=99999999, min_length=284, pad_token=151643, tokenizer_data=tokenizer_data)
+
+
+class OvisTokenizer(sd1_clip.SD1Tokenizer):
+    def __init__(self, embedding_directory=None, tokenizer_data={}):
+        super().__init__(embedding_directory=embedding_directory, tokenizer_data=tokenizer_data, name="qwen3_2b", tokenizer=Qwen3Tokenizer)
+        self.llama_template = "<|im_start|>user\nDescribe the image by detailing the color, quantity, text, shape, size, texture, spatial relationships of the objects and background: {}<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n"
+
+    def tokenize_with_weights(self, text, return_word_ids=False, llama_template=None, **kwargs):
+        if llama_template is None:
+            llama_text = self.llama_template.format(text)
+        else:
+            llama_text = llama_template.format(text)
+
+        tokens = super().tokenize_with_weights(llama_text, return_word_ids=return_word_ids, disable_weights=True, **kwargs)
+        return tokens
+
+class Ovis25_2BModel(sd1_clip.SDClipModel):
+    def __init__(self, device="cpu", layer="last", layer_idx=None, dtype=None, attention_mask=True, model_options={}):
+        super().__init__(device=device, layer=layer, layer_idx=layer_idx, textmodel_json_config={}, dtype=dtype, special_tokens={"pad": 151643}, layer_norm_hidden_state=False, model_class=comfy.text_encoders.llama.Ovis25_2B, enable_attention_masks=attention_mask, return_attention_masks=False, zero_out_masked=True, model_options=model_options)
+
+
+class OvisTEModel(sd1_clip.SD1ClipModel):
+    def __init__(self, device="cpu", dtype=None, model_options={}):
+        super().__init__(device=device, dtype=dtype, name="qwen3_2b", clip_model=Ovis25_2BModel, model_options=model_options)
+
+    def encode_token_weights(self, token_weight_pairs, template_end=-1):
+        out, pooled = super().encode_token_weights(token_weight_pairs)
+        tok_pairs = token_weight_pairs["qwen3_2b"][0]
+        count_im_start = 0
+        if template_end == -1:
+            for i, v in enumerate(tok_pairs):
+                elem = v[0]
+                if not torch.is_tensor(elem):
+                    if isinstance(elem, numbers.Integral):
+                        if elem == 4004 and count_im_start < 1:
+                            template_end = i
+                            count_im_start += 1
+
+            if out.shape[1] > (template_end + 1):
+                if tok_pairs[template_end + 1][0] == 25:
+                    template_end += 1
+
+        out = out[:, template_end:]
+        return out, pooled, {}
+
+
+def te(dtype_llama=None, llama_scaled_fp8=None, llama_quantization_metadata=None):
+    class OvisTEModel_(OvisTEModel):
+        def __init__(self, device="cpu", dtype=None, model_options={}):
+            if llama_scaled_fp8 is not None and "scaled_fp8" not in model_options:
+                model_options = model_options.copy()
+                model_options["scaled_fp8"] = llama_scaled_fp8
+            if dtype_llama is not None:
+                dtype = dtype_llama
+            if llama_quantization_metadata is not None:
+                model_options["quantization_metadata"] = llama_quantization_metadata
+            super().__init__(device=device, dtype=dtype, model_options=model_options)
+    return OvisTEModel_

comfy_api/feature_flags.py

@@ -13,6 +13,7 @@ from comfy.cli_args import args
 SERVER_FEATURE_FLAGS: Dict[str, Any] = {
     "supports_preview_metadata": True,
     "max_upload_size": args.max_upload_size * 1024 * 1024, # Convert MB to bytes
+    "extension": {"manager": {"supports_v4": True}},
 }
 
 

comfy_api/latest/_input_impl/video_types.py

@@ -336,7 +336,10 @@ class VideoFromComponents(VideoInput):
             raise ValueError("Only MP4 format is supported for now")
         if codec != VideoCodec.AUTO and codec != VideoCodec.H264:
             raise ValueError("Only H264 codec is supported for now")
-        with av.open(path, mode='w', options={'movflags': 'use_metadata_tags'}) as output:
+        extra_kwargs = {}
+        if format != VideoContainer.AUTO:
+            extra_kwargs["format"] = format.value
+        with av.open(path, mode='w', options={'movflags': 'use_metadata_tags'}, **extra_kwargs) as output:
             # Add metadata before writing any streams
             if metadata is not None:
                 for key, value in metadata.items():

comfy_api_nodes/apis/kling_api.py

@@ -0,0 +1,66 @@
+from pydantic import BaseModel, Field
+
+
+class OmniProText2VideoRequest(BaseModel):
+    model_name: str = Field(..., description="kling-video-o1")
+    aspect_ratio: str = Field(..., description="'16:9', '9:16' or '1:1'")
+    duration: str = Field(..., description="'5' or '10'")
+    prompt: str = Field(...)
+    mode: str = Field("pro")
+
+
+class OmniParamImage(BaseModel):
+    image_url: str = Field(...)
+    type: str | None = Field(None, description="Can be 'first_frame' or 'end_frame'")
+
+
+class OmniParamVideo(BaseModel):
+    video_url: str = Field(...)
+    refer_type: str | None = Field(..., description="Can be 'base' or 'feature'")
+    keep_original_sound: str = Field(..., description="'yes' or 'no'")
+
+
+class OmniProFirstLastFrameRequest(BaseModel):
+    model_name: str = Field(..., description="kling-video-o1")
+    image_list: list[OmniParamImage] = Field(..., min_length=1, max_length=7)
+    duration: str = Field(..., description="'5' or '10'")
+    prompt: str = Field(...)
+    mode: str = Field("pro")
+
+
+class OmniProReferences2VideoRequest(BaseModel):
+    model_name: str = Field(..., description="kling-video-o1")
+    aspect_ratio: str | None = Field(..., description="'16:9', '9:16' or '1:1'")
+    image_list: list[OmniParamImage] | None = Field(
+        None, max_length=7, description="Max length 4 when video is present."
+    )
+    video_list: list[OmniParamVideo] | None = Field(None, max_length=1)
+    duration: str | None = Field(..., description="From 3 to 10.")
+    prompt: str = Field(...)
+    mode: str = Field("pro")
+
+
+class TaskStatusVideoResult(BaseModel):
+    duration: str | None = Field(None, description="Total video duration")
+    id: str | None = Field(None, description="Generated video ID")
+    url: str | None = Field(None, description="URL for generated video")
+
+
+class TaskStatusVideoResults(BaseModel):
+    videos: list[TaskStatusVideoResult] | None = Field(None)
+
+
+class TaskStatusVideoResponseData(BaseModel):
+    created_at: int | None = Field(None, description="Task creation time")
+    updated_at: int | None = Field(None, description="Task update time")
+    task_status: str | None = None
+    task_status_msg: str | None = Field(None, description="Additional failure reason. Only for polling endpoint.")
+    task_id: str | None = Field(None, description="Task ID")
+    task_result: TaskStatusVideoResults | None = Field(None)
+
+
+class TaskStatusVideoResponse(BaseModel):
+    code: int | None = Field(None, description="Error code")
+    message: str | None = Field(None, description="Error message")
+    request_id: str | None = Field(None, description="Request ID")
+    data: TaskStatusVideoResponseData | None = Field(None)

comfy_api_nodes/nodes_kling.py

@@ -4,13 +4,13 @@ For source of truth on the allowed permutations of request fields, please refere
 - [Compatibility Table](https://app.klingai.com/global/dev/document-api/apiReference/model/skillsMap)
 """
 
-import math
 import logging
-
-from typing_extensions import override
+import math
 
 import torch
+from typing_extensions import override
 
+from comfy_api.latest import IO, ComfyExtension, Input, InputImpl
 from comfy_api_nodes.apis import (
     KlingCameraControl,
     KlingCameraConfig,
@@ -48,23 +48,31 @@ from comfy_api_nodes.apis import (
     KlingCharacterEffectModelName,
     KlingSingleImageEffectModelName,
 )
+from comfy_api_nodes.apis.kling_api import (
+    OmniParamImage,
+    OmniParamVideo,
+    OmniProFirstLastFrameRequest,
+    OmniProReferences2VideoRequest,
+    OmniProText2VideoRequest,
+    TaskStatusVideoResponse,
+)
 from comfy_api_nodes.util import (
-    validate_image_dimensions,
+    ApiEndpoint,
+    download_url_to_image_tensor,
+    download_url_to_video_output,
+    get_number_of_images,
+    poll_op,
+    sync_op,
+    tensor_to_base64_string,
+    upload_audio_to_comfyapi,
+    upload_images_to_comfyapi,
+    upload_video_to_comfyapi,
     validate_image_aspect_ratio,
+    validate_image_dimensions,
+    validate_string,
     validate_video_dimensions,
     validate_video_duration,
-    tensor_to_base64_string,
-    validate_string,
-    upload_audio_to_comfyapi,
-    download_url_to_image_tensor,
-    upload_video_to_comfyapi,
-    download_url_to_video_output,
-    sync_op,
-    ApiEndpoint,
-    poll_op,
 )
-from comfy_api.input_impl import VideoFromFile
-from comfy_api.latest import ComfyExtension, IO, Input
 
 KLING_API_VERSION = "v1"
 PATH_TEXT_TO_VIDEO = f"/proxy/kling/{KLING_API_VERSION}/videos/text2video"
@@ -202,6 +210,20 @@ VOICES_CONFIG = {
 }
 
 
+async def finish_omni_video_task(cls: type[IO.ComfyNode], response: TaskStatusVideoResponse) -> IO.NodeOutput:
+    if response.code:
+        raise RuntimeError(
+            f"Kling request failed. Code: {response.code}, Message: {response.message}, Data: {response.data}"
+        )
+    final_response = await poll_op(
+        cls,
+        ApiEndpoint(path=f"/proxy/kling/v1/videos/omni-video/{response.data.task_id}"),
+        response_model=TaskStatusVideoResponse,
+        status_extractor=lambda r: (r.data.task_status if r.data else None),
+    )
+    return IO.NodeOutput(await download_url_to_video_output(final_response.data.task_result.videos[0].url))
+
+
 def is_valid_camera_control_configs(configs: list[float]) -> bool:
     """Verifies that at least one camera control configuration is non-zero."""
     return any(not math.isclose(value, 0.0) for value in configs)
@@ -449,7 +471,7 @@ async def execute_video_effect(
     image_1: torch.Tensor,
     image_2: torch.Tensor | None = None,
     model_mode: KlingVideoGenMode | None = None,
-) -> tuple[VideoFromFile, str, str]:
+) -> tuple[InputImpl.VideoFromFile, str, str]:
     if dual_character:
         request_input_field = KlingDualCharacterEffectInput(
             model_name=model_name,
@@ -736,6 +758,386 @@ class KlingTextToVideoNode(IO.ComfyNode):
         )
 
 
+class OmniProTextToVideoNode(IO.ComfyNode):
+
+    @classmethod
+    def define_schema(cls) -> IO.Schema:
+        return IO.Schema(
+            node_id="KlingOmniProTextToVideoNode",
+            display_name="Kling Omni Text to Video (Pro)",
+            category="api node/video/Kling",
+            description="Use text prompts to generate videos with the latest Kling model.",
+            inputs=[
+                IO.Combo.Input("model_name", options=["kling-video-o1"]),
+                IO.String.Input(
+                    "prompt",
+                    multiline=True,
+                    tooltip="A text prompt describing the video content. "
+                    "This can include both positive and negative descriptions.",
+                ),
+                IO.Combo.Input("aspect_ratio", options=["16:9", "9:16", "1:1"]),
+                IO.Combo.Input("duration", options=[5, 10]),
+            ],
+            outputs=[
+                IO.Video.Output(),
+            ],
+            hidden=[
+                IO.Hidden.auth_token_comfy_org,
+                IO.Hidden.api_key_comfy_org,
+                IO.Hidden.unique_id,
+            ],
+            is_api_node=True,
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        model_name: str,
+        prompt: str,
+        aspect_ratio: str,
+        duration: int,
+    ) -> IO.NodeOutput:
+        validate_string(prompt, min_length=1, max_length=2500)
+        response = await sync_op(
+            cls,
+            ApiEndpoint(path="/proxy/kling/v1/videos/omni-video", method="POST"),
+            response_model=TaskStatusVideoResponse,
+            data=OmniProText2VideoRequest(
+                model_name=model_name,
+                prompt=prompt,
+                aspect_ratio=aspect_ratio,
+                duration=str(duration),
+            ),
+        )
+        return await finish_omni_video_task(cls, response)
+
+
+class OmniProFirstLastFrameNode(IO.ComfyNode):
+
+    @classmethod
+    def define_schema(cls) -> IO.Schema:
+        return IO.Schema(
+            node_id="KlingOmniProFirstLastFrameNode",
+            display_name="Kling Omni First-Last-Frame to Video (Pro)",
+            category="api node/video/Kling",
+            description="Use a start frame, an optional end frame, or reference images with the latest Kling model.",
+            inputs=[
+                IO.Combo.Input("model_name", options=["kling-video-o1"]),
+                IO.String.Input(
+                    "prompt",
+                    multiline=True,
+                    tooltip="A text prompt describing the video content. "
+                    "This can include both positive and negative descriptions.",
+                ),
+                IO.Combo.Input("duration", options=["5", "10"]),
+                IO.Image.Input("first_frame"),
+                IO.Image.Input(
+                    "end_frame",
+                    optional=True,
+                    tooltip="An optional end frame for the video. "
+                    "This cannot be used simultaneously with 'reference_images'.",
+                ),
+                IO.Image.Input(
+                    "reference_images",
+                    optional=True,
+                    tooltip="Up to 6 additional reference images.",
+                ),
+            ],
+            outputs=[
+                IO.Video.Output(),
+            ],
+            hidden=[
+                IO.Hidden.auth_token_comfy_org,
+                IO.Hidden.api_key_comfy_org,
+                IO.Hidden.unique_id,
+            ],
+            is_api_node=True,
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        model_name: str,
+        prompt: str,
+        duration: int,
+        first_frame: Input.Image,
+        end_frame: Input.Image | None = None,
+        reference_images: Input.Image | None = None,
+    ) -> IO.NodeOutput:
+        validate_string(prompt, min_length=1, max_length=2500)
+        if end_frame is not None and reference_images is not None:
+            raise ValueError("The 'end_frame' input cannot be used simultaneously with 'reference_images'.")
+        validate_image_dimensions(first_frame, min_width=300, min_height=300)
+        validate_image_aspect_ratio(first_frame, (1, 2.5), (2.5, 1))
+        image_list: list[OmniParamImage] = [
+            OmniParamImage(
+                image_url=(await upload_images_to_comfyapi(cls, first_frame, wait_label="Uploading first frame"))[0],
+                type="first_frame",
+            )
+        ]
+        if end_frame is not None:
+            validate_image_dimensions(end_frame, min_width=300, min_height=300)
+            validate_image_aspect_ratio(end_frame, (1, 2.5), (2.5, 1))
+            image_list.append(
+                OmniParamImage(
+                    image_url=(await upload_images_to_comfyapi(cls, end_frame, wait_label="Uploading end frame"))[0],
+                    type="end_frame",
+                )
+            )
+        if reference_images is not None:
+            if get_number_of_images(reference_images) > 6:
+                raise ValueError("The maximum number of reference images allowed is 6.")
+            for i in reference_images:
+                validate_image_dimensions(i, min_width=300, min_height=300)
+                validate_image_aspect_ratio(i, (1, 2.5), (2.5, 1))
+            for i in await upload_images_to_comfyapi(cls, reference_images, wait_label="Uploading reference frame(s)"):
+                image_list.append(OmniParamImage(image_url=i))
+        response = await sync_op(
+            cls,
+            ApiEndpoint(path="/proxy/kling/v1/videos/omni-video", method="POST"),
+            response_model=TaskStatusVideoResponse,
+            data=OmniProFirstLastFrameRequest(
+                model_name=model_name,
+                prompt=prompt,
+                duration=str(duration),
+                image_list=image_list,
+            ),
+        )
+        return await finish_omni_video_task(cls, response)
+
+
+class OmniProImageToVideoNode(IO.ComfyNode):
+
+    @classmethod
+    def define_schema(cls) -> IO.Schema:
+        return IO.Schema(
+            node_id="KlingOmniProImageToVideoNode",
+            display_name="Kling Omni Image to Video (Pro)",
+            category="api node/video/Kling",
+            description="Use up to 7 reference images to generate a video with the latest Kling model.",
+            inputs=[
+                IO.Combo.Input("model_name", options=["kling-video-o1"]),
+                IO.String.Input(
+                    "prompt",
+                    multiline=True,
+                    tooltip="A text prompt describing the video content. "
+                    "This can include both positive and negative descriptions.",
+                ),
+                IO.Combo.Input("aspect_ratio", options=["16:9", "9:16", "1:1"]),
+                IO.Int.Input("duration", default=3, min=3, max=10, display_mode=IO.NumberDisplay.slider),
+                IO.Image.Input(
+                    "reference_images",
+                    tooltip="Up to 7 reference images.",
+                ),
+            ],
+            outputs=[
+                IO.Video.Output(),
+            ],
+            hidden=[
+                IO.Hidden.auth_token_comfy_org,
+                IO.Hidden.api_key_comfy_org,
+                IO.Hidden.unique_id,
+            ],
+            is_api_node=True,
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        model_name: str,
+        prompt: str,
+        aspect_ratio: str,
+        duration: int,
+        reference_images: Input.Image,
+    ) -> IO.NodeOutput:
+        validate_string(prompt, min_length=1, max_length=2500)
+        if get_number_of_images(reference_images) > 7:
+            raise ValueError("The maximum number of reference images is 7.")
+        for i in reference_images:
+            validate_image_dimensions(i, min_width=300, min_height=300)
+            validate_image_aspect_ratio(i, (1, 2.5), (2.5, 1))
+        image_list: list[OmniParamImage] = []
+        for i in await upload_images_to_comfyapi(cls, reference_images, wait_label="Uploading reference image"):
+            image_list.append(OmniParamImage(image_url=i))
+        response = await sync_op(
+            cls,
+            ApiEndpoint(path="/proxy/kling/v1/videos/omni-video", method="POST"),
+            response_model=TaskStatusVideoResponse,
+            data=OmniProReferences2VideoRequest(
+                model_name=model_name,
+                prompt=prompt,
+                aspect_ratio=aspect_ratio,
+                duration=str(duration),
+                image_list=image_list,
+            ),
+        )
+        return await finish_omni_video_task(cls, response)
+
+
+class OmniProVideoToVideoNode(IO.ComfyNode):
+
+    @classmethod
+    def define_schema(cls) -> IO.Schema:
+        return IO.Schema(
+            node_id="KlingOmniProVideoToVideoNode",
+            display_name="Kling Omni Video to Video (Pro)",
+            category="api node/video/Kling",
+            description="Use a video and up to 4 reference images to generate a video with the latest Kling model.",
+            inputs=[
+                IO.Combo.Input("model_name", options=["kling-video-o1"]),
+                IO.String.Input(
+                    "prompt",
+                    multiline=True,
+                    tooltip="A text prompt describing the video content. "
+                    "This can include both positive and negative descriptions.",
+                ),
+                IO.Combo.Input("aspect_ratio", options=["16:9", "9:16", "1:1"]),
+                IO.Int.Input("duration", default=3, min=3, max=10, display_mode=IO.NumberDisplay.slider),
+                IO.Video.Input("reference_video", tooltip="Video to use as a reference."),
+                IO.Boolean.Input("keep_original_sound", default=True),
+                IO.Image.Input(
+                    "reference_images",
+                    tooltip="Up to 4 additional reference images.",
+                    optional=True,
+                ),
+            ],
+            outputs=[
+                IO.Video.Output(),
+            ],
+            hidden=[
+                IO.Hidden.auth_token_comfy_org,
+                IO.Hidden.api_key_comfy_org,
+                IO.Hidden.unique_id,
+            ],
+            is_api_node=True,
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        model_name: str,
+        prompt: str,
+        aspect_ratio: str,
+        duration: int,
+        reference_video: Input.Video,
+        keep_original_sound: bool,
+        reference_images: Input.Image | None = None,
+    ) -> IO.NodeOutput:
+        validate_string(prompt, min_length=1, max_length=2500)
+        validate_video_duration(reference_video, min_duration=3.0, max_duration=10.05)
+        validate_video_dimensions(reference_video, min_width=720, min_height=720, max_width=2160, max_height=2160)
+        image_list: list[OmniParamImage] = []
+        if reference_images is not None:
+            if get_number_of_images(reference_images) > 4:
+                raise ValueError("The maximum number of reference images allowed with a video input is 4.")
+            for i in reference_images:
+                validate_image_dimensions(i, min_width=300, min_height=300)
+                validate_image_aspect_ratio(i, (1, 2.5), (2.5, 1))
+            for i in await upload_images_to_comfyapi(cls, reference_images, wait_label="Uploading reference image"):
+                image_list.append(OmniParamImage(image_url=i))
+        video_list = [
+            OmniParamVideo(
+                video_url=await upload_video_to_comfyapi(cls, reference_video, wait_label="Uploading reference video"),
+                refer_type="feature",
+                keep_original_sound="yes" if keep_original_sound else "no",
+            )
+        ]
+        response = await sync_op(
+            cls,
+            ApiEndpoint(path="/proxy/kling/v1/videos/omni-video", method="POST"),
+            response_model=TaskStatusVideoResponse,
+            data=OmniProReferences2VideoRequest(
+                model_name=model_name,
+                prompt=prompt,
+                aspect_ratio=aspect_ratio,
+                duration=str(duration),
+                image_list=image_list if image_list else None,
+                video_list=video_list,
+            ),
+        )
+        return await finish_omni_video_task(cls, response)
+
+
+class OmniProEditVideoNode(IO.ComfyNode):
+
+    @classmethod
+    def define_schema(cls) -> IO.Schema:
+        return IO.Schema(
+            node_id="KlingOmniProEditVideoNode",
+            display_name="Kling Omni Edit Video (Pro)",
+            category="api node/video/Kling",
+            description="Edit an existing video with the latest model from Kling.",
+            inputs=[
+                IO.Combo.Input("model_name", options=["kling-video-o1"]),
+                IO.String.Input(
+                    "prompt",
+                    multiline=True,
+                    tooltip="A text prompt describing the video content. "
+                    "This can include both positive and negative descriptions.",
+                ),
+                IO.Video.Input("video", tooltip="Video for editing. The output video length will be the same."),
+                IO.Boolean.Input("keep_original_sound", default=True),
+                IO.Image.Input(
+                    "reference_images",
+                    tooltip="Up to 4 additional reference images.",
+                    optional=True,
+                ),
+            ],
+            outputs=[
+                IO.Video.Output(),
+            ],
+            hidden=[
+                IO.Hidden.auth_token_comfy_org,
+                IO.Hidden.api_key_comfy_org,
+                IO.Hidden.unique_id,
+            ],
+            is_api_node=True,
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        model_name: str,
+        prompt: str,
+        video: Input.Video,
+        keep_original_sound: bool,
+        reference_images: Input.Image | None = None,
+    ) -> IO.NodeOutput:
+        validate_string(prompt, min_length=1, max_length=2500)
+        validate_video_duration(video, min_duration=3.0, max_duration=10.05)
+        validate_video_dimensions(video, min_width=720, min_height=720, max_width=2160, max_height=2160)
+        image_list: list[OmniParamImage] = []
+        if reference_images is not None:
+            if get_number_of_images(reference_images) > 4:
+                raise ValueError("The maximum number of reference images allowed with a video input is 4.")
+            for i in reference_images:
+                validate_image_dimensions(i, min_width=300, min_height=300)
+                validate_image_aspect_ratio(i, (1, 2.5), (2.5, 1))
+            for i in await upload_images_to_comfyapi(cls, reference_images, wait_label="Uploading reference image"):
+                image_list.append(OmniParamImage(image_url=i))
+        video_list = [
+            OmniParamVideo(
+                video_url=await upload_video_to_comfyapi(cls, video, wait_label="Uploading base video"),
+                refer_type="base",
+                keep_original_sound="yes" if keep_original_sound else "no",
+            )
+        ]
+        response = await sync_op(
+            cls,
+            ApiEndpoint(path="/proxy/kling/v1/videos/omni-video", method="POST"),
+            response_model=TaskStatusVideoResponse,
+            data=OmniProReferences2VideoRequest(
+                model_name=model_name,
+                prompt=prompt,
+                aspect_ratio=None,
+                duration=None,
+                image_list=image_list if image_list else None,
+                video_list=video_list,
+            ),
+        )
+        return await finish_omni_video_task(cls, response)
+
+
 class KlingCameraControlT2VNode(IO.ComfyNode):
     """
     Kling Text to Video Camera Control Node. This node is a text to video node, but it supports controlling the camera.
@@ -1162,7 +1564,10 @@ class KlingSingleImageVideoEffectNode(IO.ComfyNode):
             category="api node/video/Kling",
             description="Achieve different special effects when generating a video based on the effect_scene.",
             inputs=[
-                IO.Image.Input("image", tooltip=" Reference Image. URL or Base64 encoded string (without data:image prefix). File size cannot exceed 10MB, resolution not less than 300*300px, aspect ratio between 1:2.5 ~ 2.5:1"),
+                IO.Image.Input(
+                    "image",
+                    tooltip=" Reference Image. URL or Base64 encoded string (without data:image prefix). File size cannot exceed 10MB, resolution not less than 300*300px, aspect ratio between 1:2.5 ~ 2.5:1",
+                ),
                 IO.Combo.Input(
                     "effect_scene",
                     options=[i.value for i in KlingSingleImageEffectsScene],
@@ -1525,6 +1930,11 @@ class KlingExtension(ComfyExtension):
             KlingImageGenerationNode,
             KlingSingleImageVideoEffectNode,
             KlingDualCharacterVideoEffectNode,
+            OmniProTextToVideoNode,
+            OmniProFirstLastFrameNode,
+            OmniProImageToVideoNode,
+            OmniProVideoToVideoNode,
+            OmniProEditVideoNode,
         ]
 
 

comfy_api_nodes/util/upload_helpers.py

@@ -103,6 +103,7 @@ async def upload_video_to_comfyapi(
     container: VideoContainer = VideoContainer.MP4,
     codec: VideoCodec = VideoCodec.H264,
     max_duration: Optional[int] = None,
+    wait_label: str | None = "Uploading",
 ) -> str:
     """
     Uploads a single video to ComfyUI API and returns its download URL.
@@ -127,7 +128,7 @@ async def upload_video_to_comfyapi(
     video.save_to(video_bytes_io, format=container, codec=codec)
     video_bytes_io.seek(0)
 
-    return await upload_file_to_comfyapi(cls, video_bytes_io, filename, upload_mime_type)
+    return await upload_file_to_comfyapi(cls, video_bytes_io, filename, upload_mime_type, wait_label)
 
 
 async def upload_file_to_comfyapi(

comfyui_version.py

@@ -1,3 +1,3 @@
 # This file is automatically generated by the build process when version is
 # updated in pyproject.toml.
-__version__ = "0.3.75"
+__version__ = "0.3.76"

main.py

@@ -15,6 +15,7 @@ from comfy_execution.progress import get_progress_state
 from comfy_execution.utils import get_executing_context
 from comfy_api import feature_flags
 
+
 if __name__ == "__main__":
     #NOTE: These do not do anything on core ComfyUI, they are for custom nodes.
     os.environ['HF_HUB_DISABLE_TELEMETRY'] = '1'
@@ -22,6 +23,23 @@ if __name__ == "__main__":
 
 setup_logger(log_level=args.verbose, use_stdout=args.log_stdout)
 
+
+def handle_comfyui_manager_unavailable():
+    if not args.windows_standalone_build:
+        logging.warning(f"\n\nYou appear to be running comfyui-manager from source, this is not recommended. Please install comfyui-manager using the following command:\ncommand:\n\t{sys.executable} -m pip install --pre comfyui_manager\n")
+    args.enable_manager = False
+
+
+if args.enable_manager:
+    if importlib.util.find_spec("comfyui_manager"):
+        import comfyui_manager
+
+        if not comfyui_manager.__file__ or not comfyui_manager.__file__.endswith('__init__.py'):
+            handle_comfyui_manager_unavailable()
+    else:
+        handle_comfyui_manager_unavailable()
+
+
 def apply_custom_paths():
     # extra model paths
     extra_model_paths_config_path = os.path.join(os.path.dirname(os.path.realpath(__file__)), "extra_model_paths.yaml")
@@ -79,6 +97,11 @@ def execute_prestartup_script():
 
         for possible_module in possible_modules:
             module_path = os.path.join(custom_node_path, possible_module)
+
+            if args.enable_manager:
+                if comfyui_manager.should_be_disabled(module_path):
+                    continue
+
             if os.path.isfile(module_path) or module_path.endswith(".disabled") or module_path == "__pycache__":
                 continue
 
@@ -101,6 +124,10 @@ def execute_prestartup_script():
         logging.info("")
 
 apply_custom_paths()
+
+if args.enable_manager:
+    comfyui_manager.prestartup()
+
 execute_prestartup_script()
 
 
@@ -323,6 +350,9 @@ def start_comfyui(asyncio_loop=None):
         asyncio.set_event_loop(asyncio_loop)
     prompt_server = server.PromptServer(asyncio_loop)
 
+    if args.enable_manager and not args.disable_manager_ui:
+        comfyui_manager.start()
+
     hook_breaker_ac10a0.save_functions()
     asyncio_loop.run_until_complete(nodes.init_extra_nodes(
         init_custom_nodes=(not args.disable_all_custom_nodes) or len(args.whitelist_custom_nodes) > 0,

manager_requirements.txt

@@ -0,0 +1 @@
+comfyui_manager==4.0.3b3

nodes.py

@@ -43,6 +43,9 @@ import folder_paths
 import latent_preview
 import node_helpers
 
+if args.enable_manager:
+    import comfyui_manager
+
 def before_node_execution():
     comfy.model_management.throw_exception_if_processing_interrupted()
 
@@ -939,7 +942,7 @@ class CLIPLoader:
     @classmethod
     def INPUT_TYPES(s):
         return {"required": { "clip_name": (folder_paths.get_filename_list("text_encoders"), ),
-                              "type": (["stable_diffusion", "stable_cascade", "sd3", "stable_audio", "mochi", "ltxv", "pixart", "cosmos", "lumina2", "wan", "hidream", "chroma", "ace", "omnigen2", "qwen_image", "hunyuan_image", "flux2"], ),
+                              "type": (["stable_diffusion", "stable_cascade", "sd3", "stable_audio", "mochi", "ltxv", "pixart", "cosmos", "lumina2", "wan", "hidream", "chroma", "ace", "omnigen2", "qwen_image", "hunyuan_image", "flux2", "ovis"], ),
                               },
                 "optional": {
                               "device": (["default", "cpu"], {"advanced": True}),
@@ -2243,6 +2246,12 @@ async def init_external_custom_nodes():
             if args.disable_all_custom_nodes and possible_module not in args.whitelist_custom_nodes:
                 logging.info(f"Skipping {possible_module} due to disable_all_custom_nodes and whitelist_custom_nodes")
                 continue
+
+            if args.enable_manager:
+                if comfyui_manager.should_be_disabled(module_path):
+                    logging.info(f"Blocked by policy: {module_path}")
+                    continue
+
             time_before = time.perf_counter()
             success = await load_custom_node(module_path, base_node_names, module_parent="custom_nodes")
             node_import_times.append((time.perf_counter() - time_before, module_path, success))

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "ComfyUI"
-version = "0.3.75"
+version = "0.3.76"
 readme = "README.md"
 license = { file = "LICENSE" }
 requires-python = ">=3.9"

requirements.txt

@@ -1,4 +1,4 @@
-comfyui-frontend-package==1.32.10
+comfyui-frontend-package==1.33.10
 comfyui-workflow-templates==0.7.25
 comfyui-embedded-docs==0.3.1
 torch

server.py

@@ -44,6 +44,9 @@ from protocol import BinaryEventTypes
 # Import cache control middleware
 from middleware.cache_middleware import cache_control
 
+if args.enable_manager:
+    import comfyui_manager
+
 async def send_socket_catch_exception(function, message):
     try:
         await function(message)
@@ -212,6 +215,9 @@ class PromptServer():
         if args.disable_api_nodes:
             middlewares.append(create_block_external_middleware())
 
+        if args.enable_manager:
+            middlewares.append(comfyui_manager.create_middleware())
+
         max_upload_size = round(args.max_upload_size * 1024 * 1024)
         self.app = web.Application(client_max_size=max_upload_size, middlewares=middlewares)
         self.sockets = dict()
@@ -599,7 +605,7 @@ class PromptServer():
 
             system_stats = {
                 "system": {
-                    "os": os.name,
+                    "os": sys.platform,
                     "ram_total": ram_total,
                     "ram_free": ram_free,
                     "comfyui_version": __version__,