Merge 5905513e32 into 9125613b53

* chore: update workflow templates to v0.8.9

* Update requirements.txt

comfy/ldm/lumina/model.py

@@ -119,6 +119,9 @@ class JointAttention(nn.Module):
             xv = xv.unsqueeze(3).repeat(1, 1, 1, n_rep, 1).flatten(2, 3)
         output = optimized_attention_masked(xq.movedim(1, 2), xk.movedim(1, 2), xv.movedim(1, 2), self.n_local_heads, x_mask, skip_reshape=True, transformer_options=transformer_options)
 
+        if output.dtype == torch.float16:
+            output.div_(4)
+
         return self.out(output)
 
 
@@ -175,8 +178,12 @@ class FeedForward(nn.Module):
     def _forward_silu_gating(self, x1, x3):
         return clamp_fp16(F.silu(x1) * x3)
 
-    def forward(self, x):
-        return self.w2(self._forward_silu_gating(self.w1(x), self.w3(x)))
+    def forward(self, x, apply_fp16_downscale=False):
+        x3 = self.w3(x)
+        if x.dtype == torch.float16 and apply_fp16_downscale:
+            x3.div_(32)
+
+        return self.w2(self._forward_silu_gating(self.w1(x), x3))
 
 
 class JointTransformerBlock(nn.Module):
@@ -287,6 +294,7 @@ class JointTransformerBlock(nn.Module):
             x = x + gate_mlp.unsqueeze(1).tanh() * self.ffn_norm2(
                 clamp_fp16(self.feed_forward(
                     modulate(self.ffn_norm1(x), scale_mlp),
+                    apply_fp16_downscale=True,
                 ))
             )
         else:

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: