A ComfyUI custom node that brings a DAW-style interactive video timeline directly into the node graph. Upload any video, scrub through it in real time, drag a loop region to define your extraction window, and pipe the resulting frame batch into any downstream node.
- Interactive timeline — filmstrip thumbnails with correct aspect ratio (including portrait video), brightness waveform, and a zoomable timecode ruler all rendered directly inside the node
- Loop region selection — drag the in/out handles or the region body to define start and end frames; the video preview seeks live as you drag
- Resizable video preview — drag the resize handle to make the preview pane as large as you need
- Collapsible preview section — hide the video preview to save canvas space while keeping the timeline accessible
- Transport controls — Play / Pause toggle, frame-step ◀ ▶ buttons, and ⏮ ⏭ Jump buttons that advance the loop window by its own span for stepping through non-overlapping clips
- Live loading status — progress bar, buffering percentage, elapsed time, and health indicators (stalled / error / timeout) shown while a video loads
- Memory-safe execution — live peak-memory estimate in the info bar updates as you adjust the loop region and
target_fps; the backend refuses to run clips that would exceed available RAM, so ComfyUI never freezes on an oversized extraction - Zoom and pan — scroll wheel to zoom the timeline, Shift + scroll to pan horizontally
- Full outputs — 10 output pins: forward frames, reversed frames, first and last frame, filename prefix, dimensions, original FPS, custom-FPS resampled frame batch, and target FPS passthrough
- Open ComfyUI Manager → Custom Nodes → Search
- Search for
Video Frame Extractor - Click Install and restart ComfyUI
cd ComfyUI/custom_nodes
git clone https://github.com/comfyuiattic-989/ComfyUI-Video-Frame-Extractor
cd ComfyUI-Video-Frame-Extractor
pip install -r requirements.txtRestart ComfyUI after installation.
| Package | Version |
|---|---|
| Python | ≥ 3.9 |
| opencv-python | ≥ 4.7 |
| Pillow | ≥ 9.0 |
| numpy | ≥ 1.22 |
| psutil | ≥ 5.9 (optional) |
| torch | (provided by ComfyUI) |
psutil is optional — when installed, the memory-safety check auto-tunes its limit to 75 % of your system's available RAM. Without it, the limit falls back to a fixed 8 GB.
- Place a video file in
ComfyUI/input/or upload one directly via the Choose Video to Upload button on the node - Add the Video Frame Extractor 🎞️ node to your graph (right-click canvas → Add Node → video)
- Select your video from the dropdown — the timeline populates automatically
- Drag the indigo loop handles to set the extraction window, or drag the loop body to pan the whole window
- Connect the output pins to your workflow
| Interaction | Action |
|---|---|
| Drag loop handle (◀ ▶) | Resize the loop window |
| Drag loop body | Slide the window without changing its span |
| ▶ / ⏸ button | Play or pause the preview within the loop |
| ◀ / ▶ step buttons | Shift the entire loop window one frame |
| ⏮ / ⏭ jump buttons | Jump the loop window backward/forward by its own span |
| Scroll wheel | Zoom the timeline centred on the cursor |
| Shift + scroll | Pan the timeline left / right |
| Drag resize bar | Adjust the height of the video preview |
| Click collapse toggle | Show / hide the video preview and transport |
| Name | Type | Description |
|---|---|---|
video |
File picker | MP4, AVI, MOV, MKV, or WebM file — upload via Choose Video to Upload button |
start_frame |
INT | Loop region start (auto-set by timeline) |
end_frame |
INT | Loop region end (auto-set by timeline) |
num_frames |
INT | Auto-computed as end − start + 1; hidden on the node, synced internally |
target_fps |
FLOAT | Target frame rate for the Clipped Frames at Target FPS output (default: 24.0) |
| Pin name | Type | Description |
|---|---|---|
Clipped Frames |
IMAGE | Batch of extracted frames, forward order (N, H, W, 3) |
Reversed Clipped Frames |
IMAGE | Same batch in reverse order — useful for ping-pong effects |
First Frame |
IMAGE | Single frame at the loop start (1, H, W, 3) |
Last Frame |
IMAGE | Single frame at the loop end (1, H, W, 3) |
Filename Prefix (no extension) |
STRING | Video filename without extension, e.g. my_clip |
Width |
INT | Video width in pixels |
Height |
INT | Video height in pixels |
Original FPS |
FLOAT | Source frame rate of the video file |
Clipped Frames at Target FPS |
IMAGE | Frame batch resampled to target_fps — count computed automatically from clip duration × target FPS |
Target FPS |
FLOAT | Passes through the target_fps input value for use downstream |
Video Frame Extractor → VAE Encode → KSampler
Feed a batch of video frames directly into a KSampler for video-to-video workflows.
Video Frame Extractor (frames) ──────────────→ ┐
Batch Concat → Video Combine
Video Frame Extractor (frames_reversed) ──────→ ┘
Concatenate forward and reversed frame batches to create a seamless loop.
Video Frame Extractor (filename_prefix) → Save Image (filename_prefix input)
Automatically name saved frames after the source video.
| Format | Extension |
|---|---|
| MP4 / H.264 | .mp4 |
| AVI | .avi |
| QuickTime | .mov |
| Matroska | .mkv |
| WebM | .webm |
Any format supported by your OpenCV build will also work.
- Initial release
- DAW-style timeline with filmstrip thumbnails and brightness waveform
- Loop region with drag handles and live video scrubbing
- Resizable / collapsible video preview
- Play / Pause toggle with frame-step buttons
- 8 output pins:
frames,frames_reversed,first_frame,last_frame,filename_prefix,width,height,fps - Zoom and pan on the timeline
start_frame/end_framespinners sync bidirectionally with the loop region
- Reduced video file opens from 6 → 2 per execution
- Halved peak memory during frame extraction by pre-allocating output buffer
- first_frame / last_frame now sliced from loaded batch instead of re-read from disk
- Fixed BytesIO buffer in thumbnail endpoint now explicitly closed after use
- Added
target_fpsinput andframes_at_fpsoutput — extracts a frame batch resampled to any target frame rate, independent ofnum_frames
- Reliable video loading — previous video is now fully stopped, unloaded, and its memory released before a new file begins loading; playing/looping videos no longer block new loads from completing
- Loading progress bar — the preview area shows a live progress bar (
Loading video… N%) while the video buffers, an animated shimmer for indeterminate states (metadata fetch), and detailed status lines (elapsed time, buffered/total seconds, HTML5readyState) - Load health indicators — stalled downloads turn the bar amber with a "Stalled at N%" label; network errors, decode failures, and unsupported formats surface a red error overlay with the specific error reason; a 30-second watchdog flags videos that never become playable
- Memory estimation & safety guard — the info bar now shows a live peak-memory estimate (
✓ MEM est. X GB / Y GB limit) that turns amber above 75% and red at the limit; the backend independently blocks execution with a clearRuntimeErrorbefore allocation to prevent ComfyUI from freezing on oversized clips. The limit adapts to 75 % of available system RAM whenpsutilis installed, falling back to 8 GB otherwise - Single-output execution fix — the node is now marked
OUTPUT_NODE = True, so workflows that only connect one pin (e.g. justClipped Frames at Target FPS) no longer fail with "Prompt has no outputs" - Jump buttons — ⏮ JUMP BACK and ⏭ JUMP FWD added to the transport bar; each press shifts the loop window by its own span so you can step through non-overlapping clips; buttons dim and disable when no full jump fits in that direction
- Dynamic filmstrip height — the filmstrip now scales its height based on the actual video aspect ratio so portrait footage is no longer squished; thumbnails also use the real aspect ratio instead of assuming 16:9
- Improved info bar — loop summary text is now larger (bold 13 px), white, and easier to read at a glance; a second row shows the memory estimate and the
N+M fr @ WxHframe breakdown - Cleaner output labels — all 10 output pins renamed to plain English:
Clipped Frames,Reversed Clipped Frames,First Frame,Last Frame,Filename Prefix (no extension),Width,Height,Original FPS,Clipped Frames at Target FPS,Target FPS - Target FPS output — added a 10th output pin that passes through the
target_fpsinput value for use downstream - Upload button label — renamed from "choose file to upload" to "Choose Video to Upload"
- Hidden num_frames widget — the auto-computed
num_framesfield is no longer shown on the node; its value is still synced internally so the backend receives the correct frame count
MIT © 2026