[Partner Nodes] hash video via memoryview to avoid memory copy

Signed-off-by: bigcat88 <bigcat88@icloud.com>

.github/workflows/backport_release.yaml

@@ -458,6 +458,41 @@ jobs:
 
           echo "Released ${NEW_VERSION} on ${RELEASE_BRANCH}."
 
+      - name: Delete remote source branch
+        env:
+          GH_TOKEN:        ${{ steps.app-token.outputs.token }}
+          REPO:            ${{ github.repository }}
+          SOURCE_BRANCH:   ${{ steps.resolve.outputs.source_branch }}
+          SOURCE_COMMIT:   ${{ inputs.commit }}
+          RELEASE_BRANCH:  ${{ steps.latest.outputs.release_branch }}
+          DEFAULT_BRANCH:  ${{ github.event.repository.default_branch }}
+        run: |
+          set -euo pipefail
+
+          # Belt-and-braces: the resolve step already refuses the default branch,
+          # but never delete the default or the release branch under any
+          # circumstances.
+          if [[ "${SOURCE_BRANCH}" == "${DEFAULT_BRANCH}" || "${SOURCE_BRANCH}" == "${RELEASE_BRANCH}" ]]; then
+            echo "::error::Refusing to delete '${SOURCE_BRANCH}' (matches default or release branch)."
+            exit 1
+          fi
+
+          # Delete the source branch on origin, but only if its tip is still the
+          # SHA we released from. If someone pushed new commits to it after we
+          # resolved it, leave it alone — those commits would be silently lost.
+          current_tip="$(git ls-remote origin "refs/heads/${SOURCE_BRANCH}" | awk '{print $1}')"
+          if [[ -z "${current_tip}" ]]; then
+            echo "Source branch '${SOURCE_BRANCH}' no longer exists on origin; nothing to delete."
+            exit 0
+          fi
+          if [[ "${current_tip}" != "${SOURCE_COMMIT}" ]]; then
+            echo "::warning::Source branch '${SOURCE_BRANCH}' tip (${current_tip}) no longer matches released commit (${SOURCE_COMMIT}). Leaving it in place."
+            exit 0
+          fi
+
+          git push origin --delete "refs/heads/${SOURCE_BRANCH}"
+          echo "Deleted remote branch '${SOURCE_BRANCH}'."
+
       - name: Summary
         if: always()
         env:

README.md

@@ -433,7 +433,7 @@ See also: [https://www.comfy.org/](https://www.comfy.org/)
 
 ## Frontend Development
 
-As of August 15, 2024, we have transitioned to a new frontend, which is now hosted in a separate repository: [ComfyUI Frontend](https://github.com/Comfy-Org/ComfyUI_frontend). This repository now hosts the compiled JS (from TS/Vue) under the `web/` directory.
+As of August 15, 2024, we have transitioned to a new frontend, which is now hosted in a separate repository: [ComfyUI Frontend](https://github.com/Comfy-Org/ComfyUI_frontend). The compiled JS files (from TS/Vue) are published to [pypi](https://pypi.org/project/comfyui-frontend-package) and installed as a dependency in ComfyUI.
 
 ### Reporting Issues and Requesting Features
 

app/logger.py

@@ -5,6 +5,40 @@ import logging
 import sys
 import threading
 
+ANSI_NAMED_COLORS = {
+    'black':   '\033[30m',
+    'red':     '\033[31m',
+    'green':   '\033[32m',
+    'yellow':  '\033[33m',
+    'blue':    '\033[34m',
+    'magenta': '\033[35m',
+    'cyan':    '\033[36m',
+    'white':   '\033[37m',
+}
+
+ANSI_LEVEL_COLORS = {
+    'DEBUG':    ANSI_NAMED_COLORS['cyan'],
+    'INFO':     ANSI_NAMED_COLORS['green'],
+    'WARNING':  ANSI_NAMED_COLORS['yellow'],
+    'ERROR':    ANSI_NAMED_COLORS['red'],
+    'CRITICAL': ANSI_NAMED_COLORS['magenta'],
+}
+
+ANSI_RESET = '\033[0m'
+ANSI_BOLD  = '\033[1m'
+
+
+class ColoredFormatter(logging.Formatter):
+    def format(self, record):
+        color = ANSI_LEVEL_COLORS.get(record.levelname, '')
+        bold  = ANSI_BOLD if record.levelno >= logging.WARNING else ''
+        level_tag = f"{bold}{color}[{record.levelname}]{ANSI_RESET} "
+        message = super().format(record)
+        line_color = ANSI_NAMED_COLORS.get(getattr(record, 'color', ''), '')
+        if line_color:
+            return f"{level_tag}{line_color}{message}{ANSI_RESET}"
+        return level_tag + message
+
 logs = None
 stdout_interceptor = None
 stderr_interceptor = None
@@ -68,8 +102,10 @@ def setup_logger(log_level: str = 'INFO', capacity: int = 300, use_stdout: bool
     logger = logging.getLogger()
     logger.setLevel(log_level)
 
+    formatter = ColoredFormatter("%(message)s")
+
     stream_handler = logging.StreamHandler()
-    stream_handler.setFormatter(logging.Formatter("%(message)s"))
+    stream_handler.setFormatter(formatter)
 
     if use_stdout:
         # Only errors and critical to stderr
@@ -77,7 +113,7 @@ def setup_logger(log_level: str = 'INFO', capacity: int = 300, use_stdout: bool
 
         # Lesser to stdout
         stdout_handler = logging.StreamHandler(sys.stdout)
-        stdout_handler.setFormatter(logging.Formatter("%(message)s"))
+        stdout_handler.setFormatter(formatter)
         stdout_handler.addFilter(lambda record: record.levelno < logging.ERROR)
         logger.addHandler(stdout_handler)
 

comfy/cli_args.py

@@ -111,7 +111,7 @@ parser.add_argument("--preview-method", type=LatentPreviewMethod, default=Latent
 parser.add_argument("--preview-size", type=int, default=512, help="Sets the maximum preview size for sampler nodes.")
 
 cache_group = parser.add_mutually_exclusive_group()
-cache_group.add_argument("--cache-ram", nargs='*', type=float, default=[], metavar="GB", help="Use RAM pressure caching with the specified headroom thresholds. This is the default caching mode. The first value sets the active-cache threshold; the optional second value sets the inactive-cache/pin threshold. Defaults when no values are provided: active 25%% of system RAM (min 4GB, max 32GB), inactive 75%% of system RAM (min 12GB, max 96GB).")
+cache_group.add_argument("--cache-ram", nargs='*', type=float, default=[], metavar="GB", help="Use RAM pressure caching with the specified headroom thresholds. This is the default caching mode. The first value sets the active-cache threshold; the optional second value sets the inactive-cache/pin threshold. Defaults when no values are provided: active 10%% of system RAM (min 2GB, max 10GB), inactive 100%% of system RAM (max 96GB).")
 cache_group.add_argument("--cache-classic", action="store_true", help="Use the old style (aggressive) caching.")
 cache_group.add_argument("--cache-lru", type=int, default=0, help="Use LRU caching with a maximum of N node results cached. May use more RAM/VRAM.")
 cache_group.add_argument("--cache-none", action="store_true", help="Reduced RAM/VRAM usage at the expense of executing every node for each run.")

comfy/ldm/modules/attention.py

@@ -741,12 +741,12 @@ optimized_attention = attention_basic
 if model_management.sage_attention_enabled():
     logging.info("Using sage attention")
     optimized_attention = attention_sage
-elif model_management.xformers_enabled():
-    logging.info("Using xformers attention")
-    optimized_attention = attention_xformers
 elif model_management.flash_attention_enabled():
     logging.info("Using Flash Attention")
     optimized_attention = attention_flash
+elif model_management.xformers_enabled():
+    logging.info("Using xformers attention")
+    optimized_attention = attention_xformers
 elif model_management.pytorch_attention_enabled():
     logging.info("Using pytorch attention")
     optimized_attention = attention_pytorch

comfy/model_management.py

@@ -1217,7 +1217,7 @@ def get_aimdo_cast_buffer(offload_stream, device):
 def get_pin_buffer(offload_stream):
     pin_buffer = STREAM_PIN_BUFFERS.get(offload_stream, None)
     if pin_buffer is None:
-        pin_buffer = comfy_aimdo.host_buffer.HostBuffer(0, 0, pinned_hostbuf_size(8 * 1024**3))
+        pin_buffer = comfy_aimdo.host_buffer.HostBuffer(0, 0, pinned_hostbuf_size(8 * 1024**3), mark_cold=False)
         STREAM_PIN_BUFFERS[offload_stream] = pin_buffer
     elif offload_stream is not None:
         event = getattr(pin_buffer, "_comfy_event", None)

comfy/samplers.py

@@ -265,7 +265,6 @@ def _calc_cond_batch(model: BaseModel, conds: list[list[dict]], x_in: torch.Tens
                 input_shape = [len(batch_amount) * first_shape[0]] + list(first_shape)[1:]
                 cond_shapes = collections.defaultdict(list)
                 for tt in batch_amount:
-                    cond = {k: v.size() for k, v in to_run[tt][0].conditioning.items()}
                     for k, v in to_run[tt][0].conditioning.items():
                         cond_shapes[k].append(v.size())
 

comfy_api_nodes/apis/bytedance.py

@@ -158,8 +158,9 @@ class SeedanceCreateAssetResponse(BaseModel):
 
 
 class SeedanceVirtualLibraryCreateAssetRequest(BaseModel):
-    url: str = Field(..., description="Publicly accessible URL of the image asset to upload.")
+    url: str = Field(..., description="Publicly accessible URL of the asset to upload.")
     hash: str = Field(..., description="Dedup key. Re-submitting the same hash returns the existing asset id.")
+    asset_type: str | None = Field(None, description="BytePlus asset type. Defaults to Image server-side when omitted.")
 
 
 # Dollars per 1K tokens, keyed by (model_id, has_video_input).

comfy_api_nodes/apis/rodin.py

@@ -1,7 +1,5 @@
-from __future__ import annotations
-
 from enum import Enum
-from typing import Optional, List
+
 from pydantic import BaseModel, Field
 
 
@@ -11,44 +9,76 @@ class Rodin3DGenerateRequest(BaseModel):
     material: str = Field(..., description="The material type.")
     quality_override: int = Field(..., description="The poly count of the mesh.")
     mesh_mode: str = Field(..., description="It controls the type of faces of generated models.")
-    TAPose: Optional[bool] = Field(None, description="")
+    TAPose: bool | None = Field(None, description="")
+
+
+class Rodin3DGen25Request(BaseModel):
+
+    tier: str = Field(..., description="Gen-2.5 tier (e.g. Gen-2.5-High).")
+    prompt: str | None = Field(None, description="Required for Text-to-3D; ignored otherwise.")
+    seed: int | None = Field(None, description="0-65535.")
+    material: str | None = Field(None, description="PBR | Shaded | All | None.")
+    geometry_file_format: str | None = Field(None, description="glb | usdz | fbx | obj | stl.")
+    texture_mode: str | None = Field(None, description="legacy | extreme-low | low | medium | high.")
+    mesh_mode: str | None = Field(None, description="Raw (triangular) | Quad.")
+    quality_override: int | None = Field(None, description="Mesh face count override.")
+    geometry_instruct_mode: str | None = Field(None, description="faithful | creative.")
+    bbox_condition: list[int] | None = Field(None, description="Bounding box [Width(Y), Height(Z), Length(X)] in cm.")
+    height: int | None = Field(None, description="Approximate model height in cm.")
+    TAPose: bool | None = Field(None, description="T/A pose for human-like models.")
+    hd_texture: bool | None = Field(None, description="Enhanced texture quality.")
+    texture_delight: bool | None = Field(None, description="Remove baked lighting from textures.")
+    is_micro: bool | None = Field(None, description="Micro detail (Extreme-High only).")
+    use_original_alpha: bool | None = Field(None, description="Preserve image transparency.")
+    preview_render: bool | None = Field(None, description="Generate high-quality preview render.")
+    addons: list[str] | None = Field(None, description='Optional addons, e.g. ["HighPack"].')
+
 
 class GenerateJobsData(BaseModel):
-    uuids: List[str] = Field(..., description="str LIST")
+    uuids: list[str] = Field(..., description="str LIST")
     subscription_key: str = Field(..., description="subscription key")
 
+
 class Rodin3DGenerateResponse(BaseModel):
-    message: Optional[str] = Field(None, description="Return message.")
-    prompt: Optional[str] = Field(None, description="Generated Prompt from image.")
-    submit_time: Optional[str] = Field(None, description="Submit Time")
-    uuid: Optional[str] = Field(None, description="Task str")
-    jobs: Optional[GenerateJobsData] = Field(None, description="Details of jobs")
+    message: str | None = Field(None, description="Return message.")
+    prompt: str | None = Field(None, description="Generated Prompt from image.")
+    submit_time: str | None = Field(None, description="Submit Time")
+    uuid: str | None = Field(None, description="Task str")
+    jobs: GenerateJobsData | None = Field(None, description="Details of jobs")
+
 
 class JobStatus(str, Enum):
     """
     Status for jobs
     """
+
     Done = "Done"
     Failed = "Failed"
     Generating = "Generating"
     Waiting = "Waiting"
 
+
 class Rodin3DCheckStatusRequest(BaseModel):
     subscription_key: str = Field(..., description="subscription from generate endpoint")
 
+
 class JobItem(BaseModel):
     uuid: str = Field(..., description="uuid")
-    status: JobStatus = Field(...,description="Status Currently")
+    status: JobStatus = Field(..., description="Status Currently")
+
 
 class Rodin3DCheckStatusResponse(BaseModel):
-    jobs: List[JobItem] = Field(..., description="Job status List")
+    jobs: list[JobItem] = Field(..., description="Job status List")
+
 
 class Rodin3DDownloadRequest(BaseModel):
     task_uuid: str = Field(..., description="Task str")
 
+
 class RodinResourceItem(BaseModel):
     url: str = Field(..., description="Download Url")
     name: str = Field(..., description="File name with ext")
 
+
 class Rodin3DDownloadResponse(BaseModel):
-    list: List[RodinResourceItem] = Field(..., description="Source List")
+    items: list[RodinResourceItem] = Field(..., alias="list", description="Source List")

comfy_api_nodes/nodes_bytedance.py

@@ -2,11 +2,12 @@ import hashlib
 import logging
 import math
 import re
+from io import BytesIO
 
 import torch
 from typing_extensions import override
 
-from comfy_api.latest import IO, ComfyExtension, Input
+from comfy_api.latest import IO, ComfyExtension, Input, Types
 from comfy_api_nodes.apis.bytedance import (
     RECOMMENDED_PRESETS,
     RECOMMENDED_PRESETS_SEEDREAM_4,
@@ -308,6 +309,26 @@ async def _seedance_virtual_library_upload_image_asset(
     return f"asset://{create_resp.asset_id}"
 
 
+async def _seedance_virtual_library_upload_video_asset(
+    cls: type[IO.ComfyNode],
+    video: Input.Video,
+    *,
+    wait_label: str = "Uploading video",
+) -> str:
+    buf = BytesIO()
+    video.save_to(buf, format=Types.VideoContainer.MP4, codec=Types.VideoCodec.H264)
+    video_hash = hashlib.sha256(buf.getbuffer()).hexdigest()
+    public_url = await upload_video_to_comfyapi(cls, video, wait_label=wait_label)
+    create_resp = await sync_op(
+        cls,
+        ApiEndpoint(path="/proxy/seedance/virtual-library/assets", method="POST"),
+        response_model=SeedanceCreateAssetResponse,
+        data=SeedanceVirtualLibraryCreateAssetRequest(url=public_url, hash=video_hash, asset_type="Video"),
+    )
+    await _wait_for_asset_active(cls, create_resp.asset_id, group_id="virtual-library")
+    return f"asset://{create_resp.asset_id}"
+
+
 def _seedance2_price_extractor(model_id: str, has_video_input: bool):
     """Returns a price_extractor closure for Seedance 2.0 poll_op."""
     rate = SEEDANCE2_PRICE_PER_1K_TOKENS.get((model_id, has_video_input))
@@ -2106,7 +2127,7 @@ class ByteDance2ReferenceNode(IO.ComfyNode):
             content.append(
                 TaskVideoContent(
                     video_url=TaskVideoContentUrl(
-                        url=await upload_video_to_comfyapi(
+                        url=await _seedance_virtual_library_upload_video_asset(
                             cls,
                             reference_videos[key],
                             wait_label=f"Uploading video {i}",

comfy_api_nodes/nodes_rodin.py

@@ -5,32 +5,37 @@ Rodin API docs: https://developer.hyper3d.ai/
 
 """
 
-from inspect import cleandoc
-import folder_paths as comfy_paths
-import os
 import logging
 import math
+import os
+from inspect import cleandoc
 from io import BytesIO
-from typing_extensions import override
+from typing import Any
+
+import aiohttp
 from PIL import Image
+from typing_extensions import override
+
+import folder_paths as comfy_paths
+from comfy_api.latest import IO, ComfyExtension, Types
 from comfy_api_nodes.apis.rodin import (
-    Rodin3DGenerateRequest,
-    Rodin3DGenerateResponse,
+    JobStatus,
     Rodin3DCheckStatusRequest,
     Rodin3DCheckStatusResponse,
     Rodin3DDownloadRequest,
     Rodin3DDownloadResponse,
-    JobStatus,
+    Rodin3DGen25Request,
+    Rodin3DGenerateRequest,
+    Rodin3DGenerateResponse,
 )
 from comfy_api_nodes.util import (
-    sync_op,
-    poll_op,
     ApiEndpoint,
     download_url_to_bytesio,
     download_url_to_file_3d,
+    poll_op,
+    sync_op,
+    validate_string,
 )
-from comfy_api.latest import ComfyExtension, IO, Types
-
 
 COMMON_PARAMETERS = [
     IO.Int.Input(
@@ -51,40 +56,30 @@ COMMON_PARAMETERS = [
 ]
 
 
-def get_quality_mode(poly_count):
-    polycount = poly_count.split("-")
-    poly = polycount[1]
-    count = polycount[0]
-    if poly == "Triangle":
-        mesh_mode = "Raw"
-    elif poly == "Quad":
-        mesh_mode = "Quad"
-    else:
-        mesh_mode = "Quad"
-
-    if count == "4K":
-        quality_override = 4000
-    elif count == "8K":
-        quality_override = 8000
-    elif count == "18K":
-        quality_override = 18000
-    elif count == "50K":
-        quality_override = 50000
-    elif count == "2K":
-        quality_override = 2000
-    elif count == "20K":
-        quality_override = 20000
-    elif count == "150K":
-        quality_override = 150000
-    elif count == "500K":
-        quality_override = 500000
-    else:
-        quality_override = 18000
-
-    return mesh_mode, quality_override
+_QUALITY_MESH_OPTIONS: dict[str, tuple[str, int]] = {
+    "4K-Quad":       ("Quad", 4000),
+    "8K-Quad":       ("Quad", 8000),
+    "18K-Quad":      ("Quad", 18000),
+    "50K-Quad":      ("Quad", 50000),
+    "200K-Quad":     ("Quad", 200000),
+    "2K-Triangle":   ("Raw", 2000),
+    "20K-Triangle":  ("Raw", 20000),
+    "150K-Triangle": ("Raw", 150000),
+    "200K-Triangle": ("Raw", 200000),
+    "500K-Triangle": ("Raw", 500000),
+    "1M-Triangle":   ("Raw", 1000000),
+}
 
 
-def tensor_to_filelike(tensor, max_pixels: int = 2048*2048):
+def get_quality_mode(poly_count: str) -> tuple[str, int]:
+    """Map a polygon-count preset like '18K-Quad' to (mesh_mode, quality_override).
+
+    Falls back to ('Quad', 18000) for unknown labels; legacy parity.
+    """
+    return _QUALITY_MESH_OPTIONS.get(poly_count, ("Quad", 18000))
+
+
+def tensor_to_filelike(tensor, max_pixels: int = 2048 * 2048):
     """
     Converts a PyTorch tensor to a file-like object.
 
@@ -96,8 +91,8 @@ def tensor_to_filelike(tensor, max_pixels: int = 2048*2048):
     - io.BytesIO: A file-like object containing the image data.
     """
     array = tensor.cpu().numpy()
-    array = (array * 255).astype('uint8')
-    image = Image.fromarray(array, 'RGB')
+    array = (array * 255).astype("uint8")
+    image = Image.fromarray(array, "RGB")
 
     original_width, original_height = image.size
     original_pixels = original_width * original_height
@@ -112,7 +107,7 @@ def tensor_to_filelike(tensor, max_pixels: int = 2048*2048):
         image = image.resize((new_width, new_height), Image.Resampling.LANCZOS)
 
     img_byte_arr = BytesIO()
-    image.save(img_byte_arr, format='PNG')  # PNG is used for lossless compression
+    image.save(img_byte_arr, format="PNG")  # PNG is used for lossless compression
     img_byte_arr.seek(0)
     return img_byte_arr
 
@@ -145,11 +140,9 @@ async def create_generate_task(
             TAPose=ta_pose,
         ),
         files=[
-            (
-                "images",
-                open(image, "rb") if isinstance(image, str) else tensor_to_filelike(image)
-            )
-            for image in images if image is not None
+            ("images", open(image, "rb") if isinstance(image, str) else tensor_to_filelike(image))
+            for image in images
+            if image is not None
         ],
         content_type="multipart/form-data",
     )
@@ -177,6 +170,7 @@ def check_rodin_status(response: Rodin3DCheckStatusResponse) -> str:
         return "DONE"
     return "Generating"
 
+
 def extract_progress(response: Rodin3DCheckStatusResponse) -> int | None:
     if not response.jobs:
         return None
@@ -214,7 +208,7 @@ async def download_files(url_list, task_uuid: str) -> tuple[str | None, Types.Fi
     model_file_path = None
     file_3d = None
 
-    for i in url_list.list:
+    for i in url_list.items:
         file_path = os.path.join(save_path, i.name)
         if i.name.lower().endswith(".glb"):
             model_file_path = os.path.join(result_folder_name, i.name)
@@ -489,7 +483,16 @@ class Rodin3D_Gen2(IO.ComfyNode):
                 IO.Combo.Input("Material_Type", options=["PBR", "Shaded"], default="PBR", optional=True),
                 IO.Combo.Input(
                     "Polygon_count",
-                    options=["4K-Quad", "8K-Quad", "18K-Quad", "50K-Quad", "2K-Triangle", "20K-Triangle", "150K-Triangle", "500K-Triangle"],
+                    options=[
+                        "4K-Quad",
+                        "8K-Quad",
+                        "18K-Quad",
+                        "50K-Quad",
+                        "2K-Triangle",
+                        "20K-Triangle",
+                        "150K-Triangle",
+                        "500K-Triangle",
+                    ],
                     default="500K-Triangle",
                     optional=True,
                 ),
@@ -542,6 +545,566 @@ class Rodin3D_Gen2(IO.ComfyNode):
         return IO.NodeOutput(model_path, file_3d)
 
 
+def _rodin_multipart_parser(data: dict[str, Any]) -> aiohttp.FormData:
+    """Convert a Rodin request dict to an aiohttp form, fixing bool/list serialization.
+
+    Booleans --> "true"/"false". Lists --> one field per element.
+    """
+    form = aiohttp.FormData(default_to_multipart=True)
+    for key, value in data.items():
+        if value is None:
+            continue
+        if isinstance(value, bool):
+            form.add_field(key, "true" if value else "false")
+        elif isinstance(value, list):
+            for item in value:
+                form.add_field(key, str(item))
+        elif isinstance(value, (bytes, bytearray)):
+            form.add_field(key, value)
+        else:
+            form.add_field(key, str(value))
+    return form
+
+
+async def _create_gen25_task(
+    cls: type[IO.ComfyNode],
+    request: Rodin3DGen25Request,
+    images: list | None,
+) -> tuple[str, str]:
+    """Submit a Gen-2.5 generate job; returns (task_uuid, subscription_key)."""
+
+    if images is not None and len(images) > 5:
+        raise ValueError("Rodin Gen-2.5 supports at most 5 input images.")
+
+    files = None
+    if images:
+        files = [
+            (
+                "images",
+                open(image, "rb") if isinstance(image, str) else tensor_to_filelike(image),
+            )
+            for image in images
+            if image is not None
+        ]
+
+    response = await sync_op(
+        cls,
+        ApiEndpoint(path="/proxy/rodin/api/v2/rodin", method="POST"),
+        response_model=Rodin3DGenerateResponse,
+        data=request,
+        files=files,
+        content_type="multipart/form-data",
+        multipart_parser=_rodin_multipart_parser,
+    )
+
+    if not response.uuid or not response.jobs or not response.jobs.subscription_key:
+        raise RuntimeError(f"Rodin Gen-2.5 submit failed: message={response.message!r}")
+    return response.uuid, response.jobs.subscription_key
+
+
+_PREVIEWABLE_3D_EXTS = {".glb", ".obj", ".fbx", ".stl", ".gltf"}
+
+
+async def _download_gen25_files(
+    download_list: Rodin3DDownloadResponse,
+    task_uuid: str,
+    geometry_file_format: str,
+) -> Types.File3D | None:
+    """Download every file in the list; return the File3D matching the chosen format."""
+
+    folder_name = f"Rodin3D_Gen25_{task_uuid}"
+    save_dir = os.path.join(comfy_paths.get_output_directory(), folder_name)
+    os.makedirs(save_dir, exist_ok=True)
+
+    target_ext = f".{geometry_file_format.lower().lstrip('.')}"
+    file_3d: Types.File3D | None = None
+
+    for item in download_list.items:
+        file_path = os.path.join(save_dir, item.name)
+        ext = os.path.splitext(item.name.lower())[1]
+        # Prefer the file matching the user's chosen format; fall back below.
+        if file_3d is None and ext == target_ext and ext in _PREVIEWABLE_3D_EXTS:
+            file_3d = await download_url_to_file_3d(item.url, target_ext.lstrip("."))
+            with open(file_path, "wb") as f:
+                f.write(file_3d.get_bytes())
+            continue
+        await download_url_to_bytesio(item.url, file_path)
+
+    # If the chosen format wasn't found, surface any model file we did get.
+    if file_3d is None:
+        for item in download_list.items:
+            ext = os.path.splitext(item.name.lower())[1]
+            if ext in _PREVIEWABLE_3D_EXTS:
+                file_3d = await download_url_to_file_3d(item.url, ext.lstrip("."))
+                break
+    return file_3d
+
+
+_MODE_REGULAR = "Regular"
+_MODE_FAST = "Fast"
+_MODE_EXTREME_HIGH = "Extreme-High"
+
+_REGULAR_POLY_OPTIONS = [
+    "Default",
+    "4K-Quad",
+    "8K-Quad",
+    "18K-Quad",
+    "50K-Quad",
+    "2K-Triangle",
+    "20K-Triangle",
+    "150K-Triangle",
+    "500K-Triangle",
+    "1M-Triangle",
+]
+
+_TEXTURE_MODE_OPTIONS = ["Default", "legacy", "extreme-low", "low", "medium", "high"]
+_GEOMETRY_FORMAT_OPTIONS = ["glb", "fbx", "obj", "stl"]
+_MATERIAL_OPTIONS = ["PBR", "Shaded", "All", "None"]
+
+
+def _build_mode_input(name: str = "mode") -> IO.DynamicCombo.Input:
+    return IO.DynamicCombo.Input(
+        name,
+        options=[
+            IO.DynamicCombo.Option(
+                _MODE_REGULAR,
+                [
+                    IO.Combo.Input(
+                        "tier",
+                        options=["Gen-2.5-Low", "Gen-2.5-Medium", "Gen-2.5-High"],
+                        default="Gen-2.5-High",
+                        tooltip="Quality tier. Higher tiers produce higher-fidelity geometry.",
+                    ),
+                    IO.Combo.Input(
+                        "polygon_count",
+                        options=_REGULAR_POLY_OPTIONS,
+                        default="Default",
+                        tooltip="Preset face count. 'Default' uses the server's default for the selected tier.",
+                    ),
+                    IO.Boolean.Input(
+                        "creative",
+                        default=False,
+                        tooltip="Creative mode (Medium/High only). Enhances generative robustness.",
+                    ),
+                ],
+            ),
+            IO.DynamicCombo.Option(
+                _MODE_FAST,
+                [
+                    IO.Combo.Input(
+                        "tier",
+                        options=[
+                            "Gen-2.5-Extreme-Low",
+                            "Gen-2.5-Low",
+                            "Gen-2.5-Medium",
+                            "Gen-2.5-High",
+                        ],
+                        default="Gen-2.5-Low",
+                    ),
+                    IO.Int.Input(
+                        "mesh_faces",
+                        default=20000,
+                        min=1000,
+                        max=20000,
+                        display_mode=IO.NumberDisplay.number,
+                        tooltip="Mesh face count (1K-20K in Fast mode).",
+                    ),
+                ],
+            ),
+            IO.DynamicCombo.Option(
+                _MODE_EXTREME_HIGH,
+                [
+                    IO.Combo.Input("mesh_mode", options=["Raw", "Quad"], default="Raw"),
+                    IO.Int.Input(
+                        "mesh_faces",
+                        default=1000000,
+                        min=20000,
+                        max=2000000,
+                        display_mode=IO.NumberDisplay.number,
+                        tooltip=(
+                            "Mesh face count. Raw mode: 20K-2M. "
+                            "Quad mode: keep under 200K (upstream may reject higher values)."
+                        ),
+                    ),
+                    IO.Boolean.Input(
+                        "is_micro",
+                        default=False,
+                        tooltip="Enable micro detail (Extreme-High only).",
+                    ),
+                    IO.Boolean.Input(
+                        "creative",
+                        default=False,
+                        tooltip="Creative mode. Enhances generative robustness.",
+                    ),
+                ],
+            ),
+        ],
+        tooltip=(
+            "Generation mode. Regular = balanced. Fast = 1K-20K faces for rapid prototyping. "
+            "Extreme-High = 20K-2M faces with optional micro details."
+        ),
+    )
+
+
+def _build_common_inputs(*, include_image_only: bool) -> list:
+    inputs: list = [
+        IO.Combo.Input("material", options=_MATERIAL_OPTIONS, default="Shaded"),
+        IO.Combo.Input("geometry_file_format", options=_GEOMETRY_FORMAT_OPTIONS, default="glb"),
+        IO.Combo.Input(
+            "texture_mode",
+            options=_TEXTURE_MODE_OPTIONS,
+            default="Default",
+            optional=True,
+            tooltip="Texture quality preset. 'Default' uses the server's default for the selected tier.",
+        ),
+        IO.Int.Input(
+            "seed",
+            default=0,
+            min=0,
+            max=65535,
+            display_mode=IO.NumberDisplay.number,
+            control_after_generate=True,
+            optional=True,
+        ),
+        IO.Boolean.Input(
+            "TAPose", default=False, optional=True, advanced=True, tooltip="T/A pose for human-like models."
+        ),
+        IO.Boolean.Input(
+            "hd_texture", default=False, optional=True, advanced=True, tooltip="High-quality texture enhancement."
+        ),
+        IO.Boolean.Input(
+            "texture_delight",
+            default=False,
+            optional=True,
+            advanced=True,
+            tooltip="Remove baked lighting from textures.",
+        ),
+    ]
+    if include_image_only:
+        inputs.append(
+            IO.Boolean.Input(
+                "use_original_alpha",
+                default=False,
+                optional=True,
+                advanced=True,
+                tooltip="Preserve image transparency.",
+            )
+        )
+    inputs.extend(
+        [
+            IO.Boolean.Input(
+                "addon_highpack",
+                default=False,
+                optional=True,
+                advanced=True,
+                tooltip="HighPack addon: 4K textures and ~16x faces in Quad mode.",
+            ),
+            IO.Int.Input(
+                "bbox_width",
+                default=0,
+                min=0,
+                max=300,
+                display_mode=IO.NumberDisplay.number,
+                optional=True,
+                advanced=True,
+                tooltip="Bounding-box width (Y axis). Set to 0 with the others to skip bbox.",
+            ),
+            IO.Int.Input(
+                "bbox_height",
+                default=0,
+                min=0,
+                max=300,
+                display_mode=IO.NumberDisplay.number,
+                optional=True,
+                advanced=True,
+                tooltip="Bounding-box height (Z axis).",
+            ),
+            IO.Int.Input(
+                "bbox_length",
+                default=0,
+                min=0,
+                max=300,
+                display_mode=IO.NumberDisplay.number,
+                optional=True,
+                advanced=True,
+                tooltip="Bounding-box length (X axis).",
+            ),
+            IO.Int.Input(
+                "height_cm",
+                default=0,
+                min=0,
+                max=10000,
+                display_mode=IO.NumberDisplay.number,
+                optional=True,
+                advanced=True,
+                tooltip="Approximate model height in centimeters (0 to skip).",
+            ),
+        ]
+    )
+    return inputs
+
+
+_PRICE_EXPR = """
+(
+  $baseCredits := widgets.mode = "extreme-high" ? 1.0 : 0.5;
+  $addonCredits := widgets.addon_highpack ? 1.0 : 0.0;
+  $total := ($baseCredits * 1.5) + ($addonCredits * 0.8);
+  {"type":"usd","usd": $total}
+)
+"""
+
+
+def _resolve_mode_params(mode_input: dict) -> dict:
+    """Translate the DynamicCombo `mode` payload into Gen-2.5 request fields.
+
+    Returns a dict with: tier, quality_override, mesh_mode, geometry_instruct_mode, is_micro.
+    Missing keys mean "do not send" (so we don't override server defaults).
+    """
+    selected = mode_input["mode"]
+    out: dict = {}
+
+    if selected == _MODE_REGULAR:
+        out["tier"] = mode_input["tier"]
+        polygon = mode_input.get("polygon_count", "Default")
+        if polygon != "Default":
+            mesh_mode, faces = get_quality_mode(polygon)
+            out["mesh_mode"] = mesh_mode
+            out["quality_override"] = faces
+        if mode_input.get("creative"):
+            out["geometry_instruct_mode"] = "creative"
+
+    elif selected == _MODE_FAST:
+        out["tier"] = mode_input["tier"]
+        out["mesh_mode"] = "Raw"
+        out["quality_override"] = int(mode_input["mesh_faces"])
+
+    elif selected == _MODE_EXTREME_HIGH:
+        out["tier"] = "Gen-2.5-Extreme-High"
+        out["mesh_mode"] = mode_input["mesh_mode"]
+        out["quality_override"] = int(mode_input["mesh_faces"])
+        if mode_input.get("is_micro"):
+            out["is_micro"] = True
+        if mode_input.get("creative"):
+            out["geometry_instruct_mode"] = "creative"
+    return out
+
+
+def _build_request(
+    *,
+    mode_input: dict,
+    material: str,
+    geometry_file_format: str,
+    texture_mode: str,
+    seed: int,
+    TAPose: bool,
+    hd_texture: bool,
+    texture_delight: bool,
+    addon_highpack: bool,
+    bbox_width: int,
+    bbox_height: int,
+    bbox_length: int,
+    height_cm: int,
+    prompt: str | None = None,
+    use_original_alpha: bool = False,
+) -> Rodin3DGen25Request:
+    mode_params = _resolve_mode_params(mode_input)
+
+    bbox = None
+    if bbox_width and bbox_height and bbox_length:
+        bbox = [bbox_width, bbox_height, bbox_length]
+
+    return Rodin3DGen25Request(
+        tier=mode_params["tier"],
+        prompt=prompt or None,
+        seed=seed,
+        material=material,
+        geometry_file_format=geometry_file_format,
+        texture_mode=None if texture_mode == "Default" else texture_mode,
+        mesh_mode=mode_params.get("mesh_mode"),
+        quality_override=mode_params.get("quality_override"),
+        geometry_instruct_mode=mode_params.get("geometry_instruct_mode"),
+        bbox_condition=bbox,
+        height=height_cm or None,
+        TAPose=TAPose or None,
+        hd_texture=hd_texture or None,
+        texture_delight=texture_delight or None,
+        is_micro=mode_params.get("is_micro"),
+        use_original_alpha=use_original_alpha or None,
+        addons=["HighPack"] if addon_highpack else None,
+    )
+
+
+class Rodin3D_Gen25_Image(IO.ComfyNode):
+
+    @classmethod
+    def define_schema(cls) -> IO.Schema:
+        return IO.Schema(
+            node_id="Rodin3D_Gen25_Image",
+            display_name="Rodin 3D Gen-2.5 - Image to 3D",
+            category="api node/3d/Rodin",
+            description=(
+                "Generate a 3D model from 1-5 reference images via Rodin Gen-2.5. "
+                "Pick a mode (Fast / Regular / Extreme-High) to tune quality vs. cost."
+            ),
+            inputs=[
+                IO.Autogrow.Input(
+                    "images",
+                    template=IO.Autogrow.TemplatePrefix(IO.Image.Input("image"), prefix="image", min=1, max=5),
+                    tooltip="1-5 images. The first image is used for materials when multi-view.",
+                ),
+                _build_mode_input(),
+                *_build_common_inputs(include_image_only=True),
+            ],
+            outputs=[IO.File3DAny.Output(display_name="model_file")],
+            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=["mode", "addon_highpack"]),
+                expr=_PRICE_EXPR,
+            ),
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        images: IO.Autogrow.Type,
+        mode: dict,
+        material: str,
+        geometry_file_format: str,
+        texture_mode: str,
+        seed: int,
+        TAPose: bool,
+        hd_texture: bool,
+        texture_delight: bool,
+        use_original_alpha: bool,
+        addon_highpack: bool,
+        bbox_width: int,
+        bbox_height: int,
+        bbox_length: int,
+        height_cm: int,
+    ) -> IO.NodeOutput:
+        image_tensors = [img for img in images.values() if img is not None]
+        if not image_tensors:
+            raise ValueError("Rodin Gen-2.5 Image-to-3D requires at least one image.")
+
+        # Flatten multi-image tensors into individual frames; the API accepts each as a separate part.
+        flat_images: list = []
+        for tensor in image_tensors:
+            if hasattr(tensor, "shape") and len(tensor.shape) == 4:
+                for i in range(tensor.shape[0]):
+                    flat_images.append(tensor[i])
+            else:
+                flat_images.append(tensor)
+
+        if len(flat_images) > 5:
+            raise ValueError(f"Rodin Gen-2.5 accepts at most 5 images; received {len(flat_images)}.")
+
+        request = _build_request(
+            mode_input=mode,
+            material=material,
+            geometry_file_format=geometry_file_format,
+            texture_mode=texture_mode,
+            seed=seed,
+            TAPose=TAPose,
+            hd_texture=hd_texture,
+            texture_delight=texture_delight,
+            addon_highpack=addon_highpack,
+            bbox_width=bbox_width,
+            bbox_height=bbox_height,
+            bbox_length=bbox_length,
+            height_cm=height_cm,
+            prompt=None,
+            use_original_alpha=use_original_alpha,
+        )
+
+        task_uuid, subscription_key = await _create_gen25_task(cls, request, flat_images)
+        await poll_for_task_status(subscription_key, cls)
+        download_list = await get_rodin_download_list(task_uuid, cls)
+        file_3d = await _download_gen25_files(download_list, task_uuid, geometry_file_format)
+        return IO.NodeOutput(file_3d)
+
+
+class Rodin3D_Gen25_Text(IO.ComfyNode):
+
+    @classmethod
+    def define_schema(cls) -> IO.Schema:
+        return IO.Schema(
+            node_id="Rodin3D_Gen25_Text",
+            display_name="Rodin 3D Gen-2.5 - Text to 3D",
+            category="api node/3d/Rodin",
+            description=(
+                "Generate a 3D model from a text prompt via Rodin Gen-2.5. "
+                "Pick a mode (Fast / Regular / Extreme-High) to tune quality vs. cost."
+            ),
+            inputs=[
+                IO.String.Input(
+                    "prompt",
+                    multiline=True,
+                    default="",
+                    tooltip="Text prompt for the 3D model.",
+                ),
+                _build_mode_input(),
+                *_build_common_inputs(include_image_only=False),
+            ],
+            outputs=[IO.File3DAny.Output(display_name="model_file")],
+            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=["mode", "addon_highpack"]),
+                expr=_PRICE_EXPR,
+            ),
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        prompt: str,
+        mode: dict,
+        material: str,
+        geometry_file_format: str,
+        texture_mode: str,
+        seed: int,
+        TAPose: bool,
+        hd_texture: bool,
+        texture_delight: bool,
+        addon_highpack: bool,
+        bbox_width: int,
+        bbox_height: int,
+        bbox_length: int,
+        height_cm: int,
+    ) -> IO.NodeOutput:
+        validate_string(prompt, field_name="prompt", min_length=1, max_length=2500)
+        request = _build_request(
+            mode_input=mode,
+            material=material,
+            geometry_file_format=geometry_file_format,
+            texture_mode=texture_mode,
+            seed=seed,
+            TAPose=TAPose,
+            hd_texture=hd_texture,
+            texture_delight=texture_delight,
+            addon_highpack=addon_highpack,
+            bbox_width=bbox_width,
+            bbox_height=bbox_height,
+            bbox_length=bbox_length,
+            height_cm=height_cm,
+            prompt=prompt,
+        )
+        task_uuid, subscription_key = await _create_gen25_task(cls, request, images=None)
+        await poll_for_task_status(subscription_key, cls)
+        download_list = await get_rodin_download_list(task_uuid, cls)
+        file_3d = await _download_gen25_files(download_list, task_uuid, geometry_file_format)
+        return IO.NodeOutput(file_3d)
+
+
 class Rodin3DExtension(ComfyExtension):
     @override
     async def get_node_list(self) -> list[type[IO.ComfyNode]]:
@@ -551,6 +1114,8 @@ class Rodin3DExtension(ComfyExtension):
             Rodin3D_Smooth,
             Rodin3D_Sketch,
             Rodin3D_Gen2,
+            Rodin3D_Gen25_Image,
+            Rodin3D_Gen25_Text,
         ]
 
 

comfy_extras/nodes_images.py

@@ -3,15 +3,23 @@ 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.
@@ -835,6 +843,405 @@ 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(
+                    "format",
+                    options=[
+                        IO.DynamicCombo.Option("png", [
+                            IO.Combo.Input("bit_depth", options=["8-bit", "16-bit"],
+                                           default="8-bit", advanced=True),
+                            IO.Combo.Input("input_color_space", 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(
+                                "input_color_space",
+                                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, format: dict) -> IO.NodeOutput:
+        file_format = format["format"]
+        bit_depth = format["bit_depth"]
+        colorspace = format.get("input_color_space", "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]]:
@@ -847,6 +1254,7 @@ class ImagesExtension(ComfyExtension):
             ImageAddNoise,
             SaveAnimatedWEBP,
             SaveAnimatedPNG,
+            SaveImageAdvanced,
             SaveSVGNode,
             ImageStitch,
             ResizeAndPadImage,

comfy_extras/nodes_logic.py

@@ -91,7 +91,7 @@ class SwitchNode(io.ComfyNode):
         return io.Schema(
             node_id="ComfySwitchNode",
             display_name="Switch",
-            category="logic",
+            category="utils/logic",
             is_experimental=True,
             inputs=[
                 io.Boolean.Input("switch"),
@@ -122,7 +122,7 @@ class SoftSwitchNode(io.ComfyNode):
         return io.Schema(
             node_id="ComfySoftSwitchNode",
             display_name="Soft Switch",
-            category="logic",
+            category="utils/logic",
             is_experimental=True,
             inputs=[
                 io.Boolean.Input("switch"),
@@ -212,7 +212,7 @@ class DCTestNode(io.ComfyNode):
         return io.Schema(
             node_id="DCTestNode",
             display_name="DCTest",
-            category="logic",
+            category="utils/logic",
             is_output_node=True,
             inputs=[io.DynamicCombo.Input("combo", options=[
                 io.DynamicCombo.Option("option1", [io.String.Input("string")]),
@@ -250,7 +250,7 @@ class AutogrowNamesTestNode(io.ComfyNode):
         return io.Schema(
             node_id="AutogrowNamesTestNode",
             display_name="AutogrowNamesTest",
-            category="logic",
+            category="utils/logic",
             inputs=[
                 _io.Autogrow.Input("autogrow", template=template)
             ],
@@ -270,7 +270,7 @@ class AutogrowPrefixTestNode(io.ComfyNode):
         return io.Schema(
             node_id="AutogrowPrefixTestNode",
             display_name="AutogrowPrefixTest",
-            category="logic",
+            category="utils/logic",
             inputs=[
                 _io.Autogrow.Input("autogrow", template=template)
             ],
@@ -289,7 +289,7 @@ class ComboOutputTestNode(io.ComfyNode):
         return io.Schema(
             node_id="ComboOptionTestNode",
             display_name="ComboOptionTest",
-            category="logic",
+            category="utils/logic",
             inputs=[io.Combo.Input("combo", options=["option1", "option2", "option3"]),
                     io.Combo.Input("combo2", options=["option4", "option5", "option6"])],
             outputs=[io.Combo.Output(), io.Combo.Output()],
@@ -306,7 +306,7 @@ class ConvertStringToComboNode(io.ComfyNode):
             node_id="ConvertStringToComboNode",
             search_aliases=["string to dropdown", "text to combo"],
             display_name="Convert String to Combo",
-            category="logic",
+            category="utils/logic",
             inputs=[io.String.Input("string")],
             outputs=[io.Combo.Output()],
         )
@@ -322,7 +322,7 @@ class InvertBooleanNode(io.ComfyNode):
             node_id="InvertBooleanNode",
             search_aliases=["not", "toggle", "negate", "flip boolean"],
             display_name="Invert Boolean",
-            category="logic",
+            category="utils/logic",
             inputs=[io.Boolean.Input("boolean")],
             outputs=[io.Boolean.Output()],
         )

comfy_extras/nodes_math.py

@@ -70,7 +70,7 @@ class MathExpressionNode(io.ComfyNode):
         return io.Schema(
             node_id="ComfyMathExpression",
             display_name="Math Expression",
-            category="logic",
+            category="utils",
             search_aliases=[
                 "expression", "formula", "calculate", "calculator",
                 "eval", "math",

comfy_extras/nodes_toolkit.py

@@ -14,7 +14,7 @@ class CreateList(io.ComfyNode):
         return io.Schema(
             node_id="CreateList",
             display_name="Create List",
-            category="logic",
+            category="utils",
             is_input_list=True,
             search_aliases=["Image Iterator", "Text Iterator", "Iterator"],
             inputs=[io.Autogrow.Input("inputs", template=template_autogrow)],

main.py

@@ -286,8 +286,8 @@ def prompt_worker(q, server_instance):
     cache_ram = 0
     cache_ram_inactive = 0
     if not args.cache_classic and not args.cache_none and args.cache_lru <= 0:
-        cache_ram = min(32.0, max(4.0, comfy.model_management.total_ram * 0.25 / 1024.0))
-        cache_ram_inactive = min(96.0, max(12.0, comfy.model_management.total_ram * 0.75 / 1024.0))
+        cache_ram = min(10.0, max(2.0, comfy.model_management.total_ram * 0.10 / 1024.0))
+        cache_ram_inactive = min(96.0, comfy.model_management.total_ram / 1024.0)
         if len(args.cache_ram) > 0:
             cache_ram = args.cache_ram[0]
         if len(args.cache_ram) > 1:
@@ -344,9 +344,9 @@ def prompt_worker(q, server_instance):
             # Log Time in a more readable way after 10 minutes
             if execution_time > 600:
                 execution_time = time.strftime("%H:%M:%S", time.gmtime(execution_time))
-                logging.info(f"Prompt executed in {execution_time}")
+                logging.info(f"Prompt executed in {execution_time}", extra={'color': 'green'})
             else:
-                logging.info("Prompt executed in {:.2f} seconds".format(execution_time))
+                logging.info("Prompt executed in {:.2f} seconds".format(execution_time), extra={'color': 'green'})
 
             if not asset_seeder.is_disabled():
                 paths = _collect_output_absolute_paths(e.history_result)

requirements.txt

@@ -1,4 +1,4 @@
-comfyui-frontend-package==1.43.18
+comfyui-frontend-package==1.44.19
 comfyui-workflow-templates==0.9.82
 comfyui-embedded-docs==0.5.0
 torch
@@ -23,7 +23,7 @@ SQLAlchemy>=2.0.0
 filelock
 av>=14.2.0
 comfy-kitchen>=0.2.8
-comfy-aimdo==0.4.3
+comfy-aimdo==0.4.5
 requests
 simpleeval>=1.0.0
 blake3