Merge fb8726244c into 3e3ed8cc2a

Co-authored-by: Jedrzej Kosinski <kosinkadink1@gmail.com>

.ci/windows_amd_base_files/run_amd_gpu_enable_dynamic_vram.bat

@@ -1,2 +1,2 @@
-.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build --disable-smart-memory
+.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build --enable-dynamic-vram
 pause

CODEOWNERS

@@ -1,2 +1,2 @@
 # Admins
-* @comfyanonymous @kosinkadink @guill
+* @comfyanonymous @kosinkadink @guill @alexisrolland @rattus128 @kijai

README.md

@@ -193,13 +193,15 @@ If you have trouble extracting it, right click the file -> properties -> unblock
 
 The portable above currently comes with python 3.13 and pytorch cuda 13.0. Update your Nvidia drivers if it doesn't start.
 
-#### Alternative Downloads:
+#### All Official Portable Downloads:
 
 [Portable for AMD GPUs](https://github.com/comfyanonymous/ComfyUI/releases/latest/download/ComfyUI_windows_portable_amd.7z)
 
-[Experimental portable for Intel GPUs](https://github.com/comfyanonymous/ComfyUI/releases/latest/download/ComfyUI_windows_portable_intel.7z)
+[Portable for Intel GPUs](https://github.com/comfyanonymous/ComfyUI/releases/latest/download/ComfyUI_windows_portable_intel.7z)
 
-[Portable with pytorch cuda 12.6 and python 3.12](https://github.com/comfyanonymous/ComfyUI/releases/latest/download/ComfyUI_windows_portable_nvidia_cu126.7z) (Supports Nvidia 10 series and older GPUs).
+[Portable for Nvidia GPUs](https://github.com/comfyanonymous/ComfyUI/releases/latest/download/ComfyUI_windows_portable_nvidia.7z) (supports 20 series and above).
+
+[Portable for Nvidia GPUs with pytorch cuda 12.6 and python 3.12](https://github.com/comfyanonymous/ComfyUI/releases/latest/download/ComfyUI_windows_portable_nvidia_cu126.7z) (Supports Nvidia 10 series and older GPUs).
 
 #### How do I share models between another UI and ComfyUI?
 

comfy/cli_args.py

@@ -90,7 +90,6 @@ parser.add_argument("--force-channels-last", action="store_true", help="Force ch
 parser.add_argument("--directml", type=int, nargs="?", metavar="DIRECTML_DEVICE", const=-1, help="Use torch-directml.")
 
 parser.add_argument("--oneapi-device-selector", type=str, default=None, metavar="SELECTOR_STRING", help="Sets the oneAPI device(s) this instance will use.")
-parser.add_argument("--disable-ipex-optimize", action="store_true", help="Disables ipex.optimize default when loading models with Intel's Extension for Pytorch.")
 parser.add_argument("--supports-fp8-compute", action="store_true", help="ComfyUI will act like if the device supports fp8 compute.")
 
 class LatentPreviewMethod(enum.Enum):

comfy/latent_formats.py

@@ -224,6 +224,7 @@ class Flux2(LatentFormat):
 
         self.latent_rgb_factors_bias = [-0.0329, -0.0718, -0.0851]
         self.latent_rgb_factors_reshape = lambda t: t.reshape(t.shape[0], 32, 2, 2, t.shape[-2], t.shape[-1]).permute(0, 1, 4, 2, 5, 3).reshape(t.shape[0], 32, t.shape[-2] * 2, t.shape[-1] * 2)
+        self.taesd_decoder_name = "taef2_decoder"
 
     def process_in(self, latent):
         return latent

comfy/lora.py

@@ -342,6 +342,12 @@ def model_lora_keys_unet(model, key_map={}):
                 key_map["base_model.model.{}".format(key_lora)] = k  # Official base model loras
                 key_map["lycoris_{}".format(key_lora.replace(".", "_"))] = k  # LyCORIS/LoKR format
 
+    if isinstance(model, comfy.model_base.ErnieImage):
+        for k in sdk:
+            if k.startswith("diffusion_model.") and k.endswith(".weight"):
+                key_lora = k[len("diffusion_model."):-len(".weight")]
+                key_map["transformer.{}".format(key_lora)] = k
+
     return key_map
 
 

comfy/model_management.py

@@ -112,10 +112,6 @@ if args.directml is not None:
     # torch_directml.disable_tiled_resources(True)
     lowvram_available = False #TODO: need to find a way to get free memory in directml before this can be enabled by default.
 
-try:
-    import intel_extension_for_pytorch as ipex  # noqa: F401
-except:
-    pass
 
 try:
     _ = torch.xpu.device_count()
@@ -583,9 +579,6 @@ class LoadedModel:
 
         real_model = self.model.model
 
-        if is_intel_xpu() and not args.disable_ipex_optimize and 'ipex' in globals() and real_model is not None:
-            with torch.no_grad():
-                real_model = ipex.optimize(real_model.eval(), inplace=True, graph_mode=True, concat_linear=True)
 
         self.real_model = weakref.ref(real_model)
         self.model_finalizer = weakref.finalize(real_model, cleanup_models)
@@ -1581,10 +1574,7 @@ def should_use_fp16(device=None, model_params=0, prioritize_performance=True, ma
         return False
 
     if is_intel_xpu():
-        if torch_version_numeric < (2, 3):
-            return True
-        else:
-            return torch.xpu.get_device_properties(device).has_fp16
+        return torch.xpu.get_device_properties(device).has_fp16
 
     if is_ascend_npu():
         return True
@@ -1650,10 +1640,7 @@ def should_use_bf16(device=None, model_params=0, prioritize_performance=True, ma
         return False
 
     if is_intel_xpu():
-        if torch_version_numeric < (2, 3):
-            return True
-        else:
-            return torch.xpu.is_bf16_supported()
+        return torch.xpu.is_bf16_supported()
 
     if is_ascend_npu():
         return True
@@ -1784,6 +1771,7 @@ def soft_empty_cache(force=False):
     if cpu_state == CPUState.MPS:
         torch.mps.empty_cache()
     elif is_intel_xpu():
+        torch.xpu.synchronize()
         torch.xpu.empty_cache()
     elif is_ascend_npu():
         torch.npu.empty_cache()

comfy/sd.py

@@ -479,7 +479,10 @@ class VAE:
                                                             encoder_config={'target': "comfy.ldm.modules.diffusionmodules.model.Encoder", 'params': encoder_config},
                                                             decoder_config={'target': "comfy.ldm.modules.temporal_ae.VideoDecoder", 'params': decoder_config})
             elif "taesd_decoder.1.weight" in sd:
-                self.latent_channels = sd["taesd_decoder.1.weight"].shape[1]
+                if isinstance(metadata, dict) and "tae_latent_channels" in metadata:
+                    self.latent_channels = metadata["tae_latent_channels"]
+                else:
+                    self.latent_channels = sd["taesd_decoder.1.weight"].shape[1]
                 self.first_stage_model = comfy.taesd.taesd.TAESD(latent_channels=self.latent_channels)
             elif "vquantizer.codebook.weight" in sd: #VQGan: stage a of stable cascade
                 self.first_stage_model = StageA()

comfy/supported_models.py

@@ -1879,6 +1879,86 @@ class CogVideoX_I2V(CogVideoX_T2V):
         out = model_base.CogVideoX(self, image_to_video=True, device=device)
         return out
 
-models = [LotusD, Stable_Zero123, SD15_instructpix2pix, SD15, SD20, SD21UnclipL, SD21UnclipH, SDXL_instructpix2pix, SDXLRefiner, SDXL, SSD1B, KOALA_700M, KOALA_1B, Segmind_Vega, SD_X4Upscaler, Stable_Cascade_C, Stable_Cascade_B, SV3D_u, SV3D_p, SD3, StableAudio, AuraFlow, PixArtAlpha, PixArtSigma, HunyuanDiT, HunyuanDiT1, FluxInpaint, Flux, LongCatImage, FluxSchnell, GenmoMochi, LTXV, LTXAV, HunyuanVideo15_SR_Distilled, HunyuanVideo15, HunyuanImage21Refiner, HunyuanImage21, HunyuanVideoSkyreelsI2V, HunyuanVideoI2V, HunyuanVideo, CosmosT2V, CosmosI2V, CosmosT2IPredict2, CosmosI2VPredict2, ZImagePixelSpace, ZImage, Lumina2, WAN22_T2V, WAN21_T2V, WAN21_I2V, WAN21_FunControl2V, WAN21_Vace, WAN21_Camera, WAN22_Camera, WAN22_S2V, WAN21_HuMo, WAN22_Animate, WAN21_FlowRVS, WAN21_SCAIL, Hunyuan3Dv2mini, Hunyuan3Dv2, Hunyuan3Dv2_1, HiDream, Chroma, ChromaRadiance, ACEStep, ACEStep15, Omnigen2, QwenImage, Flux2, Kandinsky5Image, Kandinsky5, Anima, RT_DETR_v4, ErnieImage, SAM3, SAM31, CogVideoX_I2V, CogVideoX_T2V]
 
-models += [SVD_img2vid]
+models = [
+    LotusD,
+    Stable_Zero123,
+    SD15_instructpix2pix,
+    SD15,
+    SD20,
+    SD21UnclipL,
+    SD21UnclipH,
+    SDXL_instructpix2pix,
+    SDXLRefiner,
+    SDXL,
+    SSD1B,
+    KOALA_700M,
+    KOALA_1B,
+    Segmind_Vega,
+    SD_X4Upscaler,
+    Stable_Cascade_C,
+    Stable_Cascade_B,
+    SV3D_u,
+    SV3D_p,
+    SD3,
+    StableAudio,
+    AuraFlow,
+    PixArtAlpha,
+    PixArtSigma,
+    HunyuanDiT,
+    HunyuanDiT1,
+    FluxInpaint,
+    Flux,
+    LongCatImage,
+    FluxSchnell,
+    GenmoMochi,
+    LTXV,
+    LTXAV,
+    HunyuanVideo15_SR_Distilled,
+    HunyuanVideo15,
+    HunyuanImage21Refiner,
+    HunyuanImage21,
+    HunyuanVideoSkyreelsI2V,
+    HunyuanVideoI2V,
+    HunyuanVideo,
+    CosmosT2V,
+    CosmosI2V,
+    CosmosT2IPredict2,
+    CosmosI2VPredict2,
+    ZImagePixelSpace,
+    ZImage,
+    Lumina2,
+    WAN22_T2V,
+    WAN21_T2V,
+    WAN21_I2V,
+    WAN21_FunControl2V,
+    WAN21_Vace,
+    WAN21_Camera,
+    WAN22_Camera,
+    WAN22_S2V,
+    WAN21_HuMo,
+    WAN22_Animate,
+    WAN21_FlowRVS,
+    WAN21_SCAIL,
+    Hunyuan3Dv2mini,
+    Hunyuan3Dv2,
+    Hunyuan3Dv2_1,
+    HiDream,
+    Chroma,
+    ChromaRadiance,
+    ACEStep,
+    ACEStep15,
+    Omnigen2,
+    QwenImage,
+    Flux2,
+    Kandinsky5Image,
+    Kandinsky5,
+    Anima,
+    RT_DETR_v4,
+    ErnieImage,
+    SAM3,
+    SAM31,
+    CogVideoX_I2V,
+    CogVideoX_T2V,
+    SVD_img2vid,
+]

comfy/taesd/taesd.py

@@ -17,32 +17,79 @@ class Clamp(nn.Module):
         return torch.tanh(x / 3) * 3
 
 class Block(nn.Module):
-    def __init__(self, n_in, n_out):
+    def __init__(self, n_in: int, n_out: int, use_midblock_gn: bool = False):
         super().__init__()
         self.conv = nn.Sequential(conv(n_in, n_out), nn.ReLU(), conv(n_out, n_out), nn.ReLU(), conv(n_out, n_out))
         self.skip = comfy.ops.disable_weight_init.Conv2d(n_in, n_out, 1, bias=False) if n_in != n_out else nn.Identity()
         self.fuse = nn.ReLU()
-    def forward(self, x):
+        if not use_midblock_gn:
+            self.pool = None
+            return
+        n_gn = n_in * 4
+        self.pool = nn.Sequential(
+            comfy.ops.disable_weight_init.Conv2d(n_in, n_gn, 1, bias=False),
+            comfy.ops.disable_weight_init.GroupNorm(4, n_gn),
+            nn.ReLU(inplace=True),
+            comfy.ops.disable_weight_init.Conv2d(n_gn, n_in, 1, bias=False),
+        )
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        if self.pool is not None:
+            x = x + self.pool(x)
         return self.fuse(self.conv(x) + self.skip(x))
 
-def Encoder(latent_channels=4):
-    return nn.Sequential(
-        conv(3, 64), Block(64, 64),
-        conv(64, 64, stride=2, bias=False), Block(64, 64), Block(64, 64), Block(64, 64),
-        conv(64, 64, stride=2, bias=False), Block(64, 64), Block(64, 64), Block(64, 64),
-        conv(64, 64, stride=2, bias=False), Block(64, 64), Block(64, 64), Block(64, 64),
-        conv(64, latent_channels),
-    )
+class Encoder(nn.Sequential):
+    def __init__(self, latent_channels: int = 4, use_gn: bool = False):
+        super().__init__(
+            conv(3, 64), Block(64, 64),
+            conv(64, 64, stride=2, bias=False), Block(64, 64), Block(64, 64), Block(64, 64),
+            conv(64, 64, stride=2, bias=False), Block(64, 64), Block(64, 64), Block(64, 64),
+            conv(64, 64, stride=2, bias=False), Block(64, 64, use_gn), Block(64, 64, use_gn), Block(64, 64, use_gn),
+            conv(64, latent_channels),
+        )
 
+class Decoder(nn.Sequential):
+    def __init__(self, latent_channels: int = 4, use_gn: bool = False):
+        super().__init__(
+            Clamp(), conv(latent_channels, 64), nn.ReLU(),
+            Block(64, 64, use_gn), Block(64, 64, use_gn), Block(64, 64, use_gn), nn.Upsample(scale_factor=2), conv(64, 64, bias=False),
+            Block(64, 64), Block(64, 64), Block(64, 64), nn.Upsample(scale_factor=2), conv(64, 64, bias=False),
+            Block(64, 64), Block(64, 64), Block(64, 64), nn.Upsample(scale_factor=2), conv(64, 64, bias=False),
+            Block(64, 64), conv(64, 3),
+        )
+
+class DecoderFlux2(Decoder):
+    def __init__(self, latent_channels: int = 128, use_gn: bool = True):
+        if latent_channels != 128 or not use_gn:
+            raise ValueError("Unexpected parameters for Flux2 TAE module")
+        super().__init__(latent_channels=32, use_gn=True)
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        B, C, H, W = x.shape
+        x = (
+            x
+            .reshape(B, 32, 2, 2, H, W)
+            .permute(0, 1, 4, 2, 5, 3)
+            .reshape(B, 32, H * 2, W * 2)
+        )
+        return super().forward(x)
+
+class EncoderFlux2(Encoder):
+    def __init__(self, latent_channels: int = 128, use_gn: bool = True):
+        if latent_channels != 128 or not use_gn:
+            raise ValueError("Unexpected parameters for Flux2 TAE module")
+        super().__init__(latent_channels=32, use_gn=True)
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        result = super().forward(x)
+        B, C, H, W = result.shape
+        return (
+            result
+            .reshape(B, C, H // 2, 2, W // 2, 2)
+            .permute(0, 1, 3, 5, 2, 4)
+            .reshape(B, 128, H // 2, W // 2)
+        )
 
-def Decoder(latent_channels=4):
-    return nn.Sequential(
-        Clamp(), conv(latent_channels, 64), nn.ReLU(),
-        Block(64, 64), Block(64, 64), Block(64, 64), nn.Upsample(scale_factor=2), conv(64, 64, bias=False),
-        Block(64, 64), Block(64, 64), Block(64, 64), nn.Upsample(scale_factor=2), conv(64, 64, bias=False),
-        Block(64, 64), Block(64, 64), Block(64, 64), nn.Upsample(scale_factor=2), conv(64, 64, bias=False),
-        Block(64, 64), conv(64, 3),
-    )
 
 class TAESD(nn.Module):
     latent_magnitude = 3
@@ -51,8 +98,15 @@ class TAESD(nn.Module):
     def __init__(self, encoder_path=None, decoder_path=None, latent_channels=4):
         """Initialize pretrained TAESD on the given device from the given checkpoints."""
         super().__init__()
-        self.taesd_encoder = Encoder(latent_channels=latent_channels)
-        self.taesd_decoder = Decoder(latent_channels=latent_channels)
+        if latent_channels == 128:
+            encoder_class = EncoderFlux2
+            decoder_class = DecoderFlux2
+        else:
+            encoder_class = Encoder
+            decoder_class = Decoder
+        self.taesd_encoder = encoder_class(latent_channels=latent_channels)
+        self.taesd_decoder = decoder_class(latent_channels=latent_channels)
+
         self.vae_scale = torch.nn.Parameter(torch.tensor(1.0))
         self.vae_shift = torch.nn.Parameter(torch.tensor(0.0))
         if encoder_path is not None:
@@ -61,19 +115,19 @@ class TAESD(nn.Module):
             self.taesd_decoder.load_state_dict(comfy.utils.load_torch_file(decoder_path, safe_load=True))
 
     @staticmethod
-    def scale_latents(x):
+    def scale_latents(x: torch.Tensor) -> torch.Tensor:
         """raw latents -> [0, 1]"""
         return x.div(2 * TAESD.latent_magnitude).add(TAESD.latent_shift).clamp(0, 1)
 
     @staticmethod
-    def unscale_latents(x):
+    def unscale_latents(x: torch.Tensor) -> torch.Tensor:
         """[0, 1] -> raw latents"""
         return x.sub(TAESD.latent_shift).mul(2 * TAESD.latent_magnitude)
 
-    def decode(self, x):
+    def decode(self, x: torch.Tensor) -> torch.Tensor:
         x_sample = self.taesd_decoder((x - self.vae_shift) * self.vae_scale)
         x_sample = x_sample.sub(0.5).mul(2)
         return x_sample
 
-    def encode(self, x):
+    def encode(self, x: torch.Tensor) -> torch.Tensor:
         return (self.taesd_encoder(x * 0.5 + 0.5) / self.vae_scale) + self.vae_shift

comfy_api/latest/_input_impl/video_types.py

@@ -251,6 +251,7 @@ class VideoFromFile(VideoInput):
             container.seek(start_pts, stream=video_stream)
 
         image_format = 'gbrpf32le'
+        process_image_format = lambda a: a
         audio = None
 
         streams = [video_stream]
@@ -283,11 +284,25 @@ class VideoFromFile(VideoInput):
                             break
 
                         if not checked_alpha:
+                            alpha_channel = False
                             for comp in frame.format.components:
                                 if comp.is_alpha or frame.format.name == "pal8":
                                     alphas = []
-                                    image_format = 'gbrapf32le'
+                                    alpha_channel = True
                                     break
+                            if frame.format.name in ("yuvj420p", "yuvj422p", "yuvj444p", "rgb24", "rgba", "pal8"):
+                                process_image_format = lambda a: a.float() / 255.0
+                                if alpha_channel:
+                                    image_format = 'rgba'
+                                else:
+                                    image_format = 'rgb24'
+                            else:
+                                process_image_format = lambda a: a
+                                if alpha_channel:
+                                    image_format = 'gbrapf32le'
+                                else:
+                                    image_format = 'gbrpf32le'
+
                             checked_alpha = True
 
                         img = frame.to_ndarray(format=image_format)  # shape: (H, W, 4)
@@ -323,9 +338,9 @@ class VideoFromFile(VideoInput):
                     else:
                         audio_frames.append(frame.to_ndarray())
 
-        images = torch.stack(frames) if len(frames) > 0 else torch.zeros(0, 0, 0, 3)
+        images = process_image_format(torch.stack(frames)) if len(frames) > 0 else torch.zeros(0, 0, 0, 3)
         if alphas is not None:
-            alphas = torch.stack(alphas) if len(alphas) > 0 else torch.zeros(0, 0, 0, 1)
+            alphas = process_image_format(torch.stack(alphas)) if len(alphas) > 0 else torch.zeros(0, 0, 0, 1)
 
         # Get frame rate
         frame_rate = Fraction(video_stream.average_rate) if video_stream.average_rate else Fraction(1)

comfy_api_nodes/apis/bytedance.py

@@ -157,6 +157,11 @@ class SeedanceCreateAssetResponse(BaseModel):
     asset_id: str = Field(...)
 
 
+class SeedanceVirtualLibraryCreateAssetRequest(BaseModel):
+    url: str = Field(..., description="Publicly accessible URL of the image asset to upload.")
+    hash: str = Field(..., description="Dedup key. Re-submitting the same hash returns the existing asset id.")
+
+
 # Dollars per 1K tokens, keyed by (model_id, has_video_input).
 SEEDANCE2_PRICE_PER_1K_TOKENS = {
     ("dreamina-seedance-2-0-260128", False): 0.007,

comfy_api_nodes/apis/moonvalley.py

@@ -1,152 +0,0 @@
-from enum import Enum
-from typing import Optional, Dict, Any
-
-from pydantic import BaseModel, Field, StrictBytes
-
-
-class MoonvalleyPromptResponse(BaseModel):
-    error: Optional[Dict[str, Any]] = None
-    frame_conditioning: Optional[Dict[str, Any]] = None
-    id: Optional[str] = None
-    inference_params: Optional[Dict[str, Any]] = None
-    meta: Optional[Dict[str, Any]] = None
-    model_params: Optional[Dict[str, Any]] = None
-    output_url: Optional[str] = None
-    prompt_text: Optional[str] = None
-    status: Optional[str] = None
-
-
-class MoonvalleyTextToVideoInferenceParams(BaseModel):
-    add_quality_guidance: Optional[bool] = Field(
-        True, description='Whether to add quality guidance'
-    )
-    caching_coefficient: Optional[float] = Field(
-        0.3, description='Caching coefficient for optimization'
-    )
-    caching_cooldown: Optional[int] = Field(
-        3, description='Number of caching cooldown steps'
-    )
-    caching_warmup: Optional[int] = Field(
-        3, description='Number of caching warmup steps'
-    )
-    clip_value: Optional[float] = Field(
-        3, description='CLIP value for generation control'
-    )
-    conditioning_frame_index: Optional[int] = Field(
-        0, description='Index of the conditioning frame'
-    )
-    cooldown_steps: Optional[int] = Field(
-        75, description='Number of cooldown steps (calculated based on num_frames)'
-    )
-    fps: Optional[int] = Field(
-        24, description='Frames per second of the generated video'
-    )
-    guidance_scale: Optional[float] = Field(
-        10, description='Guidance scale for generation control'
-    )
-    height: Optional[int] = Field(
-        1080, description='Height of the generated video in pixels'
-    )
-    negative_prompt: Optional[str] = Field(None, description='Negative prompt text')
-    num_frames: Optional[int] = Field(64, description='Number of frames to generate')
-    seed: Optional[int] = Field(
-        None, description='Random seed for generation (default: random)'
-    )
-    shift_value: Optional[float] = Field(
-        3, description='Shift value for generation control'
-    )
-    steps: Optional[int] = Field(80, description='Number of denoising steps')
-    use_guidance_schedule: Optional[bool] = Field(
-        True, description='Whether to use guidance scheduling'
-    )
-    use_negative_prompts: Optional[bool] = Field(
-        False, description='Whether to use negative prompts'
-    )
-    use_timestep_transform: Optional[bool] = Field(
-        True, description='Whether to use timestep transformation'
-    )
-    warmup_steps: Optional[int] = Field(
-        0, description='Number of warmup steps (calculated based on num_frames)'
-    )
-    width: Optional[int] = Field(
-        1920, description='Width of the generated video in pixels'
-    )
-
-
-class MoonvalleyTextToVideoRequest(BaseModel):
-    image_url: Optional[str] = None
-    inference_params: Optional[MoonvalleyTextToVideoInferenceParams] = None
-    prompt_text: Optional[str] = None
-    webhook_url: Optional[str] = None
-
-
-class MoonvalleyUploadFileRequest(BaseModel):
-    file: Optional[StrictBytes] = None
-
-
-class MoonvalleyUploadFileResponse(BaseModel):
-    access_url: Optional[str] = None
-
-
-class MoonvalleyVideoToVideoInferenceParams(BaseModel):
-    add_quality_guidance: Optional[bool] = Field(
-        True, description='Whether to add quality guidance'
-    )
-    caching_coefficient: Optional[float] = Field(
-        0.3, description='Caching coefficient for optimization'
-    )
-    caching_cooldown: Optional[int] = Field(
-        3, description='Number of caching cooldown steps'
-    )
-    caching_warmup: Optional[int] = Field(
-        3, description='Number of caching warmup steps'
-    )
-    clip_value: Optional[float] = Field(
-        3, description='CLIP value for generation control'
-    )
-    conditioning_frame_index: Optional[int] = Field(
-        0, description='Index of the conditioning frame'
-    )
-    cooldown_steps: Optional[int] = Field(
-        36, description='Number of cooldown steps (calculated based on num_frames)'
-    )
-    guidance_scale: Optional[float] = Field(
-        15, description='Guidance scale for generation control'
-    )
-    negative_prompt: Optional[str] = Field(None, description='Negative prompt text')
-    seed: Optional[int] = Field(
-        None, description='Random seed for generation (default: random)'
-    )
-    shift_value: Optional[float] = Field(
-        3, description='Shift value for generation control'
-    )
-    steps: Optional[int] = Field(80, description='Number of denoising steps')
-    use_guidance_schedule: Optional[bool] = Field(
-        True, description='Whether to use guidance scheduling'
-    )
-    use_negative_prompts: Optional[bool] = Field(
-        False, description='Whether to use negative prompts'
-    )
-    use_timestep_transform: Optional[bool] = Field(
-        True, description='Whether to use timestep transformation'
-    )
-    warmup_steps: Optional[int] = Field(
-        24, description='Number of warmup steps (calculated based on num_frames)'
-    )
-
-
-class ControlType(str, Enum):
-    motion_control = 'motion_control'
-    pose_control = 'pose_control'
-
-
-class MoonvalleyVideoToVideoRequest(BaseModel):
-    control_type: ControlType = Field(
-        ..., description='Supported types for video control'
-    )
-    inference_params: Optional[MoonvalleyVideoToVideoInferenceParams] = None
-    prompt_text: str = Field(..., description='Describes the video to generate')
-    video_url: str = Field(..., description='Url to control video')
-    webhook_url: Optional[str] = Field(
-        None, description='Optional webhook URL for notifications'
-    )

comfy_api_nodes/nodes_bytedance.py

@@ -1,3 +1,4 @@
+import hashlib
 import logging
 import math
 import re
@@ -20,6 +21,7 @@ from comfy_api_nodes.apis.bytedance import (
     SeedanceCreateAssetResponse,
     SeedanceCreateVisualValidateSessionResponse,
     SeedanceGetVisualValidateSessionResponse,
+    SeedanceVirtualLibraryCreateAssetRequest,
     Seedream4Options,
     Seedream4TaskCreationRequest,
     TaskAudioContent,
@@ -271,6 +273,30 @@ async def _wait_for_asset_active(cls: type[IO.ComfyNode], asset_id: str, group_i
     )
 
 
+async def _seedance_virtual_library_upload_image_asset(
+    cls: type[IO.ComfyNode],
+    image: torch.Tensor,
+    *,
+    wait_label: str = "Uploading image",
+) -> str:
+    """Upload an image into the caller's per-customer Seedance virtual library."""
+    public_url = await upload_image_to_comfyapi(cls, image, wait_label=wait_label)
+    normalized = image.detach().cpu().contiguous().to(torch.float32)
+    digest = hashlib.sha256()
+    digest.update(str(tuple(normalized.shape)).encode("utf-8"))
+    digest.update(b"\0")
+    digest.update(normalized.numpy().tobytes())
+    image_hash = digest.hexdigest()
+    create_resp = await sync_op(
+        cls,
+        ApiEndpoint(path="/proxy/seedance/virtual-library/assets", method="POST"),
+        response_model=SeedanceCreateAssetResponse,
+        data=SeedanceVirtualLibraryCreateAssetRequest(url=public_url, hash=image_hash),
+    )
+    await _wait_for_asset_active(cls, create_resp.asset_id, group_id="virtual-library")
+    return f"asset://{create_resp.asset_id}"
+
+
 def _seedance2_price_extractor(model_id: str, has_video_input: bool):
     """Returns a price_extractor closure for Seedance 2.0 poll_op."""
     rate = SEEDANCE2_PRICE_PER_1K_TOKENS.get((model_id, has_video_input))
@@ -1377,7 +1403,6 @@ class ByteDance2TextToVideoNode(IO.ComfyNode):
             status_extractor=lambda r: r.status,
             price_extractor=_seedance2_price_extractor(model_id, has_video_input=False),
             poll_interval=9,
-            max_poll_attempts=180,
         )
         return IO.NodeOutput(await download_url_to_video_output(response.content.video_url))
 
@@ -1507,7 +1532,9 @@ class ByteDance2FirstLastFrameNode(IO.ComfyNode):
         if first_frame_asset_id:
             first_frame_url = image_assets[first_frame_asset_id]
         else:
-            first_frame_url = await upload_image_to_comfyapi(cls, first_frame, wait_label="Uploading first frame.")
+            first_frame_url = await _seedance_virtual_library_upload_image_asset(
+                cls, first_frame, wait_label="Uploading first frame."
+            )
 
         content: list[TaskTextContent | TaskImageContent] = [
             TaskTextContent(text=model["prompt"]),
@@ -1527,7 +1554,9 @@ class ByteDance2FirstLastFrameNode(IO.ComfyNode):
             content.append(
                 TaskImageContent(
                     image_url=TaskImageContentUrl(
-                        url=await upload_image_to_comfyapi(cls, last_frame, wait_label="Uploading last frame.")
+                        url=await _seedance_virtual_library_upload_image_asset(
+                            cls, last_frame, wait_label="Uploading last frame."
+                        )
                     ),
                     role="last_frame",
                 ),
@@ -1555,7 +1584,6 @@ class ByteDance2FirstLastFrameNode(IO.ComfyNode):
             status_extractor=lambda r: r.status,
             price_extractor=_seedance2_price_extractor(model_id, has_video_input=False),
             poll_interval=9,
-            max_poll_attempts=180,
         )
         return IO.NodeOutput(await download_url_to_video_output(response.content.video_url))
 
@@ -1805,9 +1833,9 @@ class ByteDance2ReferenceNode(IO.ComfyNode):
             content.append(
                 TaskImageContent(
                     image_url=TaskImageContentUrl(
-                        url=await upload_image_to_comfyapi(
+                        url=await _seedance_virtual_library_upload_image_asset(
                             cls,
-                            image=reference_images[key],
+                            reference_images[key],
                             wait_label=f"Uploading image {i}",
                         ),
                     ),
@@ -1877,7 +1905,6 @@ class ByteDance2ReferenceNode(IO.ComfyNode):
             status_extractor=lambda r: r.status,
             price_extractor=_seedance2_price_extractor(model_id, has_video_input=has_video_input),
             poll_interval=9,
-            max_poll_attempts=180,
         )
         return IO.NodeOutput(await download_url_to_video_output(response.content.video_url))
 

comfy_api_nodes/nodes_hitpaw.py

@@ -178,7 +178,6 @@ class HitPawGeneralImageEnhance(IO.ComfyNode):
             status_extractor=lambda x: x.data.status,
             price_extractor=lambda x: request_price,
             poll_interval=10.0,
-            max_poll_attempts=480,
         )
         return IO.NodeOutput(await download_url_to_image_tensor(final_response.data.res_url))
 
@@ -324,7 +323,6 @@ class HitPawVideoEnhance(IO.ComfyNode):
             status_extractor=lambda x: x.data.status,
             price_extractor=lambda x: request_price,
             poll_interval=10.0,
-            max_poll_attempts=320,
         )
         return IO.NodeOutput(await download_url_to_video_output(final_response.data.res_url))
 

comfy_api_nodes/nodes_kling.py

@@ -276,7 +276,6 @@ async def finish_omni_video_task(cls: type[IO.ComfyNode], response: TaskStatusRe
         cls,
         ApiEndpoint(path=f"/proxy/kling/v1/videos/omni-video/{response.data.task_id}"),
         response_model=TaskStatusResponse,
-        max_poll_attempts=280,
         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))
@@ -3062,7 +3061,6 @@ class KlingVideoNode(IO.ComfyNode):
             cls,
             ApiEndpoint(path=poll_path),
             response_model=TaskStatusResponse,
-            max_poll_attempts=280,
             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))
@@ -3188,7 +3186,6 @@ class KlingFirstLastFrameNode(IO.ComfyNode):
             cls,
             ApiEndpoint(path=f"/proxy/kling/v1/videos/image2video/{response.data.task_id}"),
             response_model=TaskStatusResponse,
-            max_poll_attempts=280,
             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))

comfy_api_nodes/nodes_magnific.py

@@ -230,7 +230,6 @@ class MagnificImageUpscalerCreativeNode(IO.ComfyNode):
             status_extractor=lambda x: x.status,
             price_extractor=lambda _: price_usd,
             poll_interval=10.0,
-            max_poll_attempts=480,
         )
         return IO.NodeOutput(await download_url_to_image_tensor(final_response.generated[0]))
 
@@ -391,7 +390,6 @@ class MagnificImageUpscalerPreciseV2Node(IO.ComfyNode):
             status_extractor=lambda x: x.status,
             price_extractor=lambda _: price_usd,
             poll_interval=10.0,
-            max_poll_attempts=480,
         )
         return IO.NodeOutput(await download_url_to_image_tensor(final_response.generated[0]))
 
@@ -541,7 +539,6 @@ class MagnificImageStyleTransferNode(IO.ComfyNode):
             response_model=TaskResponse,
             status_extractor=lambda x: x.status,
             poll_interval=10.0,
-            max_poll_attempts=480,
         )
         return IO.NodeOutput(await download_url_to_image_tensor(final_response.generated[0]))
 
@@ -782,7 +779,6 @@ class MagnificImageRelightNode(IO.ComfyNode):
             response_model=TaskResponse,
             status_extractor=lambda x: x.status,
             poll_interval=10.0,
-            max_poll_attempts=480,
         )
         return IO.NodeOutput(await download_url_to_image_tensor(final_response.generated[0]))
 
@@ -924,7 +920,6 @@ class MagnificImageSkinEnhancerNode(IO.ComfyNode):
             response_model=TaskResponse,
             status_extractor=lambda x: x.status,
             poll_interval=10.0,
-            max_poll_attempts=480,
         )
         return IO.NodeOutput(await download_url_to_image_tensor(final_response.generated[0]))
 

comfy_api_nodes/nodes_moonvalley.py

@@ -1,534 +0,0 @@
-import logging
-
-from typing_extensions import override
-
-from comfy_api.latest import IO, ComfyExtension, Input
-from comfy_api_nodes.apis.moonvalley import (
-    MoonvalleyPromptResponse,
-    MoonvalleyTextToVideoInferenceParams,
-    MoonvalleyTextToVideoRequest,
-    MoonvalleyVideoToVideoInferenceParams,
-    MoonvalleyVideoToVideoRequest,
-)
-from comfy_api_nodes.util import (
-    ApiEndpoint,
-    download_url_to_video_output,
-    poll_op,
-    sync_op,
-    trim_video,
-    upload_images_to_comfyapi,
-    upload_video_to_comfyapi,
-    validate_container_format_is_mp4,
-    validate_image_dimensions,
-    validate_string,
-)
-
-API_UPLOADS_ENDPOINT = "/proxy/moonvalley/uploads"
-API_PROMPTS_ENDPOINT = "/proxy/moonvalley/prompts"
-API_VIDEO2VIDEO_ENDPOINT = "/proxy/moonvalley/prompts/video-to-video"
-API_TXT2VIDEO_ENDPOINT = "/proxy/moonvalley/prompts/text-to-video"
-API_IMG2VIDEO_ENDPOINT = "/proxy/moonvalley/prompts/image-to-video"
-
-MIN_WIDTH = 300
-MIN_HEIGHT = 300
-
-MAX_WIDTH = 10000
-MAX_HEIGHT = 10000
-
-MIN_VID_WIDTH = 300
-MIN_VID_HEIGHT = 300
-
-MAX_VID_WIDTH = 10000
-MAX_VID_HEIGHT = 10000
-
-MAX_VIDEO_SIZE = 1024 * 1024 * 1024  # 1 GB max for in-memory video processing
-
-MOONVALLEY_MAREY_MAX_PROMPT_LENGTH = 5000
-
-
-def is_valid_task_creation_response(response: MoonvalleyPromptResponse) -> bool:
-    """Verifies that the initial response contains a task ID."""
-    return bool(response.id)
-
-
-def validate_task_creation_response(response) -> None:
-    if not is_valid_task_creation_response(response):
-        error_msg = f"Moonvalley Marey API: Initial request failed. Code: {response.code}, Message: {response.message}, Data: {response}"
-        logging.error(error_msg)
-        raise RuntimeError(error_msg)
-
-
-def validate_video_to_video_input(video: Input.Video) -> Input.Video:
-    """
-    Validates and processes video input for Moonvalley Video-to-Video generation.
-
-    Args:
-        video: Input video to validate
-
-    Returns:
-        Validated and potentially trimmed video
-
-    Raises:
-        ValueError: If video doesn't meet requirements
-        MoonvalleyApiError: If video duration is too short
-    """
-    width, height = _get_video_dimensions(video)
-    _validate_video_dimensions(width, height)
-    validate_container_format_is_mp4(video)
-
-    return _validate_and_trim_duration(video)
-
-
-def _get_video_dimensions(video: Input.Video) -> tuple[int, int]:
-    """Extracts video dimensions with error handling."""
-    try:
-        return video.get_dimensions()
-    except Exception as e:
-        logging.error("Error getting dimensions of video: %s", e)
-        raise ValueError(f"Cannot get video dimensions: {e}") from e
-
-
-def _validate_video_dimensions(width: int, height: int) -> None:
-    """Validates video dimensions meet Moonvalley V2V requirements."""
-    supported_resolutions = {
-        (1920, 1080),
-        (1080, 1920),
-        (1152, 1152),
-        (1536, 1152),
-        (1152, 1536),
-    }
-
-    if (width, height) not in supported_resolutions:
-        supported_list = ", ".join([f"{w}x{h}" for w, h in sorted(supported_resolutions)])
-        raise ValueError(f"Resolution {width}x{height} not supported. Supported: {supported_list}")
-
-
-def _validate_and_trim_duration(video: Input.Video) -> Input.Video:
-    """Validates video duration and trims to 5 seconds if needed."""
-    duration = video.get_duration()
-    _validate_minimum_duration(duration)
-    return _trim_if_too_long(video, duration)
-
-
-def _validate_minimum_duration(duration: float) -> None:
-    """Ensures video is at least 5 seconds long."""
-    if duration < 5:
-        raise ValueError("Input video must be at least 5 seconds long.")
-
-
-def _trim_if_too_long(video: Input.Video, duration: float) -> Input.Video:
-    """Trims video to 5 seconds if longer."""
-    if duration > 5:
-        return trim_video(video, 5)
-    return video
-
-
-def parse_width_height_from_res(resolution: str):
-    # Accepts a string like "16:9 (1920 x 1080)" and returns width, height as a dict
-    res_map = {
-        "16:9 (1920 x 1080)": {"width": 1920, "height": 1080},
-        "9:16 (1080 x 1920)": {"width": 1080, "height": 1920},
-        "1:1 (1152 x 1152)": {"width": 1152, "height": 1152},
-        "4:3 (1536 x 1152)": {"width": 1536, "height": 1152},
-        "3:4 (1152 x 1536)": {"width": 1152, "height": 1536},
-        # "21:9 (2560 x 1080)": {"width": 2560, "height": 1080},
-    }
-    return res_map.get(resolution, {"width": 1920, "height": 1080})
-
-
-def parse_control_parameter(value):
-    control_map = {
-        "Motion Transfer": "motion_control",
-        "Canny": "canny_control",
-        "Pose Transfer": "pose_control",
-        "Depth": "depth_control",
-    }
-    return control_map.get(value, control_map["Motion Transfer"])
-
-
-async def get_response(cls: type[IO.ComfyNode], task_id: str) -> MoonvalleyPromptResponse:
-    return await poll_op(
-        cls,
-        ApiEndpoint(path=f"{API_PROMPTS_ENDPOINT}/{task_id}"),
-        response_model=MoonvalleyPromptResponse,
-        status_extractor=lambda r: (r.status if r and r.status else None),
-        poll_interval=16.0,
-        max_poll_attempts=240,
-    )
-
-
-class MoonvalleyImg2VideoNode(IO.ComfyNode):
-
-    @classmethod
-    def define_schema(cls) -> IO.Schema:
-        return IO.Schema(
-            node_id="MoonvalleyImg2VideoNode",
-            display_name="Moonvalley Marey Image to Video",
-            category="api node/video/Moonvalley Marey",
-            description="Moonvalley Marey Image to Video Node",
-            inputs=[
-                IO.Image.Input(
-                    "image",
-                    tooltip="The reference image used to generate the video",
-                ),
-                IO.String.Input(
-                    "prompt",
-                    multiline=True,
-                ),
-                IO.String.Input(
-                    "negative_prompt",
-                    multiline=True,
-                    default="<synthetic> <scene cut> gopro, bright, contrast, static, overexposed, vignette, "
-                    "artifacts, still, noise, texture, scanlines, videogame, 360 camera, VR, transition, "
-                    "flare, saturation, distorted, warped, wide angle, saturated, vibrant, glowing, "
-                    "cross dissolve, cheesy, ugly hands, mutated hands, mutant, disfigured, extra fingers, "
-                    "blown out, horrible, blurry, worst quality, bad, dissolve, melt, fade in, fade out, "
-                    "wobbly, weird, low quality, plastic, stock footage, video camera, boring",
-                    tooltip="Negative prompt text",
-                ),
-                IO.Combo.Input(
-                    "resolution",
-                    options=[
-                        "16:9 (1920 x 1080)",
-                        "9:16 (1080 x 1920)",
-                        "1:1 (1152 x 1152)",
-                        "4:3 (1536 x 1152)",
-                        "3:4 (1152 x 1536)",
-                        # "21:9 (2560 x 1080)",
-                    ],
-                    default="16:9 (1920 x 1080)",
-                    tooltip="Resolution of the output video",
-                ),
-                IO.Float.Input(
-                    "prompt_adherence",
-                    default=4.5,
-                    min=1.0,
-                    max=20.0,
-                    step=1.0,
-                    tooltip="Guidance scale for generation control",
-                ),
-                IO.Int.Input(
-                    "seed",
-                    default=9,
-                    min=0,
-                    max=4294967295,
-                    step=1,
-                    display_mode=IO.NumberDisplay.number,
-                    tooltip="Random seed value",
-                    control_after_generate=True,
-                ),
-                IO.Int.Input(
-                    "steps",
-                    default=80,
-                    min=75,  # steps should be greater or equal to cooldown_steps(75) + warmup_steps(0)
-                    max=100,
-                    step=1,
-                    tooltip="Number of denoising steps",
-                ),
-            ],
-            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,
-            price_badge=IO.PriceBadge(
-                depends_on=IO.PriceBadgeDepends(),
-                expr="""{"type":"usd","usd": 1.5}""",
-            ),
-        )
-
-    @classmethod
-    async def execute(
-        cls,
-        image: Input.Image,
-        prompt: str,
-        negative_prompt: str,
-        resolution: str,
-        prompt_adherence: float,
-        seed: int,
-        steps: int,
-    ) -> IO.NodeOutput:
-        validate_image_dimensions(image, min_width=300, min_height=300, max_height=MAX_HEIGHT, max_width=MAX_WIDTH)
-        validate_string(prompt, min_length=1, max_length=MOONVALLEY_MAREY_MAX_PROMPT_LENGTH)
-        validate_string(negative_prompt, field_name="negative_prompt", max_length=MOONVALLEY_MAREY_MAX_PROMPT_LENGTH)
-        width_height = parse_width_height_from_res(resolution)
-
-        inference_params = MoonvalleyTextToVideoInferenceParams(
-            negative_prompt=negative_prompt,
-            steps=steps,
-            seed=seed,
-            guidance_scale=prompt_adherence,
-            width=width_height["width"],
-            height=width_height["height"],
-            use_negative_prompts=True,
-        )
-
-        # Get MIME type from tensor - assuming PNG format for image tensors
-        mime_type = "image/png"
-        image_url = (await upload_images_to_comfyapi(cls, image, max_images=1, mime_type=mime_type))[0]
-        task_creation_response = await sync_op(
-            cls,
-            endpoint=ApiEndpoint(path=API_IMG2VIDEO_ENDPOINT, method="POST"),
-            response_model=MoonvalleyPromptResponse,
-            data=MoonvalleyTextToVideoRequest(
-                image_url=image_url, prompt_text=prompt, inference_params=inference_params
-            ),
-        )
-        validate_task_creation_response(task_creation_response)
-        final_response = await get_response(cls, task_creation_response.id)
-        video = await download_url_to_video_output(final_response.output_url)
-        return IO.NodeOutput(video)
-
-
-class MoonvalleyVideo2VideoNode(IO.ComfyNode):
-
-    @classmethod
-    def define_schema(cls) -> IO.Schema:
-        return IO.Schema(
-            node_id="MoonvalleyVideo2VideoNode",
-            display_name="Moonvalley Marey Video to Video",
-            category="api node/video/Moonvalley Marey",
-            description="",
-            inputs=[
-                IO.String.Input(
-                    "prompt",
-                    multiline=True,
-                    tooltip="Describes the video to generate",
-                ),
-                IO.String.Input(
-                    "negative_prompt",
-                    multiline=True,
-                    default="<synthetic> <scene cut> gopro, bright, contrast, static, overexposed, vignette, "
-                    "artifacts, still, noise, texture, scanlines, videogame, 360 camera, VR, transition, "
-                    "flare, saturation, distorted, warped, wide angle, saturated, vibrant, glowing, "
-                    "cross dissolve, cheesy, ugly hands, mutated hands, mutant, disfigured, extra fingers, "
-                    "blown out, horrible, blurry, worst quality, bad, dissolve, melt, fade in, fade out, "
-                    "wobbly, weird, low quality, plastic, stock footage, video camera, boring",
-                    tooltip="Negative prompt text",
-                ),
-                IO.Int.Input(
-                    "seed",
-                    default=9,
-                    min=0,
-                    max=4294967295,
-                    step=1,
-                    display_mode=IO.NumberDisplay.number,
-                    tooltip="Random seed value",
-                    control_after_generate=False,
-                ),
-                IO.Video.Input(
-                    "video",
-                    tooltip="The reference video used to generate the output video. Must be at least 5 seconds long. "
-                    "Videos longer than 5s will be automatically trimmed. Only MP4 format supported.",
-                ),
-                IO.Combo.Input(
-                    "control_type",
-                    options=["Motion Transfer", "Pose Transfer"],
-                    default="Motion Transfer",
-                    optional=True,
-                ),
-                IO.Int.Input(
-                    "motion_intensity",
-                    default=100,
-                    min=0,
-                    max=100,
-                    step=1,
-                    tooltip="Only used if control_type is 'Motion Transfer'",
-                    optional=True,
-                ),
-                IO.Int.Input(
-                    "steps",
-                    default=60,
-                    min=60,  # steps should be greater or equal to cooldown_steps(36) + warmup_steps(24)
-                    max=100,
-                    step=1,
-                    display_mode=IO.NumberDisplay.number,
-                    tooltip="Number of inference steps",
-                ),
-            ],
-            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,
-            price_badge=IO.PriceBadge(
-                depends_on=IO.PriceBadgeDepends(),
-                expr="""{"type":"usd","usd": 2.25}""",
-            ),
-        )
-
-    @classmethod
-    async def execute(
-        cls,
-        prompt: str,
-        negative_prompt: str,
-        seed: int,
-        video: Input.Video | None = None,
-        control_type: str = "Motion Transfer",
-        motion_intensity: int | None = 100,
-        steps=60,
-        prompt_adherence=4.5,
-    ) -> IO.NodeOutput:
-        validated_video = validate_video_to_video_input(video)
-        video_url = await upload_video_to_comfyapi(cls, validated_video)
-        validate_string(prompt, min_length=1, max_length=MOONVALLEY_MAREY_MAX_PROMPT_LENGTH)
-        validate_string(negative_prompt, field_name="negative_prompt", max_length=MOONVALLEY_MAREY_MAX_PROMPT_LENGTH)
-
-        # Only include motion_intensity for Motion Transfer
-        control_params = {}
-        if control_type == "Motion Transfer" and motion_intensity is not None:
-            control_params["motion_intensity"] = motion_intensity
-
-        inference_params = MoonvalleyVideoToVideoInferenceParams(
-            negative_prompt=negative_prompt,
-            seed=seed,
-            control_params=control_params,
-            steps=steps,
-            guidance_scale=prompt_adherence,
-        )
-
-        task_creation_response = await sync_op(
-            cls,
-            endpoint=ApiEndpoint(path=API_VIDEO2VIDEO_ENDPOINT, method="POST"),
-            response_model=MoonvalleyPromptResponse,
-            data=MoonvalleyVideoToVideoRequest(
-                control_type=parse_control_parameter(control_type),
-                video_url=video_url,
-                prompt_text=prompt,
-                inference_params=inference_params,
-            ),
-        )
-        validate_task_creation_response(task_creation_response)
-        final_response = await get_response(cls, task_creation_response.id)
-        return IO.NodeOutput(await download_url_to_video_output(final_response.output_url))
-
-
-class MoonvalleyTxt2VideoNode(IO.ComfyNode):
-
-    @classmethod
-    def define_schema(cls) -> IO.Schema:
-        return IO.Schema(
-            node_id="MoonvalleyTxt2VideoNode",
-            display_name="Moonvalley Marey Text to Video",
-            category="api node/video/Moonvalley Marey",
-            description="",
-            inputs=[
-                IO.String.Input(
-                    "prompt",
-                    multiline=True,
-                ),
-                IO.String.Input(
-                    "negative_prompt",
-                    multiline=True,
-                    default="<synthetic> <scene cut> gopro, bright, contrast, static, overexposed, vignette, "
-                    "artifacts, still, noise, texture, scanlines, videogame, 360 camera, VR, transition, "
-                    "flare, saturation, distorted, warped, wide angle, saturated, vibrant, glowing, "
-                    "cross dissolve, cheesy, ugly hands, mutated hands, mutant, disfigured, extra fingers, "
-                    "blown out, horrible, blurry, worst quality, bad, dissolve, melt, fade in, fade out, "
-                    "wobbly, weird, low quality, plastic, stock footage, video camera, boring",
-                    tooltip="Negative prompt text",
-                ),
-                IO.Combo.Input(
-                    "resolution",
-                    options=[
-                        "16:9 (1920 x 1080)",
-                        "9:16 (1080 x 1920)",
-                        "1:1 (1152 x 1152)",
-                        "4:3 (1536 x 1152)",
-                        "3:4 (1152 x 1536)",
-                        "21:9 (2560 x 1080)",
-                    ],
-                    default="16:9 (1920 x 1080)",
-                    tooltip="Resolution of the output video",
-                ),
-                IO.Float.Input(
-                    "prompt_adherence",
-                    default=4.0,
-                    min=1.0,
-                    max=20.0,
-                    step=1.0,
-                    tooltip="Guidance scale for generation control",
-                ),
-                IO.Int.Input(
-                    "seed",
-                    default=9,
-                    min=0,
-                    max=4294967295,
-                    step=1,
-                    display_mode=IO.NumberDisplay.number,
-                    control_after_generate=True,
-                    tooltip="Random seed value",
-                ),
-                IO.Int.Input(
-                    "steps",
-                    default=80,
-                    min=75,  # steps should be greater or equal to cooldown_steps(75) + warmup_steps(0)
-                    max=100,
-                    step=1,
-                    tooltip="Inference steps",
-                ),
-            ],
-            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,
-            price_badge=IO.PriceBadge(
-                depends_on=IO.PriceBadgeDepends(),
-                expr="""{"type":"usd","usd": 1.5}""",
-            ),
-        )
-
-    @classmethod
-    async def execute(
-        cls,
-        prompt: str,
-        negative_prompt: str,
-        resolution: str,
-        prompt_adherence: float,
-        seed: int,
-        steps: int,
-    ) -> IO.NodeOutput:
-        validate_string(prompt, min_length=1, max_length=MOONVALLEY_MAREY_MAX_PROMPT_LENGTH)
-        validate_string(negative_prompt, field_name="negative_prompt", max_length=MOONVALLEY_MAREY_MAX_PROMPT_LENGTH)
-        width_height = parse_width_height_from_res(resolution)
-
-        inference_params = MoonvalleyTextToVideoInferenceParams(
-            negative_prompt=negative_prompt,
-            steps=steps,
-            seed=seed,
-            guidance_scale=prompt_adherence,
-            num_frames=128,
-            width=width_height["width"],
-            height=width_height["height"],
-        )
-
-        task_creation_response = await sync_op(
-            cls,
-            endpoint=ApiEndpoint(path=API_TXT2VIDEO_ENDPOINT, method="POST"),
-            response_model=MoonvalleyPromptResponse,
-            data=MoonvalleyTextToVideoRequest(prompt_text=prompt, inference_params=inference_params),
-        )
-        validate_task_creation_response(task_creation_response)
-        final_response = await get_response(cls, task_creation_response.id)
-        return IO.NodeOutput(await download_url_to_video_output(final_response.output_url))
-
-
-class MoonvalleyExtension(ComfyExtension):
-    @override
-    async def get_node_list(self) -> list[type[IO.ComfyNode]]:
-        return [
-            MoonvalleyImg2VideoNode,
-            MoonvalleyTxt2VideoNode,
-            MoonvalleyVideo2VideoNode,
-        ]
-
-
-async def comfy_entrypoint() -> MoonvalleyExtension:
-    return MoonvalleyExtension()

comfy_api_nodes/nodes_openai.py

@@ -415,8 +415,9 @@ class OpenAIGPTImage1(IO.ComfyNode):
                         "1152x2048",
                         "3840x2160",
                         "2160x3840",
+                        "Custom",
                     ],
-                    tooltip="Image size",
+                    tooltip="Image size. Select 'Custom' to use the custom width and height (GPT Image 2 only).",
                     optional=True,
                 ),
                 IO.Int.Input(
@@ -445,6 +446,24 @@ class OpenAIGPTImage1(IO.ComfyNode):
                     default="gpt-image-2",
                     optional=True,
                 ),
+                IO.Int.Input(
+                    "custom_width",
+                    default=1024,
+                    min=1024,
+                    max=3840,
+                    step=16,
+                    tooltip="Used only when `size` is 'Custom'. Must be a multiple of 16 (GPT Image 2 only).",
+                    optional=True,
+                ),
+                IO.Int.Input(
+                    "custom_height",
+                    default=1024,
+                    min=1024,
+                    max=3840,
+                    step=16,
+                    tooltip="Used only when `size` is 'Custom'. Must be a multiple of 16 (GPT Image 2 only).",
+                    optional=True,
+                ),
             ],
             outputs=[
                 IO.Image.Output(),
@@ -471,9 +490,9 @@ class OpenAIGPTImage1(IO.ComfyNode):
                       "high":   [0.133, 0.22]
                     },
                     "gpt-image-2": {
-                      "low":    [0.0048, 0.012],
-                      "medium": [0.041, 0.112],
-                      "high":   [0.165, 0.43]
+                      "low":    [0.0048, 0.019],
+                      "medium": [0.041, 0.168],
+                      "high":   [0.165, 0.67]
                     }
                   };
                   $range := $lookup($lookup($ranges, widgets.model), widgets.quality);
@@ -503,6 +522,8 @@ class OpenAIGPTImage1(IO.ComfyNode):
         mask: Input.Image | None = None,
         n: int = 1,
         size: str = "1024x1024",
+        custom_width: int = 1024,
+        custom_height: int = 1024,
         model: str = "gpt-image-1",
     ) -> IO.NodeOutput:
         validate_string(prompt, strip_whitespace=False)
@@ -510,7 +531,25 @@ class OpenAIGPTImage1(IO.ComfyNode):
         if mask is not None and image is None:
             raise ValueError("Cannot use a mask without an input image")
 
-        if model in ("gpt-image-1", "gpt-image-1.5"):
+        if size == "Custom":
+            if model != "gpt-image-2":
+                raise ValueError("Custom resolution is only supported by GPT Image 2 model")
+            if custom_width % 16 != 0 or custom_height % 16 != 0:
+                raise ValueError(f"Custom width and height must be multiples of 16, got {custom_width}x{custom_height}")
+            if max(custom_width, custom_height) > 3840:
+                raise ValueError(f"Custom resolution max edge must be <= 3840, got {custom_width}x{custom_height}")
+            ratio = max(custom_width, custom_height) / min(custom_width, custom_height)
+            if ratio > 3:
+                raise ValueError(
+                    f"Custom resolution aspect ratio must not exceed 3:1, got {custom_width}x{custom_height}"
+                )
+            total_pixels = custom_width * custom_height
+            if not 655_360 <= total_pixels <= 8_294_400:
+                raise ValueError(
+                    f"Custom resolution total pixels must be between 655,360 and 8,294,400, got {total_pixels}"
+                )
+            size = f"{custom_width}x{custom_height}"
+        elif model in ("gpt-image-1", "gpt-image-1.5"):
             if size not in ("auto", "1024x1024", "1024x1536", "1536x1024"):
                 raise ValueError(f"Resolution {size} is only supported by GPT Image 2 model")
 

comfy_api_nodes/nodes_topaz.py

@@ -453,7 +453,6 @@ class TopazVideoEnhance(IO.ComfyNode):
             progress_extractor=lambda x: getattr(x, "progress", 0),
             price_extractor=lambda x: (x.estimates.cost[0] * 0.08 if x.estimates and x.estimates.cost[0] else None),
             poll_interval=10.0,
-            max_poll_attempts=320,
         )
         return IO.NodeOutput(await download_url_to_video_output(final_response.download.url))
 

comfy_api_nodes/nodes_vidu.py

@@ -38,7 +38,7 @@ async def execute_task(
     cls: type[IO.ComfyNode],
     vidu_endpoint: str,
     payload: TaskCreationRequest | TaskExtendCreationRequest | TaskMultiFrameCreationRequest,
-    max_poll_attempts: int = 320,
+    max_poll_attempts: int = 480,
 ) -> list[TaskResult]:
     task_creation_response = await sync_op(
         cls,
@@ -1097,7 +1097,6 @@ class ViduExtendVideoNode(IO.ComfyNode):
                 video_url=await upload_video_to_comfyapi(cls, video, wait_label="Uploading video"),
                 images=[image_url] if image_url else None,
             ),
-            max_poll_attempts=480,
         )
         return IO.NodeOutput(await download_url_to_video_output(results[0].url))
 

comfy_api_nodes/nodes_wan.py

@@ -818,7 +818,6 @@ class WanReferenceVideoApi(IO.ComfyNode):
             response_model=VideoTaskStatusResponse,
             status_extractor=lambda x: x.output.task_status,
             poll_interval=6,
-            max_poll_attempts=280,
         )
         return IO.NodeOutput(await download_url_to_video_output(response.output.video_url))
 

comfy_api_nodes/nodes_wavespeed.py

@@ -84,7 +84,6 @@ class WavespeedFlashVSRNode(IO.ComfyNode):
             response_model=TaskResultResponse,
             status_extractor=lambda x: "failed" if x.data is None else x.data.status,
             poll_interval=10.0,
-            max_poll_attempts=480,
         )
         if final_response.code != 200:
             raise ValueError(
@@ -156,7 +155,6 @@ class WavespeedImageUpscaleNode(IO.ComfyNode):
             response_model=TaskResultResponse,
             status_extractor=lambda x: "failed" if x.data is None else x.data.status,
             poll_interval=10.0,
-            max_poll_attempts=480,
         )
         if final_response.code != 200:
             raise ValueError(

comfy_api_nodes/util/client.py

@@ -148,7 +148,7 @@ async def poll_op(
     queued_statuses: list[str | int] | None = None,
     data: BaseModel | None = None,
     poll_interval: float = 5.0,
-    max_poll_attempts: int = 160,
+    max_poll_attempts: int = 480,
     timeout_per_poll: float = 120.0,
     max_retries_per_poll: int = 10,
     retry_delay_per_poll: float = 1.0,
@@ -254,7 +254,7 @@ async def poll_op_raw(
     queued_statuses: list[str | int] | None = None,
     data: dict[str, Any] | BaseModel | None = None,
     poll_interval: float = 5.0,
-    max_poll_attempts: int = 160,
+    max_poll_attempts: int = 480,
     timeout_per_poll: float = 120.0,
     max_retries_per_poll: int = 10,
     retry_delay_per_poll: float = 1.0,

comfy_extras/nodes_sdpose.py

@@ -459,27 +459,23 @@ class SDPoseKeypointExtractor(io.ComfyNode):
         total_images = image.shape[0]
         captured_feat = None
 
-        model_h = int(head.heatmap_size[0]) * 4   # e.g. 192 * 4 = 768
-        model_w = int(head.heatmap_size[1]) * 4   # e.g. 256 * 4 = 1024
+        model_w = int(head.heatmap_size[0]) * 4   # 192 * 4 = 768
+        model_h = int(head.heatmap_size[1]) * 4   # 256 * 4 = 1024
 
         def _resize_to_model(imgs):
-            """Aspect-preserving resize + zero-pad BHWC images to (model_h, model_w). Returns (resized_bhwc, scale, pad_top, pad_left)."""
+            """Stretch BHWC images to (model_h, model_w), model expects no aspect preservation."""
             h, w = imgs.shape[-3], imgs.shape[-2]
-            scale = min(model_h / h, model_w / w)
-            sh, sw = int(round(h * scale)), int(round(w * scale))
-            pt, pl = (model_h - sh) // 2, (model_w - sw) // 2
+            method = "area" if (model_h <= h and model_w <= w) else "bilinear"
             chw = imgs.permute(0, 3, 1, 2).float()
-            scaled = comfy.utils.common_upscale(chw, sw, sh, upscale_method="bilinear", crop="disabled")
-            padded = torch.zeros(scaled.shape[0], scaled.shape[1], model_h, model_w, dtype=scaled.dtype, device=scaled.device)
-            padded[:, :, pt:pt + sh, pl:pl + sw] = scaled
-            return padded.permute(0, 2, 3, 1), scale, pt, pl
+            scaled = comfy.utils.common_upscale(chw, model_w, model_h, upscale_method=method, crop="disabled")
+            return scaled.permute(0, 2, 3, 1), model_w / w, model_h / h
 
-        def _remap_keypoints(kp, scale, pad_top, pad_left, offset_x=0, offset_y=0):
+        def _remap_keypoints(kp, scale_x, scale_y, offset_x=0, offset_y=0):
             """Remap keypoints from model space back to original image space."""
             kp = kp.copy() if isinstance(kp, np.ndarray) else np.array(kp, dtype=np.float32)
             invalid = kp[..., 0] < 0
-            kp[..., 0] = (kp[..., 0] - pad_left) / scale + offset_x
-            kp[..., 1] = (kp[..., 1] - pad_top)  / scale + offset_y
+            kp[..., 0] = kp[..., 0] / scale_x + offset_x
+            kp[..., 1] = kp[..., 1] / scale_y + offset_y
             kp[invalid] = -1
             return kp
 
@@ -529,18 +525,18 @@ class SDPoseKeypointExtractor(io.ComfyNode):
                             continue
 
                         crop = img[:, y1:y2, x1:x2, :]  # (1, crop_h, crop_w, C)
-                        crop_resized, scale, pad_top, pad_left = _resize_to_model(crop)
+                        crop_resized, sx, sy = _resize_to_model(crop)
 
                         latent_crop = vae.encode(crop_resized)
                         kp_batch, sc_batch = _run_on_latent(latent_crop)
-                        kp = _remap_keypoints(kp_batch[0], scale, pad_top, pad_left, x1, y1)
+                        kp = _remap_keypoints(kp_batch[0], sx, sy, x1, y1)
                         img_keypoints.append(kp)
                         img_scores.append(sc_batch[0])
                 else:
-                    img_resized, scale, pad_top, pad_left = _resize_to_model(img)
+                    img_resized, sx, sy = _resize_to_model(img)
                     latent_img = vae.encode(img_resized)
                     kp_batch, sc_batch = _run_on_latent(latent_img)
-                    img_keypoints.append(_remap_keypoints(kp_batch[0], scale, pad_top, pad_left))
+                    img_keypoints.append(_remap_keypoints(kp_batch[0], sx, sy))
                     img_scores.append(sc_batch[0])
 
                 all_keypoints.append(img_keypoints)
@@ -549,12 +545,12 @@ class SDPoseKeypointExtractor(io.ComfyNode):
 
         else: # full-image mode, batched
             for batch_start in tqdm(range(0, total_images, batch_size), desc="Extracting keypoints"):
-                batch_resized, scale, pad_top, pad_left = _resize_to_model(image[batch_start:batch_start + batch_size])
+                batch_resized, sx, sy = _resize_to_model(image[batch_start:batch_start + batch_size])
                 latent_batch = vae.encode(batch_resized)
                 kp_batch, sc_batch = _run_on_latent(latent_batch)
 
                 for kp, sc in zip(kp_batch, sc_batch):
-                    all_keypoints.append([_remap_keypoints(kp, scale, pad_top, pad_left)])
+                    all_keypoints.append([_remap_keypoints(kp, sx, sy)])
                     all_scores.append([sc])
 
                 pbar.update(len(kp_batch))
@@ -727,13 +723,13 @@ class CropByBBoxes(io.ComfyNode):
                 scale = min(output_width / crop_w, output_height / crop_h)
                 scaled_w = int(round(crop_w * scale))
                 scaled_h = int(round(crop_h * scale))
-                scaled = comfy.utils.common_upscale(crop_chw, scaled_w, scaled_h, upscale_method="bilinear", crop="disabled")
+                scaled = comfy.utils.common_upscale(crop_chw, scaled_w, scaled_h, upscale_method="area", crop="disabled")
                 pad_left = (output_width  - scaled_w) // 2
                 pad_top  = (output_height - scaled_h) // 2
                 resized = torch.zeros(1, num_ch, output_height, output_width, dtype=image.dtype, device=image.device)
                 resized[:, :, pad_top:pad_top + scaled_h, pad_left:pad_left + scaled_w] = scaled
             else:  # "stretch"
-                resized = comfy.utils.common_upscale(crop_chw, output_width, output_height, upscale_method="bilinear", crop="disabled")
+                resized = comfy.utils.common_upscale(crop_chw, output_width, output_height, upscale_method="area", crop="disabled")
             crops.append(resized)
 
         if not crops:

nodes.py

@@ -728,50 +728,26 @@ class LoraLoaderModelOnly(LoraLoader):
 
 class VAELoader:
     video_taes = ["taehv", "lighttaew2_2", "lighttaew2_1", "lighttaehy1_5", "taeltx_2"]
-    image_taes = ["taesd", "taesdxl", "taesd3", "taef1"]
+    image_taes = ["taesd", "taesdxl", "taesd3", "taef1", "taef2"]
+
     @staticmethod
     def vae_list(s):
         vaes = folder_paths.get_filename_list("vae")
         approx_vaes = folder_paths.get_filename_list("vae_approx")
-        sdxl_taesd_enc = False
-        sdxl_taesd_dec = False
-        sd1_taesd_enc = False
-        sd1_taesd_dec = False
-        sd3_taesd_enc = False
-        sd3_taesd_dec = False
-        f1_taesd_enc = False
-        f1_taesd_dec = False
-
+        have_img_encoder, have_img_decoder = set(), set()
         for v in approx_vaes:
-            if v.startswith("taesd_decoder."):
-                sd1_taesd_dec = True
-            elif v.startswith("taesd_encoder."):
-                sd1_taesd_enc = True
-            elif v.startswith("taesdxl_decoder."):
-                sdxl_taesd_dec = True
-            elif v.startswith("taesdxl_encoder."):
-                sdxl_taesd_enc = True
-            elif v.startswith("taesd3_decoder."):
-                sd3_taesd_dec = True
-            elif v.startswith("taesd3_encoder."):
-                sd3_taesd_enc = True
-            elif v.startswith("taef1_encoder."):
-                f1_taesd_dec = True
-            elif v.startswith("taef1_decoder."):
-                f1_taesd_enc = True
-            else:
+            parts = v.split("_", 1)
+            if len(parts) != 2 or parts[0] not in s.image_taes:
                 for tae in s.video_taes:
                     if v.startswith(tae):
                         vaes.append(v)
-
-        if sd1_taesd_dec and sd1_taesd_enc:
-            vaes.append("taesd")
-        if sdxl_taesd_dec and sdxl_taesd_enc:
-            vaes.append("taesdxl")
-        if sd3_taesd_dec and sd3_taesd_enc:
-            vaes.append("taesd3")
-        if f1_taesd_dec and f1_taesd_enc:
-            vaes.append("taef1")
+                        break
+                continue
+            if parts[1].startswith("encoder."):
+                have_img_encoder.add(parts[0])
+            elif parts[1].startswith("decoder."):
+                have_img_decoder.add(parts[0])
+        vaes += [k for k in have_img_decoder if k in have_img_encoder]
         vaes.append("pixel_space")
         return vaes
 
@@ -827,6 +803,11 @@ class VAELoader:
             else:
                 vae_path = folder_paths.get_full_path_or_raise("vae", vae_name)
             sd, metadata = comfy.utils.load_torch_file(vae_path, return_metadata=True)
+        if vae_name == "taef2":
+            if metadata is None:
+                metadata = {"tae_latent_channels": 128}
+            else:
+                metadata["tae_latent_channels"] = 128
         vae = comfy.sd.VAE(sd=sd, metadata=metadata)
         vae.throw_exception_if_invalid()
         return (vae,)

requirements.txt

@@ -1,5 +1,5 @@
 comfyui-frontend-package==1.42.15
-comfyui-workflow-templates==0.9.63
+comfyui-workflow-templates==0.9.66
 comfyui-embedded-docs==0.4.4
 torch
 torchsde
@@ -19,7 +19,7 @@ scipy
 tqdm
 psutil
 alembic
-SQLAlchemy>=2.0
+SQLAlchemy>=2.0.0
 filelock
 av>=14.2.0
 comfy-kitchen>=0.2.8