Merge 7d600fe114 into 9125613b53

* chore: update workflow templates to v0.8.9

* Update requirements.txt

comfy/ldm/chroma_radiance/model.py

@@ -9,12 +9,13 @@ import torch
 from torch import Tensor, nn
 from einops import repeat
 import comfy.ldm.common_dit
+import comfy.patcher_extension
 
-from comfy.ldm.flux.layers import EmbedND, DoubleStreamBlock, SingleStreamBlock
+from comfy.ldm.flux.layers import EmbedND, timestep_embedding, DoubleStreamBlock, SingleStreamBlock
 
-from comfy.ldm.chroma.model import Chroma, ChromaParams
 from comfy.ldm.chroma.layers import (
     Approximator,
+    ChromaModulationOut,
 )
 from .layers import (
     NerfEmbedder,
@@ -25,7 +26,26 @@ from .layers import (
 
 
 @dataclass
-class ChromaRadianceParams(ChromaParams):
+class ChromaRadianceParams:
+    # Fields from ChromaParams (now independent)
+    in_channels: int
+    out_channels: int
+    context_in_dim: int
+    hidden_size: int
+    mlp_ratio: float
+    num_heads: int
+    depth: int
+    depth_single_blocks: int
+    axes_dim: list
+    theta: int
+    qkv_bias: bool
+    in_dim: int
+    out_dim: int
+    hidden_dim: int
+    n_layers: int
+    txt_ids_dims: list
+    vec_in_dim: int
+    # ChromaRadiance-specific fields
     patch_size: int
     nerf_hidden_size: int
     nerf_mlp_ratio: int
@@ -39,7 +59,7 @@ class ChromaRadianceParams(ChromaParams):
     nerf_embedder_dtype: Optional[torch.dtype]
     use_x0: bool
 
-class ChromaRadiance(Chroma):
+class ChromaRadiance(nn.Module):
     """
     Transformer model for flow matching on sequences.
     """
@@ -47,7 +67,7 @@ class ChromaRadiance(Chroma):
     def __init__(self, image_model=None, final_layer=True, dtype=None, device=None, operations=None, **kwargs):
         if operations is None:
             raise RuntimeError("Attempt to create ChromaRadiance object without setting operations")
-        nn.Module.__init__(self)
+        super().__init__()
         self.dtype = dtype
         params = ChromaRadianceParams(**kwargs)
         self.params = params
@@ -176,6 +196,155 @@ class ChromaRadiance(Chroma):
         # flatten into a sequence for the transformer.
         return img.flatten(2).transpose(1, 2) # -> [B, NumPatches, Hidden]
 
+    def get_modulations(self, tensor: torch.Tensor, block_type: str, *, idx: int = 0):
+        # This function slices up the modulations tensor which has the following layout:
+        #   single     : num_single_blocks * 3 elements
+        #   double_img : num_double_blocks * 6 elements
+        #   double_txt : num_double_blocks * 6 elements
+        #   final      : 2 elements
+        if block_type == "final":
+            return (tensor[:, -2:-1, :], tensor[:, -1:, :])
+        single_block_count = self.params.depth_single_blocks
+        double_block_count = self.params.depth
+        offset = 3 * idx
+        if block_type == "single":
+            return ChromaModulationOut.from_offset(tensor, offset)
+        # Double block modulations are 6 elements so we double 3 * idx.
+        offset *= 2
+        if block_type in {"double_img", "double_txt"}:
+            # Advance past the single block modulations.
+            offset += 3 * single_block_count
+            if block_type == "double_txt":
+                # Advance past the double block img modulations.
+                offset += 6 * double_block_count
+            return (
+                ChromaModulationOut.from_offset(tensor, offset),
+                ChromaModulationOut.from_offset(tensor, offset + 3),
+            )
+        raise ValueError("Bad block_type")
+
+    def forward_orig(
+        self,
+        img: Tensor,
+        img_ids: Tensor,
+        txt: Tensor,
+        txt_ids: Tensor,
+        timesteps: Tensor,
+        guidance: Tensor = None,
+        control = None,
+        transformer_options={},
+        attn_mask: Tensor = None,
+    ) -> Tensor:
+        patches_replace = transformer_options.get("patches_replace", {})
+
+        # running on sequences img
+        img = self.img_in(img)
+
+        # distilled vector guidance
+        mod_index_length = 344
+        distill_timestep = timestep_embedding(timesteps.detach().clone(), 16).to(img.device, img.dtype)
+        # guidance = guidance *
+        distil_guidance = timestep_embedding(guidance.detach().clone(), 16).to(img.device, img.dtype)
+
+        # get all modulation index
+        modulation_index = timestep_embedding(torch.arange(mod_index_length, device=img.device), 32).to(img.device, img.dtype)
+        # we need to broadcast the modulation index here so each batch has all of the index
+        modulation_index = modulation_index.unsqueeze(0).repeat(img.shape[0], 1, 1).to(img.device, img.dtype)
+        # and we need to broadcast timestep and guidance along too
+        timestep_guidance = torch.cat([distill_timestep, distil_guidance], dim=1).unsqueeze(1).repeat(1, mod_index_length, 1).to(img.dtype).to(img.device, img.dtype)
+        # then and only then we could concatenate it together
+        input_vec = torch.cat([timestep_guidance, modulation_index], dim=-1).to(img.device, img.dtype)
+
+        mod_vectors = self.distilled_guidance_layer(input_vec)
+
+        txt = self.txt_in(txt)
+
+        ids = torch.cat((txt_ids, img_ids), dim=1)
+        pe = self.pe_embedder(ids)
+
+        blocks_replace = patches_replace.get("dit", {})
+        transformer_options["total_blocks"] = len(self.double_blocks)
+        transformer_options["block_type"] = "double"
+        for i, block in enumerate(self.double_blocks):
+            transformer_options["block_index"] = i
+            if i not in self.skip_mmdit:
+                double_mod = (
+                    self.get_modulations(mod_vectors, "double_img", idx=i),
+                    self.get_modulations(mod_vectors, "double_txt", idx=i),
+                )
+                if ("double_block", i) in blocks_replace:
+                    def block_wrap(args):
+                        out = {}
+                        out["img"], out["txt"] = block(img=args["img"],
+                                                       txt=args["txt"],
+                                                       vec=args["vec"],
+                                                       pe=args["pe"],
+                                                       attn_mask=args.get("attn_mask"),
+                                                       transformer_options=args.get("transformer_options"))
+                        return out
+
+                    out = blocks_replace[("double_block", i)]({"img": img,
+                                                               "txt": txt,
+                                                               "vec": double_mod,
+                                                               "pe": pe,
+                                                               "attn_mask": attn_mask,
+                                                               "transformer_options": transformer_options},
+                                                              {"original_block": block_wrap})
+                    txt = out["txt"]
+                    img = out["img"]
+                else:
+                    img, txt = block(img=img,
+                                     txt=txt,
+                                     vec=double_mod,
+                                     pe=pe,
+                                     attn_mask=attn_mask,
+                                     transformer_options=transformer_options)
+
+                if control is not None: # Controlnet
+                    control_i = control.get("input")
+                    if i < len(control_i):
+                        add = control_i[i]
+                        if add is not None:
+                            img += add
+
+        img = torch.cat((txt, img), 1)
+
+        transformer_options["total_blocks"] = len(self.single_blocks)
+        transformer_options["block_type"] = "single"
+        for i, block in enumerate(self.single_blocks):
+            transformer_options["block_index"] = i
+            if i not in self.skip_dit:
+                single_mod = self.get_modulations(mod_vectors, "single", idx=i)
+                if ("single_block", i) in blocks_replace:
+                    def block_wrap(args):
+                        out = {}
+                        out["img"] = block(args["img"],
+                                           vec=args["vec"],
+                                           pe=args["pe"],
+                                           attn_mask=args.get("attn_mask"),
+                                           transformer_options=args.get("transformer_options"))
+                        return out
+
+                    out = blocks_replace[("single_block", i)]({"img": img,
+                                                               "vec": single_mod,
+                                                               "pe": pe,
+                                                               "attn_mask": attn_mask,
+                                                               "transformer_options": transformer_options},
+                                                              {"original_block": block_wrap})
+                    img = out["img"]
+                else:
+                    img = block(img, vec=single_mod, pe=pe, attn_mask=attn_mask, transformer_options=transformer_options)
+
+                if control is not None: # Controlnet
+                    control_o = control.get("output")
+                    if i < len(control_o):
+                        add = control_o[i]
+                        if add is not None:
+                            img[:, txt.shape[1] :, ...] += add
+
+        img = img[:, txt.shape[1] :, ...]
+        return img
+
     def forward_nerf(
         self,
         img_orig: Tensor,
@@ -285,6 +454,13 @@ class ChromaRadiance(Chroma):
         eps = 0.0
         return (noisy - predicted) / (timesteps.view(-1,1,1,1) + eps)
 
+    def forward(self, x, timestep, context, guidance, control=None, transformer_options={}, **kwargs):
+        return comfy.patcher_extension.WrapperExecutor.new_class_executor(
+            self._forward,
+            self,
+            comfy.patcher_extension.get_all_wrappers(comfy.patcher_extension.WrappersMP.DIFFUSION_MODEL, transformer_options)
+        ).execute(x, timestep, context, guidance, control, transformer_options, **kwargs)
+
     def _forward(
         self,
         x: Tensor,
@@ -332,4 +508,3 @@ class ChromaRadiance(Chroma):
         if hasattr(self, "__x0__"):
             out = self._apply_x0_residual(out, img, timestep)
         return out
-

comfy/supported_models.py

@@ -763,7 +763,7 @@ class Flux2(Flux):
 
     def __init__(self, unet_config):
         super().__init__(unet_config)
-        self.memory_usage_factor = self.memory_usage_factor * (2.0 * 2.0) * 2.36
+        self.memory_usage_factor = self.memory_usage_factor * (2.0 * 2.0) * (unet_config['hidden_size'] / 2604)
 
     def get_model(self, state_dict, prefix="", device=None):
         out = model_base.Flux2(self, device=device)

comfy_api_nodes/apis/bytedance_api.py

@@ -65,11 +65,13 @@ class TaskImageContent(BaseModel):
 class Text2VideoTaskCreationRequest(BaseModel):
     model: str = Field(...)
     content: list[TaskTextContent] = Field(..., min_length=1)
+    generate_audio: bool | None = Field(...)
 
 
 class Image2VideoTaskCreationRequest(BaseModel):
     model: str = Field(...)
     content: list[TaskTextContent | TaskImageContent] = Field(..., min_length=2)
+    generate_audio: bool | None = Field(...)
 
 
 class TaskCreationResponse(BaseModel):
@@ -141,4 +143,9 @@ VIDEO_TASKS_EXECUTION_TIME = {
         "720p": 65,
         "1080p": 100,
     },
+    "seedance-1-5-pro-251215": {
+        "480p": 80,
+        "720p": 100,
+        "1080p": 150,
+    },
 }

comfy_api_nodes/nodes_bytedance.py

@@ -477,7 +477,12 @@ class ByteDanceTextToVideoNode(IO.ComfyNode):
             inputs=[
                 IO.Combo.Input(
                     "model",
-                    options=["seedance-1-0-pro-250528", "seedance-1-0-lite-t2v-250428", "seedance-1-0-pro-fast-251015"],
+                    options=[
+                        "seedance-1-5-pro-251215",
+                        "seedance-1-0-pro-250528",
+                        "seedance-1-0-lite-t2v-250428",
+                        "seedance-1-0-pro-fast-251015",
+                    ],
                     default="seedance-1-0-pro-fast-251015",
                 ),
                 IO.String.Input(
@@ -528,6 +533,12 @@ class ByteDanceTextToVideoNode(IO.ComfyNode):
                     tooltip='Whether to add an "AI generated" watermark to the video.',
                     optional=True,
                 ),
+                IO.Boolean.Input(
+                    "generate_audio",
+                    default=False,
+                    tooltip="This parameter is ignored for any model except seedance-1-5-pro.",
+                    optional=True,
+                ),
             ],
             outputs=[
                 IO.Video.Output(),
@@ -552,7 +563,10 @@ class ByteDanceTextToVideoNode(IO.ComfyNode):
         seed: int,
         camera_fixed: bool,
         watermark: bool,
+        generate_audio: bool = False,
     ) -> IO.NodeOutput:
+        if model == "seedance-1-5-pro-251215" and duration < 4:
+            raise ValueError("Minimum supported duration for Seedance 1.5 Pro is 4 seconds.")
         validate_string(prompt, strip_whitespace=True, min_length=1)
         raise_if_text_params(prompt, ["resolution", "ratio", "duration", "seed", "camerafixed", "watermark"])
 
@@ -567,7 +581,11 @@ class ByteDanceTextToVideoNode(IO.ComfyNode):
         )
         return await process_video_task(
             cls,
-            payload=Text2VideoTaskCreationRequest(model=model, content=[TaskTextContent(text=prompt)]),
+            payload=Text2VideoTaskCreationRequest(
+                model=model,
+                content=[TaskTextContent(text=prompt)],
+                generate_audio=generate_audio if model == "seedance-1-5-pro-251215" else None,
+            ),
             estimated_duration=max(1, math.ceil(VIDEO_TASKS_EXECUTION_TIME[model][resolution] * (duration / 10.0))),
         )
 
@@ -584,7 +602,12 @@ class ByteDanceImageToVideoNode(IO.ComfyNode):
             inputs=[
                 IO.Combo.Input(
                     "model",
-                    options=["seedance-1-0-pro-250528", "seedance-1-0-lite-t2v-250428", "seedance-1-0-pro-fast-251015"],
+                    options=[
+                        "seedance-1-5-pro-251215",
+                        "seedance-1-0-pro-250528",
+                        "seedance-1-0-lite-i2v-250428",
+                        "seedance-1-0-pro-fast-251015",
+                    ],
                     default="seedance-1-0-pro-fast-251015",
                 ),
                 IO.String.Input(
@@ -639,6 +662,12 @@ class ByteDanceImageToVideoNode(IO.ComfyNode):
                     tooltip='Whether to add an "AI generated" watermark to the video.',
                     optional=True,
                 ),
+                IO.Boolean.Input(
+                    "generate_audio",
+                    default=False,
+                    tooltip="This parameter is ignored for any model except seedance-1-5-pro.",
+                    optional=True,
+                ),
             ],
             outputs=[
                 IO.Video.Output(),
@@ -664,7 +693,10 @@ class ByteDanceImageToVideoNode(IO.ComfyNode):
         seed: int,
         camera_fixed: bool,
         watermark: bool,
+        generate_audio: bool = False,
     ) -> IO.NodeOutput:
+        if model == "seedance-1-5-pro-251215" and duration < 4:
+            raise ValueError("Minimum supported duration for Seedance 1.5 Pro is 4 seconds.")
         validate_string(prompt, strip_whitespace=True, min_length=1)
         raise_if_text_params(prompt, ["resolution", "ratio", "duration", "seed", "camerafixed", "watermark"])
         validate_image_dimensions(image, min_width=300, min_height=300, max_width=6000, max_height=6000)
@@ -686,6 +718,7 @@ class ByteDanceImageToVideoNode(IO.ComfyNode):
             payload=Image2VideoTaskCreationRequest(
                 model=model,
                 content=[TaskTextContent(text=prompt), TaskImageContent(image_url=TaskImageContentUrl(url=image_url))],
+                generate_audio=generate_audio if model == "seedance-1-5-pro-251215" else None,
             ),
             estimated_duration=max(1, math.ceil(VIDEO_TASKS_EXECUTION_TIME[model][resolution] * (duration / 10.0))),
         )
@@ -703,7 +736,7 @@ class ByteDanceFirstLastFrameNode(IO.ComfyNode):
             inputs=[
                 IO.Combo.Input(
                     "model",
-                    options=["seedance-1-0-pro-250528", "seedance-1-0-lite-i2v-250428"],
+                    options=["seedance-1-5-pro-251215", "seedance-1-0-pro-250528", "seedance-1-0-lite-i2v-250428"],
                     default="seedance-1-0-lite-i2v-250428",
                 ),
                 IO.String.Input(
@@ -762,6 +795,12 @@ class ByteDanceFirstLastFrameNode(IO.ComfyNode):
                     tooltip='Whether to add an "AI generated" watermark to the video.',
                     optional=True,
                 ),
+                IO.Boolean.Input(
+                    "generate_audio",
+                    default=False,
+                    tooltip="This parameter is ignored for any model except seedance-1-5-pro.",
+                    optional=True,
+                ),
             ],
             outputs=[
                 IO.Video.Output(),
@@ -788,7 +827,10 @@ class ByteDanceFirstLastFrameNode(IO.ComfyNode):
         seed: int,
         camera_fixed: bool,
         watermark: bool,
+        generate_audio: bool = False,
     ) -> IO.NodeOutput:
+        if model == "seedance-1-5-pro-251215" and duration < 4:
+            raise ValueError("Minimum supported duration for Seedance 1.5 Pro is 4 seconds.")
         validate_string(prompt, strip_whitespace=True, min_length=1)
         raise_if_text_params(prompt, ["resolution", "ratio", "duration", "seed", "camerafixed", "watermark"])
         for i in (first_frame, last_frame):
@@ -821,6 +863,7 @@ class ByteDanceFirstLastFrameNode(IO.ComfyNode):
                     TaskImageContent(image_url=TaskImageContentUrl(url=str(download_urls[0])), role="first_frame"),
                     TaskImageContent(image_url=TaskImageContentUrl(url=str(download_urls[1])), role="last_frame"),
                 ],
+                generate_audio=generate_audio if model == "seedance-1-5-pro-251215" else None,
             ),
             estimated_duration=max(1, math.ceil(VIDEO_TASKS_EXECUTION_TIME[model][resolution] * (duration / 10.0))),
         )
@@ -896,7 +939,41 @@ class ByteDanceImageReferenceNode(IO.ComfyNode):
                 IO.Hidden.unique_id,
             ],
             is_api_node=True,
-            price_badge=PRICE_BADGE_VIDEO,
+            price_badge=IO.PriceBadge(
+                depends_on=IO.PriceBadgeDepends(widgets=["model", "duration", "resolution"]),
+                expr="""
+                (
+                  $priceByModel := {
+                    "seedance-1-0-pro": {
+                      "480p":[0.23,0.24],
+                      "720p":[0.51,0.56]
+                    },
+                    "seedance-1-0-lite": {
+                      "480p":[0.17,0.18],
+                      "720p":[0.37,0.41]
+                    }
+                  };
+                  $model := widgets.model;
+                  $modelKey :=
+                    $contains($model, "seedance-1-0-pro")  ? "seedance-1-0-pro" :
+                    "seedance-1-0-lite";
+                  $resolution := widgets.resolution;
+                  $resKey :=
+                    $contains($resolution, "720") ? "720p" :
+                    "480p";
+                  $modelPrices := $lookup($priceByModel, $modelKey);
+                  $baseRange := $lookup($modelPrices, $resKey);
+                  $min10s := $baseRange[0];
+                  $max10s := $baseRange[1];
+                  $scale := widgets.duration / 10;
+                  $minCost := $min10s * $scale;
+                  $maxCost := $max10s * $scale;
+                  ($minCost = $maxCost)
+                    ? {"type":"usd","usd": $minCost}
+                    : {"type":"range_usd","min_usd": $minCost, "max_usd": $maxCost}
+                )
+                """,
+            ),
         )
 
     @classmethod
@@ -967,10 +1044,15 @@ def raise_if_text_params(prompt: str, text_params: list[str]) -> None:
 
 
 PRICE_BADGE_VIDEO = IO.PriceBadge(
-    depends_on=IO.PriceBadgeDepends(widgets=["model", "duration", "resolution"]),
+    depends_on=IO.PriceBadgeDepends(widgets=["model", "duration", "resolution", "generate_audio"]),
     expr="""
     (
       $priceByModel := {
+        "seedance-1-5-pro": {
+          "480p":[0.12,0.12],
+          "720p":[0.26,0.26],
+          "1080p":[0.58,0.59]
+        },
         "seedance-1-0-pro": {
           "480p":[0.23,0.24],
           "720p":[0.51,0.56],
@@ -989,6 +1071,7 @@ PRICE_BADGE_VIDEO = IO.PriceBadge(
       };
       $model := widgets.model;
       $modelKey :=
+        $contains($model, "seedance-1-5-pro")      ? "seedance-1-5-pro" :
         $contains($model, "seedance-1-0-pro-fast") ? "seedance-1-0-pro-fast" :
         $contains($model, "seedance-1-0-pro")      ? "seedance-1-0-pro" :
         "seedance-1-0-lite";
@@ -1002,11 +1085,12 @@ PRICE_BADGE_VIDEO = IO.PriceBadge(
       $min10s := $baseRange[0];
       $max10s := $baseRange[1];
       $scale := widgets.duration / 10;
-      $minCost := $min10s * $scale;
-      $maxCost := $max10s * $scale;
+      $audioMultiplier := ($modelKey = "seedance-1-5-pro" and widgets.generate_audio) ? 2 : 1;
+      $minCost := $min10s * $scale * $audioMultiplier;
+      $maxCost := $max10s * $scale * $audioMultiplier;
       ($minCost = $maxCost)
-        ? {"type":"usd","usd": $minCost}
-        : {"type":"range_usd","min_usd": $minCost, "max_usd": $maxCost}
+        ? {"type":"usd","usd": $minCost, "format": { "approximate": true }}
+        : {"type":"range_usd","min_usd": $minCost, "max_usd": $maxCost, "format": { "approximate": true }}
     )
     """,
 )

requirements.txt

@@ -1,5 +1,5 @@
 comfyui-frontend-package==1.36.14
-comfyui-workflow-templates==0.8.4
+comfyui-workflow-templates==0.8.10
 comfyui-embedded-docs==0.4.0
 torch
 torchsde

server.py

@@ -686,7 +686,10 @@ class PromptServer():
 
         @routes.get("/object_info")
         async def get_object_info(request):
-            seed_assets(["models"])
+            try:
+                seed_assets(["models"])
+            except Exception as e:
+                logging.error(f"Failed to seed assets: {e}")
             with folder_paths.cache_helper:
                 out = {}
                 for x in nodes.NODE_CLASS_MAPPINGS: