Four ways to choose frames
| Mode | What it does | Best for |
|---|---|---|
| Every N seconds | Takes a frame at a fixed interval | Time-lapse views, long recordings, datasets |
| Number of frames | Takes a fixed number of frames, evenly spaced | Storyboards and quick overviews |
| At each scene change | Finds each cut and takes the first frame of each shot | Edited videos, ads, trailers, highlights |
| Keyframes only | Takes the frames that the video stores as complete pictures | Fast previews of very long files |
Scene change mode uses the same method as our free scene detection tool. To see how often a video has keyframes, use the keyframe checker.
Frames for AI and computer vision
Frames are the input for most image models. Teams use extracted frames to:
- Label data for object detection and classification.
- Show a video to an AI chat app. For ChatGPT or Claude, a grid of frames with timestamps works better than single images. See Video to AI.
- Make thumbnails and storyboards for a video library.
For a large library, frame extraction is only the first step. The video understanding API turns each shot into structured data, and shot detection explains how cuts are found.
Image formats and sizes
- PNG keeps every pixel. Files are large.
- JPEG is the most compatible format. A quality of 85 to 92 looks the same as the original for most videos.
- WebP gives smaller files than JPEG at the same quality.
The tool never makes a frame larger than the video. If you choose 1920 px for a 1280 px video, you get 1280 px frames.