ComfyUI-vid2frames

ComfyUI custom node for selecting one useful still frame from a decoded video frame batch.

Node

• AIA_SelectVideoFrame / AIA Select Video Frame

The node accepts a batched ComfyUI IMAGE tensor and a selector string, then returns exactly one IMAGE frame.

Supported selector targets:

• empty / unknown / best still -> best still-frame scoring
• no face -> lowest face score while still preferring usable frames
• blurry -> blurriest reasonable frame
• moving -> highest-motion frame with blur/light penalties
• static -> lowest-motion sharp frame
• flicker / lighting -> strongest lighting fluctuation

Metrics

• Sharpness: Laplacian variance
• Motion: neighbor frame pixel delta, smoothed over a 5-frame window
• Lighting fluctuation: luminance delta, smoothed over a 5-frame window
• Face score: MediaPipe face detection first, OpenCV Haar fallback
• Edge penalty: first/last 8% of frames are penalized for longer videos

Install

Clone into ComfyUI custom nodes:

cd /home/ComfyUI/custom_nodes
git clone https://github.com/Pmobilee/ComfyUI-vid2frames.git
pip install -r ComfyUI-vid2frames/requirements.txt

Restart ComfyUI after installation.

Workflow Use

Wire decoded video frames into AIA_SelectVideoFrame, pass selector text such as moving or no face, then connect the output to SaveImage.