ComfyUI-Wudd

V1 is disabled by default. Use the wudd_v3 branch / ComfyUI-Wudd-V3 package for active nodes. To temporarily load legacy V1 nodes, set WUDD_ENABLE_V1=1 before starting ComfyUI.

Custom nodes for ComfyUI, focused on practical image, text, audio/video, path, and API utility workflows.

Installation

Clone this repository into ComfyUI/custom_nodes/:

cd ComfyUI/custom_nodes
git clone https://github.com/qwqgong-ui/Comfyui-Wudd.git

Install dependencies with the Python environment used by ComfyUI:

python -m pip install -r Comfyui-Wudd/requirements.txt

For the ComfyUI Windows portable build, use its embedded Python:

C:\path\to\ComfyUI\python_embeded\python.exe -m pip install -r C:\path\to\ComfyUI\ComfyUI\custom_nodes\Comfyui-Wudd\requirements.txt

Restart ComfyUI after installation.

Dependencies

requirements.txt explicitly lists:

• numpy
• Pillow
• scipy
• av
• imageio-ffmpeg
• playwright

imageio-ffmpeg provides an internal ffmpeg executable for the video and audio-video nodes. The nodes resolve ffmpeg automatically and do not require a path widget.

playwright is used only by the ChatGPT browser automation node. The node connects to Chrome/Edge through the local Chrome DevTools Protocol and can use your installed browser, so a separate playwright install browser download is not required for the default workflow.

If you need an offline/local override, place the Windows ffmpeg executables in this repository's bin/ folder:

• bin/ffmpeg.exe
• bin/ffprobe.exe
• bin/ffplay.exe

On Linux, place the no-extension executables in the same folder:

• bin/ffmpeg
• bin/ffprobe
• bin/ffplay

The node uses the local bin/ ffmpeg set only when all three platform executables exist. The bin/ folder is kept in Git with .gitkeep, while copied executables are ignored by .gitignore and should not be committed. If the local bin/ set is missing or incomplete, the node falls back to imageio-ffmpeg, then to ffmpeg on PATH.

To refresh the local Windows ffmpeg override, run:

powershell -ExecutionPolicy Bypass -File .\scripts\update_ffmpeg.ps1

To refresh the local Linux ffmpeg override, run:

bash scripts/update_ffmpeg_linux.sh

The scripts download BtbN's latest x64 GPL autobuild by default, extract the three ffmpeg executables, and place them in bin/. The executables remain ignored by Git.

For Linux Jpegli support, run:

bash scripts/update_jpegli_linux.sh

This downloads the newest available jxl-linux-x86_64-static release asset from libjxl/libjxl and places cjpegli in bin/. The image save node checks bin/cjpegli, then the bundled Windows Jpegli tools, then cjpegli on PATH.

Nodes

All nodes are registered under Wudd Nodes.

Wudd Multi Save

Batch save one or more IMAGE inputs.

• Dynamic image input ports.
• Save as PNG or Jpegli JPEG.
• Supports append and overwrite naming modes.
• Preserves ComfyUI workflow metadata in PNG output.
• Jpegli uses the bundled jxl-x64-windows-static/bin/cjpegli.exe on Windows and falls back to PIL JPEG if unavailable.
• Saved images are backed up in temp/image/wudd_save_cache before copying to the final output path.

Inputs:

• image_1, dynamic extra image inputs
• filename_prefix
• save_mode: append, overwrite
• extension: png, jpegli
• quality
• progressive
• enable_xyb
• chroma_subsampling

Wudd Save Video

Save one or more VIDEO inputs to the ComfyUI output directory.

• Dynamic video input ports.
• Encodes video as AV1 or H.265/HEVC through ffmpeg.
• Supports CPU encoding plus NVIDIA NVENC, Intel Quick Sync, and AMD AMF when the resolved ffmpeg build and local hardware support the selected encoder.
• Supports append and overwrite naming modes matching Wudd Multi Save.
• Audio can be copied, re-encoded to AAC with the source sample rate/channels, or removed.
• ffmpeg is resolved internally from a complete local bin/ set, imageio-ffmpeg, or system PATH.

Inputs:

• video_1, dynamic extra video inputs
• filename_prefix
• save_mode: append, overwrite
• codec: av1, h265
• encoder: cpu, nvidia, intel, amd
• container: mp4, mkv
• crf
• preset: fast, medium, slow
• audio_mode: copy, aac, none

Wudd Video Fast Forward

Speed up a VIDEO input and output a new VIDEO.

• speed_multiplier mode uses a direct playback multiplier such as 2.0 for 2x speed.
• target_seconds mode derives the speed from the source duration so the result lasts the requested number of seconds.
• Video is written as MP4 for downstream video nodes.
• Audio can be tempo-adjusted with the video and re-encoded as AAC, or removed.

Inputs:

• video
• mode: speed_multiplier, target_seconds
• speed_multiplier
• target_seconds
• audio_mode: keep, none

Wudd Concat Videos

Concatenate VIDEO inputs in port order and output a new VIDEO.

• Dynamic video input ports.
• Concatenates as video_1, video_2, video_3, and so on.
• Normalizes every segment to the first video's dimensions and frame rate before concatenating.
• Uses FFV1 video and PCM float audio in MKV for lossless internal processing.
• Returns an MP4 video encoded for downstream nodes that require an MP4 container.
• fit_to_first keeps aspect ratio with padding; stretch_to_first fills the first video's size.
• Audio can be kept with PCM normalization and silence for mute clips, or removed.
• Normalized segments and MP4 outputs are backed up in temp/video/wudd_concat_cache.
• The output stays in ComfyUI temp storage and can be connected to Wudd Save Video.

Inputs:

• video_1, video_2, dynamic extra video inputs
• resize_mode: fit_to_first, stretch_to_first
• audio_mode: keep, none

Wudd Drop Alpha

Composite transparent image areas over a generated background and return RGB IMAGE.

• Optional MASK input where ComfyUI mask value 1 means transparent.
• Background modes: checkerboard or fill color.
• Optional auto crop around non-transparent content.

Inputs:

• image
• optional mask
• mode: checkerboard, fill_color
• fill_color
• tile_size
• auto_crop
• padding

Wudd Image Expand

Expand an image by whole-image blocks in one direction.

• Directions: right, down, left, up.
• Fill modes match Wudd Drop Alpha: checkerboard or solid color.
• Outputs expanded image, width, and height.

Wudd Edge Pad

Create natural vertical edge padding for panorama or long-image preprocessing.

• Supports up to 16 image inputs.
• Cross-blends neighboring image edges.
• Adds controllable blur and chamfering at pad junctions.
• Returns up to 16 padded image outputs.

Wudd Image List Importer

Load multiple images from ComfyUI input/.

• File mode: choose individual uploaded/input images.
• Folder mode: load images from a folder path.
• Supports up to 50 outputs.
• Folder paths may be absolute or relative to ComfyUI input/.

Wudd Image Stitch

Stitch multiple images linearly.

• Directions: right, down, left, up.
• Supports up to 16 image inputs.
• Fits images to the first image's height or width depending on stitch direction.
• Optional gap between images.

Wudd Text Splitter

Extract one line from a multiline string.

• Zero-based index.
• Optional skip_empty filtering.
• Returns empty string when out of range.

Wudd Multi Text Splitter

Split multiline text into up to 16 string outputs.

• count controls how many output slots are intended for use.
• Optional skip_empty.
• Unused or missing lines return empty strings.

Wudd Path Joiner

Join up to 5 path segments with /.

• Ignores blank segments.
• Useful for building portable path-like prompt strings or API inputs.

Wudd ChatGPT Browser

Submit a prompt and optional IMAGE to chatgpt.com through a local Chrome/Edge browser, then return the latest assistant text and response images to ComfyUI.

• Uses your own browser session; no ChatGPT password or API key is handled by the node.
• Default mode connects to http://127.0.0.1:9222 if a CDP browser is already running, otherwise launches Edge with a persistent profile under ComfyUI user/wudd_browser_profiles/.
• Uploads the first input image as a temporary PNG, fills the ChatGPT composer, submits with Enter by default, and waits for the response text to stabilize.
• Exports images from the latest assistant response as a ComfyUI IMAGE batch. When the response contains no images, the node returns one 1x1 black placeholder image and image_count = 0.
• Returns:
  • text
  • conversation_url
  • images
  • image_count

Inputs:

• prompt
• optional image
• connection_mode: connect_or_launch_edge, connect_or_launch_chrome, connect_cdp, launch_chrome, launch_edge
• new_chat
• submit_action: press_enter, click_send_button
• advanced chatgpt_url, cdp_url, wait_timeout_seconds, stable_seconds, upload_wait_seconds, keep_browser_open, browser_executable, user_data_dir

To connect to an already-running Edge or Chrome profile, start the browser with remote debugging enabled, for example:

msedge.exe --remote-debugging-port=9222 --user-data-dir="$env:USERPROFILE\.wudd-chatgpt-browser"

Then set connection_mode to connect_cdp. If ChatGPT asks you to log in, complete the login in the opened browser and run the node again.

Wudd Group Switch

Enable or disable all nodes inside a ComfyUI canvas group from one switch node.

• Leave group_name empty to automatically add toggles for every canvas group.
• Set group_name to self, current, or auto to control only the group containing this node.
• Or fill group_name with an exact group title; the node also has right-click menu helpers for choosing a target group.
• enabled sets all listed groups on or off when group_name is empty.
• Each generated group toggle can independently turn one canvas group on or off.
• off_mode: mute sets group nodes to Never so they do not run; bypass sets them to Bypass.
• Other Wudd Group Switch nodes inside the group stay active so a switch placed inside its own group can re-enable it.

Wudd Wireless Input / Output

Virtual wireless routing nodes for hiding long links while keeping ComfyUI's execution graph connected.

• Wudd Wireless Input has real input slots and publishes multiple channels from one node.
• Wudd Wireless Output has real output slots and exposes matching channels from one node.
• Channels are matched by namespace + channel name.
• Input channel names are auto-renamed with _1, _2, and so on when another input publisher in the same namespace already uses the name.
• Use Wudd: Create matching output on an input node to create a receiver with the same namespace and channels.
• Use Wudd: Refresh from inputs on an output node to rebuild its channel list from existing input nodes.
• The nodes are frontend virtual nodes; they resolve to the original upstream links when ComfyUI builds the prompt.

Wudd Extract Audio From Video

Extract audio directly from a VIDEO input.

• Connect the output of ComfyUI's built-in Load Video node.
• No file picker or path input is required.
• Supports audio_stream_index for videos with multiple audio streams.
• Outputs:
  • AUDIO
  • sample_rate
  • duration_seconds

Wudd Replace Video Audio

Replace a VIDEO input's audio track with a supplied AUDIO input.

• Video stream is copied with ffmpeg where possible.
• New audio is encoded as AAC.
• Output duration follows the video duration: long audio is trimmed, short audio is padded with silence.
• ffmpeg is resolved internally from a complete local bin/ set, imageio-ffmpeg, or system PATH.
• Outputs a new VIDEO object that can be passed to ComfyUI's built-in Save Video.

Inputs:

• video
• audio
• output_format: mp4, mkv, mov
• audio_bitrate: 128k, 192k, 256k, 320k
• end_mode: keep_video_length (shortest is accepted for legacy workflows and behaves the same)

Wudd OpenAI GPT-5.4

Call an OpenAI-compatible API endpoint using only Python standard HTTP libraries.

• Supports Responses API and Chat Completions API modes.
• Optional image input.
• Supports custom base_url.
• Supports SSL verification toggle.
• Responses mode supports polling by response id.

Inputs include:

• prompt
• instructions
• api_key
• base_url
• model
• api_mode
• reasoning_effort
• verbosity
• max_output_tokens
• poll_interval
• max_wait_seconds
• optional images

Outputs:

• text
• response_id

Release Notes

2026-05-28

• Added Chinese locale files for ComfyUI node definitions and UI strings.
• Added OpenRouter-compatible GPT, Claude, and Gemini text nodes plus GPT and Gemini image nodes.
• Added Wudd Video Fast Forward for multiplier-based or target-duration video speed changes.
• Expanded video output support with AV1/H.265 CPU encoding plus NVIDIA NVENC, Intel Quick Sync, and AMD AMF encoder options.
• Updated Wudd Concat Videos to return MP4 output for better downstream and Topaz compatibility.
• Added backup caching for image saves and concatenated video intermediates/outputs.
• Simplified Wudd Save Video audio handling by removing the separate audio bitrate UI and preserving source audio settings where appropriate.

Notes

• PNG image saves include metadata for workflow restoration.
• Jpegli is Windows-bundled in this repository; other platforms fall back to PIL JPEG.
• AV1/H.265 video preview support depends on the browser and operating system codec support.
• Video concatenation normalizes intermediate segments as FFV1/PCM in MKV for lossless internal processing, then downstream save nodes can encode final AV1/H.265 outputs.
• Video audio replacement keeps the output video in ComfyUI's temp directory until it is consumed by downstream nodes.
• ComfyUI itself provides torch, folder_paths, and the VIDEO object model used by these nodes.