ComfyUI v0.21.1

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

.spectral.yaml

@@ -89,3 +89,12 @@ rules:
     then:
       field: description
       function: truthy
+
+overrides:
+  # /ws uses HTTP 101 (Switching Protocols) — a legitimate response for a
+  # WebSocket upgrade, but not a 2xx, so operation-success-response fires
+  # as a false positive. OpenAPI 3.x has no native WebSocket support.
+  - files:
+      - "openapi.yaml#/paths/~1ws"
+    rules:
+      operation-success-response: off

comfy/supported_models.py

@@ -1443,7 +1443,7 @@ class HiDreamO1(supported_models_base.BASE):
     }
 
     latent_format = latent_formats.HiDreamO1Pixel
-    memory_usage_factor = 0.6
+    memory_usage_factor = 0.033
     # fp16 not supported: LM MLP down_proj activations fp16 overflow, causing NaNs
     supported_inference_dtypes = [torch.bfloat16, torch.float32]
 

comfy/utils.py

@@ -1164,12 +1164,18 @@ def tiled_scale_multidim(samples, function, tile=(64, 64), overlap=8, upscale_am
 
             o = out
             o_d = out_div
+            ps_view = ps
+            mask_view = mask
             for d in range(dims):
-                o = o.narrow(d + 2, upscaled[d], mask.shape[d + 2])
-                o_d = o_d.narrow(d + 2, upscaled[d], mask.shape[d + 2])
+                l = min(ps_view.shape[d + 2], o.shape[d + 2] - upscaled[d])
+                o = o.narrow(d + 2, upscaled[d], l)
+                o_d = o_d.narrow(d + 2, upscaled[d], l)
+                if l < ps_view.shape[d + 2]:
+                    ps_view = ps_view.narrow(d + 2, 0, l)
+                    mask_view = mask_view.narrow(d + 2, 0, l)
 
-            o.add_(ps * mask)
-            o_d.add_(mask)
+            o.add_(ps_view * mask_view)
+            o_d.add_(mask_view)
 
             if pbar is not None:
                 pbar.update(1)

comfy_api/latest/_util/geometry_types.py

@@ -12,9 +12,24 @@ class VOXEL:
 
 
 class MESH:
-    def __init__(self, vertices: torch.Tensor, faces: torch.Tensor):
-        self.vertices = vertices
-        self.faces = faces
+    def __init__(self, vertices: torch.Tensor, faces: torch.Tensor,
+                 uvs: torch.Tensor | None = None,
+                 vertex_colors: torch.Tensor | None = None,
+                 texture: torch.Tensor | None = None,
+                 vertex_counts: torch.Tensor | None = None,
+                 face_counts: torch.Tensor | None = None):
+
+        assert (vertex_counts is None) == (face_counts is None), \
+            "vertex_counts and face_counts must be provided together (both or neither)"
+        self.vertices = vertices            # vertices: (B, N, 3)
+        self.faces = faces                  # faces: (B, M, 3)
+        self.uvs = uvs                      # uvs: (B, N, 2)
+        self.vertex_colors = vertex_colors  # vertex_colors: (B, N, 3 or 4)
+        self.texture = texture              # texture: (B, H, W, 3)
+        # When vertices/faces are zero-padded to a common N/M across the batch (variable-size mesh batch),
+        # these hold the real per-item lengths (B,). None means rows are uniform and no slicing is needed.
+        self.vertex_counts = vertex_counts
+        self.face_counts = face_counts
 
 
 class File3D:

comfy_api_nodes/apis/anthropic.py

@@ -0,0 +1,75 @@
+from enum import Enum
+from typing import Literal
+
+from pydantic import BaseModel, Field
+
+
+class AnthropicRole(str, Enum):
+    user = "user"
+    assistant = "assistant"
+
+
+class AnthropicTextContent(BaseModel):
+    type: Literal["text"] = "text"
+    text: str = Field(...)
+
+
+class AnthropicImageSourceBase64(BaseModel):
+    type: Literal["base64"] = "base64"
+    media_type: str = Field(..., description="MIME type of the image, e.g. image/png, image/jpeg")
+    data: str = Field(..., description="Base64-encoded image data")
+
+
+class AnthropicImageSourceUrl(BaseModel):
+    type: Literal["url"] = "url"
+    url: str = Field(...)
+
+
+class AnthropicImageContent(BaseModel):
+    type: Literal["image"] = "image"
+    source: AnthropicImageSourceBase64 | AnthropicImageSourceUrl = Field(...)
+
+
+class AnthropicMessage(BaseModel):
+    role: AnthropicRole = Field(...)
+    content: list[AnthropicTextContent | AnthropicImageContent] = Field(...)
+
+
+class AnthropicMessagesRequest(BaseModel):
+    model: str = Field(...)
+    messages: list[AnthropicMessage] = Field(...)
+    max_tokens: int = Field(..., ge=1)
+    system: str | None = Field(None, description="Top-level system prompt")
+    temperature: float | None = Field(None, ge=0.0, le=1.0)
+    top_p: float | None = Field(None, ge=0.0, le=1.0)
+    top_k: int | None = Field(None, ge=0)
+    stop_sequences: list[str] | None = Field(None)
+
+
+class AnthropicResponseTextBlock(BaseModel):
+    type: Literal["text"] = "text"
+    text: str = Field(...)
+
+
+class AnthropicCacheCreationUsage(BaseModel):
+    ephemeral_5m_input_tokens: int | None = Field(None)
+    ephemeral_1h_input_tokens: int | None = Field(None)
+
+
+class AnthropicMessagesUsage(BaseModel):
+    input_tokens: int | None = Field(None)
+    output_tokens: int | None = Field(None)
+    cache_creation_input_tokens: int | None = Field(None)
+    cache_read_input_tokens: int | None = Field(None)
+    cache_creation: AnthropicCacheCreationUsage | None = Field(None)
+
+
+class AnthropicMessagesResponse(BaseModel):
+    id: str | None = Field(None)
+    type: str | None = Field(None)
+    role: str | None = Field(None)
+    model: str | None = Field(None)
+    content: list[AnthropicResponseTextBlock] | None = Field(None)
+    stop_reason: str | None = Field(None)
+    stop_sequence: str | None = Field(None)
+    usage: AnthropicMessagesUsage | None = Field(None)

comfy_api_nodes/nodes_anthropic.py

@@ -0,0 +1,245 @@
+"""API Nodes for Anthropic Claude (Messages API). See: https://docs.anthropic.com/en/api/messages"""
+
+from typing_extensions import override
+
+from comfy_api.latest import IO, ComfyExtension, Input
+from comfy_api_nodes.apis.anthropic import (
+    AnthropicImageContent,
+    AnthropicImageSourceUrl,
+    AnthropicMessage,
+    AnthropicMessagesRequest,
+    AnthropicMessagesResponse,
+    AnthropicRole,
+    AnthropicTextContent,
+)
+from comfy_api_nodes.util import (
+    ApiEndpoint,
+    get_number_of_images,
+    sync_op,
+    upload_images_to_comfyapi,
+    validate_string,
+)
+
+ANTHROPIC_MESSAGES_ENDPOINT = "/proxy/anthropic/v1/messages"
+ANTHROPIC_IMAGE_MAX_PIXELS = 1568 * 1568
+CLAUDE_MAX_IMAGES = 20
+
+CLAUDE_MODELS: dict[str, str] = {
+    "Opus 4.7": "claude-opus-4-7",
+    "Opus 4.6": "claude-opus-4-6",
+    "Sonnet 4.6": "claude-sonnet-4-6",
+    "Sonnet 4.5": "claude-sonnet-4-5-20250929",
+    "Haiku 4.5": "claude-haiku-4-5-20251001",
+}
+
+
+def _claude_model_inputs():
+    return [
+        IO.Int.Input(
+            "max_tokens",
+            default=16000,
+            min=32,
+            max=32000,
+            tooltip="Maximum number of tokens to generate before stopping.",
+            advanced=True,
+        ),
+        IO.Float.Input(
+            "temperature",
+            default=1.0,
+            min=0.0,
+            max=1.0,
+            step=0.01,
+            tooltip="Controls randomness. 0.0 is deterministic, 1.0 is most random.",
+            advanced=True,
+        ),
+    ]
+
+
+def _model_price_per_million(model: str) -> tuple[float, float] | None:
+    """Return (input_per_1M, output_per_1M) USD for a Claude model, or None if unknown."""
+    if "opus-4-7" in model or "opus-4-6" in model or "opus-4-5" in model:
+        return 5.0, 25.0
+    if "sonnet-4" in model:
+        return 3.0, 15.0
+    if "haiku-4-5" in model:
+        return 1.0, 5.0
+    return None
+
+
+def calculate_tokens_price(response: AnthropicMessagesResponse) -> float | None:
+    """Compute approximate USD price from response usage. Server-side billing is authoritative."""
+    if not response.usage or not response.model:
+        return None
+    rates = _model_price_per_million(response.model)
+    if rates is None:
+        return None
+    input_rate, output_rate = rates
+    input_tokens = response.usage.input_tokens or 0
+    output_tokens = response.usage.output_tokens or 0
+    cache_read = response.usage.cache_read_input_tokens or 0
+    cache_5m = 0
+    cache_1h = 0
+    if response.usage.cache_creation:
+        cache_5m = response.usage.cache_creation.ephemeral_5m_input_tokens or 0
+        cache_1h = response.usage.cache_creation.ephemeral_1h_input_tokens or 0
+    total = (
+        input_tokens * input_rate
+        + output_tokens * output_rate
+        + cache_read * input_rate * 0.1
+        + cache_5m * input_rate * 1.25
+        + cache_1h * input_rate * 2.0
+    )
+    return total / 1_000_000.0
+
+
+def _get_text_from_response(response: AnthropicMessagesResponse) -> str:
+    if not response.content:
+        return ""
+    return "\n".join(block.text for block in response.content if block.text)
+
+
+async def _build_image_content_blocks(
+    cls: type[IO.ComfyNode],
+    image_tensors: list[Input.Image],
+) -> list[AnthropicImageContent]:
+    urls = await upload_images_to_comfyapi(
+        cls,
+        image_tensors,
+        max_images=CLAUDE_MAX_IMAGES,
+        total_pixels=ANTHROPIC_IMAGE_MAX_PIXELS,
+        wait_label="Uploading reference images",
+    )
+    return [AnthropicImageContent(source=AnthropicImageSourceUrl(url=url)) for url in urls]
+
+
+class ClaudeNode(IO.ComfyNode):
+    """Generate text responses from an Anthropic Claude model."""
+
+    @classmethod
+    def define_schema(cls):
+        return IO.Schema(
+            node_id="ClaudeNode",
+            display_name="Anthropic Claude",
+            category="api node/text/Anthropic",
+            essentials_category="Text Generation",
+            description="Generate text responses with Anthropic's Claude models. "
+            "Provide a text prompt and optionally one or more images for multimodal context.",
+            inputs=[
+                IO.String.Input(
+                    "prompt",
+                    multiline=True,
+                    default="",
+                    tooltip="Text input to the model.",
+                ),
+                IO.DynamicCombo.Input(
+                    "model",
+                    options=[IO.DynamicCombo.Option(label, _claude_model_inputs()) for label in CLAUDE_MODELS],
+                    tooltip="The Claude model used to generate the response.",
+                ),
+                IO.Int.Input(
+                    "seed",
+                    default=0,
+                    min=0,
+                    max=2147483647,
+                    control_after_generate=True,
+                    tooltip="Seed controls whether the node should re-run; "
+                    "results are non-deterministic regardless of seed.",
+                ),
+                IO.Autogrow.Input(
+                    "images",
+                    template=IO.Autogrow.TemplateNames(
+                        IO.Image.Input("image"),
+                        names=[f"image_{i}" for i in range(1, CLAUDE_MAX_IMAGES + 1)],
+                        min=0,
+                    ),
+                    tooltip=f"Optional image(s) to use as context for the model. Up to {CLAUDE_MAX_IMAGES} images.",
+                ),
+                IO.String.Input(
+                    "system_prompt",
+                    multiline=True,
+                    default="",
+                    optional=True,
+                    advanced=True,
+                    tooltip="Foundational instructions that dictate the model's behavior.",
+                ),
+            ],
+            outputs=[IO.String.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(widgets=["model"]),
+                expr="""
+                (
+                  $m := widgets.model;
+                  $contains($m, "opus") ? {
+                    "type": "list_usd",
+                    "usd": [0.005, 0.025],
+                    "format": { "approximate": true, "separator": "-", "suffix": " per 1K tokens" }
+                  }
+                  : $contains($m, "sonnet") ? {
+                    "type": "list_usd",
+                    "usd": [0.003, 0.015],
+                    "format": { "approximate": true, "separator": "-", "suffix": " per 1K tokens" }
+                  }
+                  : $contains($m, "haiku") ? {
+                    "type": "list_usd",
+                    "usd": [0.001, 0.005],
+                    "format": { "approximate": true, "separator": "-", "suffix": " per 1K tokens" }
+                  }
+                  : {"type":"text", "text":"Token-based"}
+                )
+                """,
+            ),
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        prompt: str,
+        model: dict,
+        seed: int,
+        images: dict | None = None,
+        system_prompt: str = "",
+    ) -> IO.NodeOutput:
+        validate_string(prompt, strip_whitespace=True, min_length=1)
+        model_label = model["model"]
+        max_tokens = model["max_tokens"]
+        temperature = model["temperature"]
+
+        image_tensors: list[Input.Image] = [t for t in (images or {}).values() if t is not None]
+        if sum(get_number_of_images(t) for t in image_tensors) > CLAUDE_MAX_IMAGES:
+            raise ValueError(f"Up to {CLAUDE_MAX_IMAGES} images are supported per request.")
+
+        content: list[AnthropicTextContent | AnthropicImageContent] = []
+        if image_tensors:
+            content.extend(await _build_image_content_blocks(cls, image_tensors))
+        content.append(AnthropicTextContent(text=prompt))
+
+        response = await sync_op(
+            cls,
+            ApiEndpoint(path=ANTHROPIC_MESSAGES_ENDPOINT, method="POST"),
+            response_model=AnthropicMessagesResponse,
+            data=AnthropicMessagesRequest(
+                model=CLAUDE_MODELS[model_label],
+                max_tokens=max_tokens,
+                messages=[AnthropicMessage(role=AnthropicRole.user, content=content)],
+                system=system_prompt or None,
+                temperature=temperature,
+            ),
+            price_extractor=calculate_tokens_price,
+        )
+        return IO.NodeOutput(_get_text_from_response(response) or "Empty response from Claude model.")
+
+
+class AnthropicExtension(ComfyExtension):
+    @override
+    async def get_node_list(self) -> list[type[IO.ComfyNode]]:
+        return [ClaudeNode]
+
+
+async def comfy_entrypoint() -> AnthropicExtension:
+    return AnthropicExtension()

comfy_api_nodes/nodes_quiver.py

@@ -143,7 +143,7 @@ class QuiverTextToSVGNode(IO.ComfyNode):
         if reference_images:
             references = []
             for key in reference_images:
-                url = await upload_image_to_comfyapi(cls, reference_images[key])
+                url = await upload_image_to_comfyapi(cls, reference_images[key], mime_type="image/png")
                 references.append(QuiverImageObject(url=url))
             if len(references) > 4:
                 raise ValueError("Maximum 4 reference images are allowed.")
@@ -252,7 +252,7 @@ class QuiverImageToSVGNode(IO.ComfyNode):
         model: dict,
         seed: int,
     ) -> IO.NodeOutput:
-        image_url = await upload_image_to_comfyapi(cls, image)
+        image_url = await upload_image_to_comfyapi(cls, image, mime_type="image/png")
 
         response = await sync_op(
             cls,

comfy_extras/nodes_audio.py

@@ -297,6 +297,7 @@ class LoadAudio(IO.ComfyNode):
     @classmethod
     def define_schema(cls):
         input_dir = folder_paths.get_input_directory()
+        os.makedirs(input_dir, exist_ok=True)
         files = folder_paths.filter_files_content_types(os.listdir(input_dir), ["audio", "video"])
         return IO.Schema(
             node_id="LoadAudio",

comfy_extras/nodes_hunyuan3d.py

@@ -1,12 +1,7 @@
 import torch
-import os
-import json
-import struct
-import numpy as np
 from comfy.ldm.modules.diffusionmodules.mmdit import get_1d_sincos_pos_embed_from_grid_torch
-import folder_paths
 import comfy.model_management
-from comfy.cli_args import args
+from comfy_extras.nodes_save_3d import pack_variable_mesh_batch
 from typing_extensions import override
 from comfy_api.latest import ComfyExtension, IO, Types
 from comfy_api.latest._util import MESH, VOXEL  # only for backward compatibility if someone import it from this file (will be removed later) # noqa
@@ -444,7 +439,9 @@ class VoxelToMeshBasic(IO.ComfyNode):
             vertices.append(v)
             faces.append(f)
 
-        return IO.NodeOutput(Types.MESH(torch.stack(vertices), torch.stack(faces)))
+        if vertices and all(v.shape == vertices[0].shape for v in vertices) and all(f.shape == faces[0].shape for f in faces):
+            return IO.NodeOutput(Types.MESH(torch.stack(vertices), torch.stack(faces)))
+        return IO.NodeOutput(pack_variable_mesh_batch(vertices, faces))
 
     decode = execute  # TODO: remove
 
@@ -481,206 +478,13 @@ class VoxelToMesh(IO.ComfyNode):
             vertices.append(v)
             faces.append(f)
 
-        return IO.NodeOutput(Types.MESH(torch.stack(vertices), torch.stack(faces)))
+        if vertices and all(v.shape == vertices[0].shape for v in vertices) and all(f.shape == faces[0].shape for f in faces):
+            return IO.NodeOutput(Types.MESH(torch.stack(vertices), torch.stack(faces)))
+        return IO.NodeOutput(pack_variable_mesh_batch(vertices, faces))
 
     decode = execute  # TODO: remove
 
 
-def save_glb(vertices, faces, filepath, metadata=None):
-    """
-    Save PyTorch tensor vertices and faces as a GLB file without external dependencies.
-
-    Parameters:
-    vertices: torch.Tensor of shape (N, 3) - The vertex coordinates
-    faces: torch.Tensor of shape (M, 3) - The face indices (triangle faces)
-    filepath: str - Output filepath (should end with .glb)
-    """
-
-    # Convert tensors to numpy arrays
-    vertices_np = vertices.cpu().numpy().astype(np.float32)
-    faces_np = faces.cpu().numpy().astype(np.uint32)
-
-    vertices_buffer = vertices_np.tobytes()
-    indices_buffer = faces_np.tobytes()
-
-    def pad_to_4_bytes(buffer):
-        padding_length = (4 - (len(buffer) % 4)) % 4
-        return buffer + b'\x00' * padding_length
-
-    vertices_buffer_padded = pad_to_4_bytes(vertices_buffer)
-    indices_buffer_padded = pad_to_4_bytes(indices_buffer)
-
-    buffer_data = vertices_buffer_padded + indices_buffer_padded
-
-    vertices_byte_length = len(vertices_buffer)
-    vertices_byte_offset = 0
-    indices_byte_length = len(indices_buffer)
-    indices_byte_offset = len(vertices_buffer_padded)
-
-    gltf = {
-        "asset": {"version": "2.0", "generator": "ComfyUI"},
-        "buffers": [
-            {
-                "byteLength": len(buffer_data)
-            }
-        ],
-        "bufferViews": [
-            {
-                "buffer": 0,
-                "byteOffset": vertices_byte_offset,
-                "byteLength": vertices_byte_length,
-                "target": 34962  # ARRAY_BUFFER
-            },
-            {
-                "buffer": 0,
-                "byteOffset": indices_byte_offset,
-                "byteLength": indices_byte_length,
-                "target": 34963  # ELEMENT_ARRAY_BUFFER
-            }
-        ],
-        "accessors": [
-            {
-                "bufferView": 0,
-                "byteOffset": 0,
-                "componentType": 5126,  # FLOAT
-                "count": len(vertices_np),
-                "type": "VEC3",
-                "max": vertices_np.max(axis=0).tolist(),
-                "min": vertices_np.min(axis=0).tolist()
-            },
-            {
-                "bufferView": 1,
-                "byteOffset": 0,
-                "componentType": 5125,  # UNSIGNED_INT
-                "count": faces_np.size,
-                "type": "SCALAR"
-            }
-        ],
-        "meshes": [
-            {
-                "primitives": [
-                    {
-                        "attributes": {
-                            "POSITION": 0
-                        },
-                        "indices": 1,
-                        "mode": 4  # TRIANGLES
-                    }
-                ]
-            }
-        ],
-        "nodes": [
-            {
-                "mesh": 0
-            }
-        ],
-        "scenes": [
-            {
-                "nodes": [0]
-            }
-        ],
-        "scene": 0
-    }
-
-    if metadata is not None:
-        gltf["asset"]["extras"] = metadata
-
-    # Convert the JSON to bytes
-    gltf_json = json.dumps(gltf).encode('utf8')
-
-    def pad_json_to_4_bytes(buffer):
-        padding_length = (4 - (len(buffer) % 4)) % 4
-        return buffer + b' ' * padding_length
-
-    gltf_json_padded = pad_json_to_4_bytes(gltf_json)
-
-    # Create the GLB header
-    # Magic glTF
-    glb_header = struct.pack('<4sII', b'glTF', 2, 12 + 8 + len(gltf_json_padded) + 8 + len(buffer_data))
-
-    # Create JSON chunk header (chunk type 0)
-    json_chunk_header = struct.pack('<II', len(gltf_json_padded), 0x4E4F534A)  # "JSON" in little endian
-
-    # Create BIN chunk header (chunk type 1)
-    bin_chunk_header = struct.pack('<II', len(buffer_data), 0x004E4942)  # "BIN\0" in little endian
-
-    # Write the GLB file
-    with open(filepath, 'wb') as f:
-        f.write(glb_header)
-        f.write(json_chunk_header)
-        f.write(gltf_json_padded)
-        f.write(bin_chunk_header)
-        f.write(buffer_data)
-
-    return filepath
-
-
-class SaveGLB(IO.ComfyNode):
-    @classmethod
-    def define_schema(cls):
-        return IO.Schema(
-            node_id="SaveGLB",
-            display_name="Save 3D Model",
-            search_aliases=["export 3d model", "save mesh"],
-            category="3d",
-            essentials_category="Basics",
-            is_output_node=True,
-            inputs=[
-                IO.MultiType.Input(
-                    IO.Mesh.Input("mesh"),
-                    types=[
-                        IO.File3DGLB,
-                        IO.File3DGLTF,
-                        IO.File3DOBJ,
-                        IO.File3DFBX,
-                        IO.File3DSTL,
-                        IO.File3DUSDZ,
-                        IO.File3DAny,
-                    ],
-                    tooltip="Mesh or 3D file to save",
-                ),
-                IO.String.Input("filename_prefix", default="3d/ComfyUI"),
-            ],
-            hidden=[IO.Hidden.prompt, IO.Hidden.extra_pnginfo]
-        )
-
-    @classmethod
-    def execute(cls, mesh: Types.MESH | Types.File3D, filename_prefix: str) -> IO.NodeOutput:
-        full_output_folder, filename, counter, subfolder, filename_prefix = folder_paths.get_save_image_path(filename_prefix, folder_paths.get_output_directory())
-        results = []
-
-        metadata = {}
-        if not args.disable_metadata:
-            if cls.hidden.prompt is not None:
-                metadata["prompt"] = json.dumps(cls.hidden.prompt)
-            if cls.hidden.extra_pnginfo is not None:
-                for x in cls.hidden.extra_pnginfo:
-                    metadata[x] = json.dumps(cls.hidden.extra_pnginfo[x])
-
-        if isinstance(mesh, Types.File3D):
-            # Handle File3D input - save BytesIO data to output folder
-            ext = mesh.format or "glb"
-            f = f"{filename}_{counter:05}_.{ext}"
-            mesh.save_to(os.path.join(full_output_folder, f))
-            results.append({
-                "filename": f,
-                "subfolder": subfolder,
-                "type": "output"
-            })
-        else:
-            # Handle Mesh input - save vertices and faces as GLB
-            for i in range(mesh.vertices.shape[0]):
-                f = f"{filename}_{counter:05}_.glb"
-                save_glb(mesh.vertices[i], mesh.faces[i], os.path.join(full_output_folder, f), metadata)
-                results.append({
-                    "filename": f,
-                    "subfolder": subfolder,
-                    "type": "output"
-                })
-                counter += 1
-        return IO.NodeOutput(ui={"3d": results})
-
-
 class Hunyuan3dExtension(ComfyExtension):
     @override
     async def get_node_list(self) -> list[type[IO.ComfyNode]]:
@@ -691,7 +495,6 @@ class Hunyuan3dExtension(ComfyExtension):
             VAEDecodeHunyuan3D,
             VoxelToMeshBasic,
             VoxelToMesh,
-            SaveGLB,
         ]
 
 

comfy_extras/nodes_images.py

@@ -3,23 +3,15 @@ from __future__ import annotations
 import nodes
 import folder_paths
 
-import av
 import json
-
 import os
 import re
 import math
-import numpy as np
-import struct
 import torch
-
-import zlib
 import comfy.utils
-from fractions import Fraction
 
 from server import PromptServer
 from comfy_api.latest import ComfyExtension, IO, UI
-from comfy.cli_args import args
 from typing_extensions import override
 
 SVG = IO.SVG.Type  # TODO: temporary solution for backward compatibility, will be removed later.
@@ -838,405 +830,6 @@ class ImageMergeTileList(IO.ComfyNode):
         return IO.NodeOutput(merged_image)
 
 
-# ---------------------------------------------------------------------------
-# Format specifications
-# ---------------------------------------------------------------------------
-
-# Maps (file_format, bit_depth, has_alpha) -> (numpy dtype scale, av pixel format,
-# stream pix_fmt). Keeps the encode path declarative instead of branchy.
-_FORMAT_SPECS = {
-    ("png", "8-bit", False):  {"scale": 255.0,   "dtype": np.uint8,   "frame_fmt": "rgb24",     "stream_fmt": "rgb24"},
-    ("png", "8-bit", True):   {"scale": 255.0,   "dtype": np.uint8,   "frame_fmt": "rgba",      "stream_fmt": "rgba"},
-    ("png", "16-bit", False): {"scale": 65535.0, "dtype": np.uint16,  "frame_fmt": "rgb48le",   "stream_fmt": "rgb48be"},
-    ("png", "16-bit", True):  {"scale": 65535.0, "dtype": np.uint16,  "frame_fmt": "rgba64le",  "stream_fmt": "rgba64be"},
-    ("exr", "32-bit float", False): {"scale": 1.0, "dtype": np.float32, "frame_fmt": "gbrpf32le",  "stream_fmt": "gbrpf32le"},
-    ("exr", "32-bit float", True):  {"scale": 1.0, "dtype": np.float32, "frame_fmt": "gbrapf32le", "stream_fmt": "gbrapf32le"},
-}
-
-
-# ---------------------------------------------------------------------------
-# Color transforms
-# ---------------------------------------------------------------------------
-
-def srgb_to_linear(t: torch.Tensor) -> torch.Tensor:
-    """Inverse sRGB EOTF (IEC 61966-2-1). Operates on RGB channels only;
-    alpha (if present as the 4th channel) is passed through unchanged."""
-    if t.shape[-1] == 4:
-        rgb, alpha = t[..., :3], t[..., 3:]
-        return torch.cat([srgb_to_linear(rgb), alpha], dim=-1)
-
-    # Piecewise: linear toe below 0.04045, gamma curve above.
-    low = t / 12.92
-    high = ((t.clamp(min=0.0) + 0.055) / 1.055) ** 2.4
-    return torch.where(t <= 0.04045, low, high)
-
-
-# HLG OETF constants from BT.2100 Table 5.
-_HLG_A = 0.17883277
-_HLG_B = 0.28466892
-_HLG_C = 0.55991072928   # = 0.5 - a*ln(4*a)
-
-
-def hlg_to_linear(t: torch.Tensor) -> torch.Tensor:
-    """Inverse HLG OETF (BT.2100). Maps a non-linear HLG signal in [0, 1] to
-    *scene*-linear light in [0, 1]. Per BT.2100 Note 5a, this is the correct
-    transform when converting HLG to a linear scene-light representation
-    (rather than display-light, which would also involve the HLG OOTF).
-
-    Operates on RGB channels only; alpha is passed through unchanged."""
-    if t.shape[-1] == 4:
-        rgb, alpha = t[..., :3], t[..., 3:]
-        return torch.cat([hlg_to_linear(rgb), alpha], dim=-1)
-
-    # Piecewise: sqrt branch below 0.5, log branch above.
-    # Clamp inside the log branch so negative / out-of-range values don't blow up;
-    # values above 1.0 are allowed and extrapolate naturally.
-    low = (t ** 2) / 3.0
-    high = (torch.exp((t.clamp(min=_HLG_C) - _HLG_C) / _HLG_A) + _HLG_B) / 12.0
-    return torch.where(t <= 0.5, low, high)
-
-
-# ---------------------------------------------------------------------------
-# Metadata injection
-# ---------------------------------------------------------------------------
-
-_PNG_SIGNATURE = b"\x89PNG\r\n\x1a\n"
-
-
-def _png_chunk(chunk_type: bytes, data: bytes) -> bytes:
-    """Build a single PNG chunk: length | type | data | CRC32(type+data)."""
-    crc = zlib.crc32(chunk_type + data) & 0xFFFFFFFF
-    return struct.pack(">I", len(data)) + chunk_type + data + struct.pack(">I", crc)
-
-
-def _png_text_chunk(keyword: str, text: str) -> bytes:
-    """tEXt chunk: latin-1 keyword + NUL + latin-1 text."""
-    payload = keyword.encode("latin-1") + b"\x00" + text.encode("latin-1", errors="replace")
-    return _png_chunk(b"tEXt", payload)
-
-
-def inject_png_metadata(png_bytes: bytes, prompt: dict | None, extra_pnginfo: dict | None) -> bytes:
-    """Insert ComfyUI prompt/workflow as tEXt chunks right after IHDR."""
-    if not png_bytes.startswith(_PNG_SIGNATURE):
-        return png_bytes
-
-    chunks: list[bytes] = []
-    if prompt is not None:
-        chunks.append(_png_text_chunk("prompt", json.dumps(prompt)))
-    if extra_pnginfo:
-        for key, value in extra_pnginfo.items():
-            chunks.append(_png_text_chunk(key, json.dumps(value)))
-    if not chunks:
-        return png_bytes
-
-    # IHDR is always the first chunk; insert ours immediately after it.
-    ihdr_length = struct.unpack(">I", png_bytes[8:12])[0]
-    ihdr_end = 8 + 8 + ihdr_length + 4  # signature + (len+type) + data + crc
-    return png_bytes[:ihdr_end] + b"".join(chunks) + png_bytes[ihdr_end:]
-
-
-# Standard chromaticities (CIE 1931 xy) for the colorspaces this node writes.
-# Each tuple is (Rx, Ry, Gx, Gy, Bx, By, Wx, Wy). All share D65 white point.
-_CHROMATICITIES = {
-    # ITU-R BT.709 / sRGB primaries
-    "Rec.709":  (0.6400, 0.3300, 0.3000, 0.6000, 0.1500, 0.0600, 0.3127, 0.3290),
-    # ITU-R BT.2020 (UHDTV / wide-gamut HDR) primaries
-    "Rec.2020": (0.7080, 0.2920, 0.1700, 0.7970, 0.1310, 0.0460, 0.3127, 0.3290),
-}
-
-
-def _pack_chromaticities(primaries: tuple) -> bytes:
-    """Serialize 8 chromaticity floats into the EXR `chromaticities` payload."""
-    return struct.pack("<8f", *primaries)
-
-
-def _exr_attribute(name: str, attr_type: str, value: bytes) -> bytes:
-    """Serialize one EXR header attribute: name\\0 type\\0 size:int32 value."""
-    return (
-        name.encode("utf-8") + b"\x00"
-        + attr_type.encode("utf-8") + b"\x00"
-        + struct.pack("<i", len(value))
-        + value
-    )
-
-
-def inject_exr_metadata(
-    exr_bytes: bytes,
-    prompt: dict | None,
-    extra_pnginfo: dict | None,
-    colorspace: str | None = None,
-) -> bytes:
-    """Insert ComfyUI metadata and color-space info into an EXR header.
-
-    Color: EXR pixels are linear by convention. The standard way to describe
-    their RGB→XYZ relationship is the `chromaticities` attribute. We pick the
-    primaries that match what the user told us their input was:
-
-      colorspace="sRGB" → Rec. 709 / sRGB primaries (D65)
-      colorspace="HDR"  → Rec. 2020 / BT.2100 primaries (D65)
-
-    Pixels are always converted to linear scene light upstream (sRGB EOTF
-    inverse for sRGB; HLG OETF inverse for HDR), so the file content is
-    scene-linear in the indicated gamut. OpenEXR has no standard transfer-
-    function attribute (the OpenEXR TSC has discussed adding one but it
-    doesn't exist), so we don't invent one — `chromaticities` plus the EXR
-    linear-by-convention rule fully specifies the color.
-
-    Prompt/workflow: written as plain `string` attributes using the same keys
-    (`prompt`, `workflow`, ...) that Comfy uses for PNG tEXt chunks, so the
-    same readers can pull them out symmetrically.
-
-    Implementation note: the chunk-offset table that follows the header stores
-    *absolute* byte offsets into the file. Inserting N bytes into the header
-    means every offset must be incremented by N or the file becomes unreadable.
-    """
-    if len(exr_bytes) < 8 or exr_bytes[:4] != b"\x76\x2f\x31\x01":
-        return exr_bytes
-
-    new_blob = b""
-    if prompt is not None:
-        new_blob += _exr_attribute("prompt", "string", json.dumps(prompt).encode("utf-8"))
-    if extra_pnginfo:
-        for key, value in extra_pnginfo.items():
-            new_blob += _exr_attribute(key, "string", json.dumps(value).encode("utf-8"))
-    if colorspace is not None:
-        # Map each colorspace option to the RGB primaries the linear pixels
-        # are now in. "sRGB" and "linear" both produce Rec. 709 linear; "HDR"
-        # (HLG-encoded Rec. 2020 input) produces Rec. 2020 linear.
-        primaries_name = {
-            "sRGB":   "Rec.709",
-            "linear": "Rec.709",
-            "HDR":    "Rec.2020",
-        }.get(colorspace, "Rec.709")
-        new_blob += _exr_attribute(
-            "chromaticities",
-            "chromaticities",
-            _pack_chromaticities(_CHROMATICITIES[primaries_name]),
-        )
-    if not new_blob:
-        return exr_bytes
-
-    # Walk header attributes to find the terminating null byte, and pick up
-    # dataWindow + compression so we know how many chunks the offset table has.
-    pos = 8  # past magic (4) + version (4)
-    data_window = None
-    compression = 0
-    while pos < len(exr_bytes) and exr_bytes[pos] != 0:
-        name_end = exr_bytes.index(b"\x00", pos)
-        attr_name = exr_bytes[pos:name_end].decode("latin-1", errors="replace")
-        type_end = exr_bytes.index(b"\x00", name_end + 1)
-        attr_type = exr_bytes[name_end + 1:type_end].decode("latin-1", errors="replace")
-        size = struct.unpack("<i", exr_bytes[type_end + 1:type_end + 5])[0]
-        value_start = type_end + 5
-        value = exr_bytes[value_start:value_start + size]
-
-        if attr_name == "dataWindow" and attr_type == "box2i":
-            data_window = struct.unpack("<iiii", value)  # xMin, yMin, xMax, yMax
-        elif attr_name == "compression" and attr_type == "compression":
-            compression = value[0]
-
-        pos = value_start + size
-
-    if data_window is None:
-        return exr_bytes  # required attribute missing — don't risk corrupting
-
-    # Scanlines per chunk by compression, from the OpenEXR spec.
-    scanlines_per_block = {
-        0: 1,   # NO_COMPRESSION
-        1: 1,   # RLE
-        2: 1,   # ZIPS
-        3: 16,  # ZIP
-        4: 32,  # PIZ
-        5: 16,  # PXR24
-        6: 32,  # B44
-        7: 32,  # B44A
-        8: 256, # DWAA
-        9: 256, # DWAB
-    }.get(compression, 1)
-
-    _, y_min, _, y_max = data_window
-    height = y_max - y_min + 1
-    num_chunks = (height + scanlines_per_block - 1) // scanlines_per_block
-
-    header_end = pos  # position of the terminating null byte
-    table_start = header_end + 1
-    pixel_start = table_start + num_chunks * 8
-    delta = len(new_blob)
-
-    old_offsets = struct.unpack(f"<{num_chunks}Q", exr_bytes[table_start:pixel_start])
-    new_table = struct.pack(f"<{num_chunks}Q", *(o + delta for o in old_offsets))
-
-    return (
-        exr_bytes[:header_end]                # header attributes
-        + new_blob                            # our new attributes
-        + exr_bytes[header_end:table_start]   # terminating null byte
-        + new_table                           # shifted offset table
-        + exr_bytes[pixel_start:]             # pixel data, untouched
-    )
-
-
-# ---------------------------------------------------------------------------
-# Encoding
-# ---------------------------------------------------------------------------
-
-def _encode_image(
-    img_tensor: torch.Tensor,
-    file_format: str,
-    bit_depth: str,
-    colorspace: str,
-) -> bytes:
-    """Encode a single HxWxC tensor to PNG or EXR bytes in memory.
-
-    For EXR the input is interpreted according to `colorspace` and converted
-    to scene-linear (EXR's convention) before writing:
-
-      "sRGB"   → input is sRGB-encoded Rec. 709; apply inverse sRGB EOTF.
-      "HDR"    → input is HLG-encoded Rec. 2020 (BT.2100); apply inverse HLG
-                 OETF to get scene-linear, per BT.2100 Note 5a.
-      "linear" → input is already scene-linear (Rec. 709 primaries); write
-                 through unchanged. Use this for renderer/compositor output.
-
-    For PNG, colorspace selection does not modify pixels — PNG is delivered
-    sRGB-encoded and there is no PNG path for wide-gamut HDR in this node.
-    """
-    height, width, num_channels = img_tensor.shape
-    has_alpha = num_channels == 4
-
-    spec = _FORMAT_SPECS[(file_format, bit_depth, has_alpha)]
-
-    if spec["dtype"] == np.float32:
-        # EXR path: preserve full range, no clamp.
-        if colorspace == "sRGB":
-            img_tensor = srgb_to_linear(img_tensor)
-        elif colorspace == "HDR":
-            img_tensor = hlg_to_linear(img_tensor)
-        img_np = img_tensor.cpu().numpy().astype(np.float32)
-    else:
-        # PNG path: quantize to integer range.
-        scaled = (img_tensor * spec["scale"]).clamp(0, spec["scale"])
-        img_np = scaled.to(torch.int32).cpu().numpy().astype(spec["dtype"])
-
-    # Encode directly via CodecContext. PyAV's `image2` muxer does NOT write to
-    # BytesIO (it expects a real file path), so we bypass the container entirely.
-    # For single-frame PNG/EXR the raw codec output IS the file.
-    codec = av.CodecContext.create(file_format, "w")
-    codec.width = width
-    codec.height = height
-    codec.pix_fmt = spec["stream_fmt"]
-    codec.time_base = Fraction(1, 1)
-
-    frame = av.VideoFrame.from_ndarray(img_np, format=spec["frame_fmt"])
-    if spec["frame_fmt"] != spec["stream_fmt"]:
-        frame = frame.reformat(format=spec["stream_fmt"])
-    frame.pts = 0
-    frame.time_base = codec.time_base
-
-    packets = list(codec.encode(frame)) + list(codec.encode(None))  # flush with None
-    return b"".join(bytes(p) for p in packets)
-
-
-# ---------------------------------------------------------------------------
-# Node
-# ---------------------------------------------------------------------------
-
-class SaveImageAdvanced(IO.ComfyNode):
-    @classmethod
-    def define_schema(cls):
-        return IO.Schema(
-            node_id="SaveImageAdvanced",
-            search_aliases=["save", "save image", "export image", "output image", "write image"],
-            display_name="Save Image (Advanced)",
-            description="Saves the input images to your ComfyUI output directory.",
-            category="image",
-            essentials_category="Basics",
-            inputs=[
-                IO.Image.Input("images", tooltip="The images to save."),
-                IO.String.Input(
-                    "filename_prefix",
-                    default="ComfyUI",
-                    tooltip=(
-                        "The prefix for the file to save. May include formatting tokens "
-                        "such as %date:yyyy-MM-dd% or %Empty Latent Image.width%."
-                    ),
-                ),
-                IO.DynamicCombo.Input(
-                    "image_format",
-                    options=[
-                        IO.DynamicCombo.Option("png", [
-                            IO.Combo.Input("bit_depth", options=["8-bit", "16-bit"],
-                                           default="8-bit", advanced=True),
-                            IO.Combo.Input("colorspace", options=["sRGB"],
-                                           default="sRGB", advanced=True),
-                        ]),
-                        IO.DynamicCombo.Option("exr", [
-                            IO.Combo.Input("bit_depth", options=["32-bit float"],
-                                           default="32-bit float", advanced=True),
-                            IO.Combo.Input(
-                                "colorspace",
-                                options=["sRGB", "HDR", "linear"],
-                                default="sRGB",
-                                advanced=True,
-                                tooltip=(
-                                    "Colorspace of the input tensor. The EXR is "
-                                    "always written as scene-linear in the matching "
-                                    "gamut.\n"
-                                    "  'sRGB'   — input is sRGB-encoded Rec.709; "
-                                    "the inverse sRGB EOTF is applied.\n"
-                                    "  'HDR'    — input is HLG-encoded Rec.2020 "
-                                    "(BT.2100); the inverse HLG OETF is applied "
-                                    "to get scene-linear light.\n"
-                                    "  'linear' — input is already scene-linear "
-                                    "(Rec.709 primaries); written through unchanged. "
-                                    "Use this for renderer/compositor output."
-                                ),
-                            ),
-                        ]),
-                    ],
-                    tooltip="The file format in which to save the image.",
-                ),
-            ],
-            hidden=[IO.Hidden.prompt, IO.Hidden.extra_pnginfo],
-            is_output_node=True,
-        )
-
-    @classmethod
-    def execute(cls, images, filename_prefix: str, image_format: dict) -> IO.NodeOutput:
-        file_format = image_format["image_format"]
-        bit_depth = image_format["bit_depth"]
-        colorspace = image_format.get("colorspace", "sRGB")
-
-        output_dir = folder_paths.get_output_directory()
-        full_output_folder, filename, counter, subfolder, filename_prefix = (
-            folder_paths.get_save_image_path(
-                filename_prefix, output_dir, images[0].shape[1], images[0].shape[0]
-            )
-        )
-
-        prompt = cls.hidden.prompt
-        extra_pnginfo = cls.hidden.extra_pnginfo
-        write_metadata = not args.disable_metadata
-
-        results = []
-        for batch_number, image in enumerate(images):
-            encoded = _encode_image(image, file_format, bit_depth, colorspace)
-
-            if write_metadata:
-                if file_format == "png":
-                    encoded = inject_png_metadata(encoded, prompt, extra_pnginfo)
-                elif file_format == "exr":
-                    encoded = inject_exr_metadata(encoded, prompt, extra_pnginfo, colorspace)
-
-            name = filename.replace("%batch_num%", str(batch_number))
-            file = f"{name}_{counter:05}.{file_format}"
-            with open(os.path.join(full_output_folder, file), "wb") as f:
-                f.write(encoded)
-
-            results.append({"filename": file, "subfolder": subfolder, "type": "output"})
-            counter += 1
-
-        return IO.NodeOutput(ui={"images": results})
-
-
 class ImagesExtension(ComfyExtension):
     @override
     async def get_node_list(self) -> list[type[IO.ComfyNode]]:
@@ -1249,7 +842,6 @@ class ImagesExtension(ComfyExtension):
             ImageAddNoise,
             SaveAnimatedWEBP,
             SaveAnimatedPNG,
-            SaveImageAdvanced,
             SaveSVGNode,
             ImageStitch,
             ResizeAndPadImage,

comfy_extras/nodes_lt.py

@@ -338,8 +338,25 @@ class LTXVAddGuide(io.ComfyNode):
         noise_mask = get_noise_mask(latent)
 
         _, _, latent_length, latent_height, latent_width = latent_image.shape
+
+        # For mid-video multi-frame guides, prepend+strip a throwaway first frame so the VAE's "first latent = 1 pixel frame" asymmetry lands on the discarded slot
+        time_scale_factor = scale_factors[0]
+        num_frames_to_keep = ((image.shape[0] - 1) // time_scale_factor) * time_scale_factor + 1
+        resolved_frame_idx = frame_idx
+        if frame_idx < 0:
+            _, num_keyframes = get_keyframe_idxs(positive)
+            resolved_frame_idx = max((latent_length - num_keyframes - 1) * time_scale_factor + 1 + frame_idx, 0)
+        causal_fix = resolved_frame_idx == 0 or num_frames_to_keep == 1
+
+        if not causal_fix:
+            image = torch.cat([image[:1], image], dim=0)
+
         image, t = cls.encode(vae, latent_width, latent_height, image, scale_factors)
 
+        if not causal_fix:
+            t = t[:, :, 1:, :, :]
+            image = image[1:]
+
         frame_idx, latent_idx = cls.get_latent_index(positive, latent_length, len(image), frame_idx, scale_factors)
         assert latent_idx + t.shape[2] <= latent_length, "Conditioning frames exceed the length of the latent sequence."
 
@@ -352,6 +369,7 @@ class LTXVAddGuide(io.ComfyNode):
             t,
             strength,
             scale_factors,
+            causal_fix=causal_fix,
         )
 
         # Track this guide for per-reference attention control.

comfy_extras/nodes_mask.py

@@ -40,23 +40,13 @@ def composite(destination, source, x, y, mask = None, multiplier = 8, resize_sou
 
     inverse_mask = torch.ones_like(mask) - mask
 
-    source_rgb = source[:, :3, :visible_height, :visible_width]
-    dest_slice = destination[..., top:bottom, left:right]
-
-    if destination.shape[1] == 4:
-        if torch.max(dest_slice) == 0:
-            destination[:, :3, top:bottom, left:right] = source_rgb
-            destination[:, 3:4, top:bottom, left:right] = mask
-        else:
-            destination[:, :3, top:bottom, left:right] = (mask * source_rgb) + (inverse_mask * dest_slice[:, :3])
-            destination[:, 3:4, top:bottom, left:right] = torch.max(mask, dest_slice[:, 3:4])
-    else:
-        source_portion = mask * source_rgb
-        destination_portion = inverse_mask * dest_slice
-        destination[..., top:bottom, left:right] = source_portion + destination_portion
+    source_portion = mask * source[..., :visible_height, :visible_width]
+    destination_portion = inverse_mask  * destination[..., top:bottom, left:right]
 
+    destination[..., top:bottom, left:right] = source_portion + destination_portion
     return destination
 
+
 class LatentCompositeMasked(IO.ComfyNode):
     @classmethod
     def define_schema(cls):
@@ -95,23 +85,18 @@ class ImageCompositeMasked(IO.ComfyNode):
             display_name="Image Composite Masked",
             category="image",
             inputs=[
+                IO.Image.Input("destination"),
                 IO.Image.Input("source"),
                 IO.Int.Input("x", default=0, min=0, max=nodes.MAX_RESOLUTION, step=1),
                 IO.Int.Input("y", default=0, min=0, max=nodes.MAX_RESOLUTION, step=1),
                 IO.Boolean.Input("resize_source", default=False),
-                IO.Image.Input("destination", optional=True),
                 IO.Mask.Input("mask", optional=True),
             ],
             outputs=[IO.Image.Output()],
         )
 
     @classmethod
-    def execute(cls, source, x, y, resize_source, destination = None, mask = None) -> IO.NodeOutput:
-        if destination is None: # transparent rgba
-            B, H, W, C = source.shape
-            destination = torch.zeros((B, H, W, 4), dtype=source.dtype, device=source.device)
-            if C == 3:
-                source = torch.nn.functional.pad(source, (0, 1), value=1.0)
+    def execute(cls, destination, source, x, y, resize_source, mask = None) -> IO.NodeOutput:
         destination, source = node_helpers.image_alpha_fix(destination, source)
         destination = destination.clone().movedim(-1, 1)
         output = composite(destination, source.movedim(-1, 1), x, y, mask, 1, resize_source).movedim(1, -1)

comfy_extras/nodes_save_3d.py

@@ -0,0 +1,396 @@
+"""Save-side 3D nodes: mesh packing/slicing helpers + GLB writer + SaveGLB node."""
+
+import json
+import logging
+import os
+import struct
+from io import BytesIO
+
+import numpy as np
+from PIL import Image
+import torch
+from typing_extensions import override
+
+import folder_paths
+from comfy.cli_args import args
+from comfy_api.latest import ComfyExtension, IO, Types
+
+
+def pack_variable_mesh_batch(vertices, faces, colors=None, uvs=None, texture=None):
+    # Pack lists of (Nᵢ, *) vertex/face/color/uv tensors into padded batched tensors,
+    # stashing per-item lengths as runtime attrs so consumers can recover the real slice.
+    # colors and uvs are 1:1 with vertices, so they're padded to max_vertices and read with vertex_counts.
+    # texture is (B, H, W, 3) — passed through unchanged
+    batch_size = len(vertices)
+    max_vertices = max(v.shape[0] for v in vertices)
+    max_faces = max(f.shape[0] for f in faces)
+
+    packed_vertices = vertices[0].new_zeros((batch_size, max_vertices, vertices[0].shape[1]))
+    packed_faces = faces[0].new_zeros((batch_size, max_faces, faces[0].shape[1]))
+    vertex_counts = torch.tensor([v.shape[0] for v in vertices], device=vertices[0].device, dtype=torch.int64)
+    face_counts = torch.tensor([f.shape[0] for f in faces], device=faces[0].device, dtype=torch.int64)
+
+    for i, (v, f) in enumerate(zip(vertices, faces)):
+        packed_vertices[i, :v.shape[0]] = v
+        packed_faces[i, :f.shape[0]] = f
+
+    packed_colors = None
+    if colors is not None:
+        packed_colors = colors[0].new_zeros((batch_size, max_vertices, colors[0].shape[1]))
+        for i, c in enumerate(colors):
+            assert c.shape[0] == vertices[i].shape[0], (
+                f"vertex_colors[{i}] has {c.shape[0]} entries, expected {vertices[i].shape[0]} (1:1 with vertices)"
+            )
+            packed_colors[i, :c.shape[0]] = c
+
+    packed_uvs = None
+    if uvs is not None:
+        packed_uvs = uvs[0].new_zeros((batch_size, max_vertices, uvs[0].shape[1]))
+        for i, u in enumerate(uvs):
+            assert u.shape[0] == vertices[i].shape[0], (
+                f"uvs[{i}] has {u.shape[0]} entries, expected {vertices[i].shape[0]} (1:1 with vertices)"
+            )
+            packed_uvs[i, :u.shape[0]] = u
+
+    return Types.MESH(packed_vertices, packed_faces,
+                      uvs=packed_uvs, vertex_colors=packed_colors, texture=texture,
+                      vertex_counts=vertex_counts, face_counts=face_counts)
+
+
+def get_mesh_batch_item(mesh, index):
+    # Returns (vertices, faces, colors, uvs) for batch index, slicing to real lengths
+    # if the mesh carries per-item counts (variable-size batch).
+    v_colors = getattr(mesh, "vertex_colors", None)
+    v_uvs = getattr(mesh, "uvs", None)
+    if getattr(mesh, "vertex_counts", None) is not None:
+        vertex_count = int(mesh.vertex_counts[index].item())
+        face_count = int(mesh.face_counts[index].item())
+        vertices = mesh.vertices[index, :vertex_count]
+        faces = mesh.faces[index, :face_count]
+        colors = v_colors[index, :vertex_count] if v_colors is not None else None
+        uvs = v_uvs[index, :vertex_count] if v_uvs is not None else None
+        return vertices, faces, colors, uvs
+
+    colors = v_colors[index] if v_colors is not None else None
+    uvs = v_uvs[index] if v_uvs is not None else None
+    return mesh.vertices[index], mesh.faces[index], colors, uvs
+
+
+def save_glb(vertices, faces, filepath, metadata=None,
+             uvs=None, vertex_colors=None, texture_image=None):
+    """
+    Save PyTorch tensor vertices and faces as a GLB file without external dependencies.
+
+    Parameters:
+    vertices: torch.Tensor of shape (N, 3) - The vertex coordinates
+    faces: torch.Tensor of shape (M, 3) - The face indices (triangle faces)
+    filepath: str - Output filepath (should end with .glb)
+    metadata: dict - Optional asset.extras metadata
+    uvs: torch.Tensor of shape (N, 2) - Optional per-vertex texture coordinates
+    vertex_colors: torch.Tensor of shape (N, 3) or (N, 4) - Optional per-vertex colors in [0, 1]
+    texture_image: PIL.Image - Optional baseColor texture, embedded as PNG
+    """
+
+    # Convert tensors to numpy arrays
+    vertices_np = vertices.cpu().numpy().astype(np.float32)
+    faces_signed = faces.cpu().numpy().astype(np.int64)
+    uvs_np = uvs.cpu().numpy().astype(np.float32) if uvs is not None else None
+    colors_np = vertex_colors.cpu().numpy().astype(np.float32) if vertex_colors is not None else None
+    if colors_np is not None:
+        colors_np = np.clip(colors_np, 0.0, 1.0)
+
+    n_verts = vertices_np.shape[0]
+    if n_verts == 0:
+        raise ValueError("save_glb: vertices is empty")
+    if faces_signed.size > 0:
+        fmin = int(faces_signed.min())
+        fmax = int(faces_signed.max())
+        if fmin < 0 or fmax >= n_verts:
+            raise ValueError(
+                f"save_glb: face index out of range [0, {n_verts}): min={fmin}, max={fmax}"
+            )
+    if uvs_np is not None and uvs_np.shape[0] != n_verts:
+        raise ValueError(
+            f"save_glb: uvs has {uvs_np.shape[0]} entries but vertex count is {n_verts}"
+        )
+    if colors_np is not None and colors_np.shape[0] != n_verts:
+        raise ValueError(
+            f"save_glb: vertex_colors has {colors_np.shape[0]} entries but vertex count is {n_verts}"
+        )
+    faces_np = faces_signed.astype(np.uint32)
+    texture_png_bytes = None
+    if texture_image is not None:
+        buf = BytesIO()
+        texture_image.save(buf, format="PNG")
+        texture_png_bytes = buf.getvalue()
+
+    vertices_buffer = vertices_np.tobytes()
+    indices_buffer = faces_np.tobytes()
+    uvs_buffer = uvs_np.tobytes() if uvs_np is not None else b""
+    colors_buffer = colors_np.tobytes() if colors_np is not None else b""
+    texture_buffer = texture_png_bytes if texture_png_bytes is not None else b""
+
+    def pad_to_4_bytes(buffer):
+        padding_length = (4 - (len(buffer) % 4)) % 4
+        return buffer + b'\x00' * padding_length
+
+    vertices_buffer_padded = pad_to_4_bytes(vertices_buffer)
+    indices_buffer_padded = pad_to_4_bytes(indices_buffer)
+    uvs_buffer_padded = pad_to_4_bytes(uvs_buffer)
+    colors_buffer_padded = pad_to_4_bytes(colors_buffer)
+    texture_buffer_padded = pad_to_4_bytes(texture_buffer)
+
+    buffer_data = b"".join([
+        vertices_buffer_padded,
+        indices_buffer_padded,
+        uvs_buffer_padded,
+        colors_buffer_padded,
+        texture_buffer_padded,
+    ])
+
+    vertices_byte_length = len(vertices_buffer)
+    vertices_byte_offset = 0
+    indices_byte_length = len(indices_buffer)
+    indices_byte_offset = len(vertices_buffer_padded)
+    uvs_byte_offset = indices_byte_offset + len(indices_buffer_padded)
+    colors_byte_offset = uvs_byte_offset + len(uvs_buffer_padded)
+    texture_byte_offset = colors_byte_offset + len(colors_buffer_padded)
+
+    buffer_views = [
+        {
+            "buffer": 0,
+            "byteOffset": vertices_byte_offset,
+            "byteLength": vertices_byte_length,
+            "target": 34962  # ARRAY_BUFFER
+        },
+        {
+            "buffer": 0,
+            "byteOffset": indices_byte_offset,
+            "byteLength": indices_byte_length,
+            "target": 34963  # ELEMENT_ARRAY_BUFFER
+        }
+    ]
+    accessors = [
+        {
+            "bufferView": 0,
+            "byteOffset": 0,
+            "componentType": 5126,  # FLOAT
+            "count": len(vertices_np),
+            "type": "VEC3",
+            "max": vertices_np.max(axis=0).tolist(),
+            "min": vertices_np.min(axis=0).tolist()
+        },
+        {
+            "bufferView": 1,
+            "byteOffset": 0,
+            "componentType": 5125,  # UNSIGNED_INT
+            "count": faces_np.size,
+            "type": "SCALAR"
+        }
+    ]
+    primitive_attributes = {"POSITION": 0}
+
+    if uvs_np is not None and len(uvs_np) > 0:
+        buffer_views.append({
+            "buffer": 0,
+            "byteOffset": uvs_byte_offset,
+            "byteLength": len(uvs_buffer),
+            "target": 34962
+        })
+        accessor_idx = len(accessors)
+        accessors.append({
+            "bufferView": len(buffer_views) - 1,
+            "byteOffset": 0,
+            "componentType": 5126,
+            "count": len(uvs_np),
+            "type": "VEC2",
+        })
+        primitive_attributes["TEXCOORD_0"] = accessor_idx
+
+    if colors_np is not None and len(colors_np) > 0:
+        buffer_views.append({
+            "buffer": 0,
+            "byteOffset": colors_byte_offset,
+            "byteLength": len(colors_buffer),
+            "target": 34962
+        })
+        accessor_idx = len(accessors)
+        accessors.append({
+            "bufferView": len(buffer_views) - 1,
+            "byteOffset": 0,
+            "componentType": 5126,
+            "count": len(colors_np),
+            "type": "VEC3" if colors_np.shape[1] == 3 else "VEC4",
+        })
+        primitive_attributes["COLOR_0"] = accessor_idx
+
+    primitive = {
+        "attributes": primitive_attributes,
+        "indices": 1,
+        "mode": 4  # TRIANGLES
+    }
+
+    images = []
+    textures = []
+    samplers = []
+    materials = []
+    if texture_png_bytes is not None and "TEXCOORD_0" in primitive_attributes:
+        buffer_views.append({
+            "buffer": 0,
+            "byteOffset": texture_byte_offset,
+            "byteLength": len(texture_buffer),
+        })
+        images.append({"bufferView": len(buffer_views) - 1, "mimeType": "image/png"})
+        samplers.append({"magFilter": 9729, "minFilter": 9729, "wrapS": 33071, "wrapT": 33071})
+        textures.append({"source": 0, "sampler": 0})
+        materials.append({
+            "pbrMetallicRoughness": {
+                "baseColorTexture": {"index": 0, "texCoord": 0},
+                "metallicFactor": 0.0,
+                "roughnessFactor": 1.0,
+            },
+            "doubleSided": True,
+        })
+        primitive["material"] = 0
+
+    gltf = {
+        "asset": {"version": "2.0", "generator": "ComfyUI"},
+        "buffers": [{"byteLength": len(buffer_data)}],
+        "bufferViews": buffer_views,
+        "accessors": accessors,
+        "meshes": [{"primitives": [primitive]}],
+        "nodes": [{"mesh": 0}],
+        "scenes": [{"nodes": [0]}],
+        "scene": 0,
+    }
+    if images:
+        gltf["images"] = images
+    if samplers:
+        gltf["samplers"] = samplers
+    if textures:
+        gltf["textures"] = textures
+    if materials:
+        gltf["materials"] = materials
+
+    if metadata:
+        gltf["asset"]["extras"] = metadata
+
+    # Convert the JSON to bytes
+    gltf_json = json.dumps(gltf).encode('utf8')
+
+    def pad_json_to_4_bytes(buffer):
+        padding_length = (4 - (len(buffer) % 4)) % 4
+        return buffer + b' ' * padding_length
+
+    gltf_json_padded = pad_json_to_4_bytes(gltf_json)
+
+    # Create the GLB header (a 4-byte ASCII magic identifier glTF)
+    glb_header = struct.pack('<4sII', b'glTF', 2, 12 + 8 + len(gltf_json_padded) + 8 + len(buffer_data))
+
+    # Create JSON chunk header (chunk type 0)
+    json_chunk_header = struct.pack('<II', len(gltf_json_padded), 0x4E4F534A)  # "JSON" in little endian
+
+    # Create BIN chunk header (chunk type 1)
+    bin_chunk_header = struct.pack('<II', len(buffer_data), 0x004E4942)  # "BIN\0" in little endian
+
+    # Write the GLB file
+    with open(filepath, 'wb') as f:
+        f.write(glb_header)
+        f.write(json_chunk_header)
+        f.write(gltf_json_padded)
+        f.write(bin_chunk_header)
+        f.write(buffer_data)
+
+    return filepath
+
+
+class SaveGLB(IO.ComfyNode):
+    @classmethod
+    def define_schema(cls):
+        return IO.Schema(
+            node_id="SaveGLB",
+            display_name="Save 3D Model",
+            search_aliases=["export 3d model", "save mesh"],
+            category="3d",
+            essentials_category="Basics",
+            is_output_node=True,
+            inputs=[
+                IO.MultiType.Input(
+                    IO.Mesh.Input("mesh"),
+                    types=[
+                        IO.File3DGLB,
+                        IO.File3DGLTF,
+                        IO.File3DOBJ,
+                        IO.File3DFBX,
+                        IO.File3DSTL,
+                        IO.File3DUSDZ,
+                        IO.File3DAny,
+                    ],
+                    tooltip="Mesh or 3D file to save",
+                ),
+                IO.String.Input("filename_prefix", default="3d/ComfyUI"),
+            ],
+            hidden=[IO.Hidden.prompt, IO.Hidden.extra_pnginfo]
+        )
+
+    @classmethod
+    def execute(cls, mesh: Types.MESH | Types.File3D, filename_prefix: str) -> IO.NodeOutput:
+        full_output_folder, filename, counter, subfolder, filename_prefix = folder_paths.get_save_image_path(filename_prefix, folder_paths.get_output_directory())
+        results = []
+
+        metadata = {}
+        if not args.disable_metadata:
+            if cls.hidden.prompt is not None:
+                metadata["prompt"] = json.dumps(cls.hidden.prompt)
+            if cls.hidden.extra_pnginfo is not None:
+                for x in cls.hidden.extra_pnginfo:
+                    metadata[x] = json.dumps(cls.hidden.extra_pnginfo[x])
+
+        if isinstance(mesh, Types.File3D):
+            # Handle File3D input - save BytesIO data to output folder
+            ext = mesh.format or "glb"
+            f = f"{filename}_{counter:05}_.{ext}"
+            mesh.save_to(os.path.join(full_output_folder, f))
+            results.append({
+                "filename": f,
+                "subfolder": subfolder,
+                "type": "output"
+            })
+            counter += 1
+        else:
+            # Handle Mesh input - save vertices and faces as GLB; carry optional UVs / colors / texture.
+            texture_b = getattr(mesh, "texture", None)
+            texture_np = None
+            if texture_b is not None:
+                texture_np = (texture_b.clamp(0.0, 1.0).cpu().numpy() * 255).astype(np.uint8)
+                assert texture_np.ndim == 4 and texture_np.shape[-1] == 3, (
+                    f"texture must be (B, H, W, 3) RGB, got shape {tuple(texture_np.shape)}"
+                )
+            for i in range(mesh.vertices.shape[0]):
+                vertices_i, faces_i, v_colors, uvs_i = get_mesh_batch_item(mesh, i)
+                if vertices_i.shape[0] == 0 or faces_i.shape[0] == 0:
+                    logging.warning(f"SaveGLB: skipping empty mesh at batch index {i}")
+                    continue
+                tex_img = Image.fromarray(texture_np[i], mode="RGB") if texture_np is not None else None
+                f = f"{filename}_{counter:05}_.glb"
+                save_glb(vertices_i, faces_i, os.path.join(full_output_folder, f), metadata,
+                         uvs=uvs_i,
+                         vertex_colors=v_colors,
+                         texture_image=tex_img)
+                results.append({
+                    "filename": f,
+                    "subfolder": subfolder,
+                    "type": "output"
+                })
+                counter += 1
+        return IO.NodeOutput(ui={"3d": results})
+
+
+class Save3DExtension(ComfyExtension):
+    @override
+    async def get_node_list(self) -> list[type[IO.ComfyNode]]:
+        return [SaveGLB]
+
+
+async def comfy_entrypoint() -> Save3DExtension:
+    return Save3DExtension()

comfy_extras/nodes_video.py

@@ -123,6 +123,7 @@ class CreateVideo(io.ComfyNode):
             search_aliases=["images to video"],
             display_name="Create Video",
             category="video",
+            essentials_category="Video Tools",
             description="Create a video from images.",
             inputs=[
                 io.Image.Input("images", tooltip="The images to create a video from."),

comfyui_version.py

@@ -1,3 +1,3 @@
 # This file is automatically generated by the build process when version is
 # updated in pyproject.toml.
-__version__ = "0.21.0"
+__version__ = "0.21.1"

nodes.py

@@ -2436,6 +2436,7 @@ async def init_builtin_extra_nodes():
         "nodes_void.py",
         "nodes_wandancer.py",
         "nodes_hidream_o1.py",
+        "nodes_save_3d.py",
     ]
 
     import_failed = []

openapi.yaml

@@ -2071,7 +2071,6 @@ paths:
                     type: integer
                     description: Number of assets marked as missing
 
-
   # ===========================================================================
   # Cloud-runtime FE-facing operations
   #
@@ -2122,7 +2121,11 @@ paths:
       operationId: getCloudJobStatus
       tags: [queue]
       summary: Get status of a cloud job
-      description: "[cloud-only] Returns the current execution status of a cloud job."
+      deprecated: true
+      description: |
+        **Deprecated.** This endpoint is superseded by `GET /api/jobs/{job_id}`.
+        Clients should migrate; the endpoint is retained for backward
+        compatibility but will be removed in a future release.
       x-runtime: [cloud]
       parameters:
         - name: job_id
@@ -2192,7 +2195,11 @@ paths:
       operationId: getHistoryV2
       tags: [history]
       summary: Get paginated execution history (v2)
-      description: "[cloud-only] Returns a paginated list of execution history entries in the v2 format, with richer metadata than the legacy history endpoint."
+      deprecated: true
+      description: |
+        **Deprecated.** This endpoint is superseded by `GET /api/jobs`.
+        Clients should migrate; the endpoint is retained for backward
+        compatibility but will be removed in a future release.
       x-runtime: [cloud]
       parameters:
         - name: limit
@@ -2231,7 +2238,11 @@ paths:
       operationId: getHistoryV2ByPromptId
       tags: [history]
       summary: Get v2 history for a specific prompt
-      description: "[cloud-only] Returns the v2 history entry for a specific prompt execution."
+      deprecated: true
+      description: |
+        **Deprecated.** This endpoint is superseded by `GET /api/jobs/{prompt_id}`.
+        Clients should migrate; the endpoint is retained for backward
+        compatibility but will be removed in a future release.
       x-runtime: [cloud]
       parameters:
         - name: prompt_id
@@ -2266,7 +2277,12 @@ paths:
       operationId: getCloudLogs
       tags: [system]
       summary: Get cloud execution logs
-      description: "[cloud-only] Returns execution logs for the authenticated user's cloud jobs."
+      deprecated: true
+      description: |
+        **Deprecated.** This endpoint returns a static placeholder response and
+        provides no real log data. It is retained only to avoid breaking clients
+        that still call it. Clients should remove their dependency; the endpoint
+        will be removed in a future release.
       x-runtime: [cloud]
       parameters:
         - name: job_id
@@ -5370,7 +5386,12 @@ paths:
       operationId: viewVideo
       tags: [view]
       summary: View or download a video file
-      description: "[cloud-only] Serves a video file from the output directory. Used by the frontend video player."
+      deprecated: true
+      description: |
+        **Deprecated.** This endpoint is an alias of `GET /api/view` added for
+        legacy history-queue video playback. Callers should use `/api/view`
+        directly; the endpoint is retained for backward compatibility but will
+        be removed in a future release.
       x-runtime: [cloud]
       parameters:
         - name: filename
@@ -5523,7 +5544,6 @@ paths:
               schema:
                 $ref: "#/components/schemas/CloudError"
 
-
 components:
   parameters:
     ComfyUserHeader:
@@ -6875,7 +6895,6 @@ components:
         error:
           type: string
 
-
     # -------------------------------------------------------------------
     # Cloud-runtime schemas
     #

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "ComfyUI"
-version = "0.21.0"
+version = "0.21.1"
 readme = "README.md"
 license = { file = "LICENSE" }
 requires-python = ">=3.10"

requirements.txt

@@ -1,6 +1,6 @@
 comfyui-frontend-package==1.43.18
-comfyui-workflow-templates==0.9.73
-comfyui-embedded-docs==0.4.4
+comfyui-workflow-templates==0.9.75
+comfyui-embedded-docs==0.5.0
 torch
 torchsde
 torchvision