Privacy First
Video decoding and image encoding run in your browser. Your video is not uploaded.
Local Video Frame Extractor
Extract frames from video and save separate JPG, PNG, or WebP images in a ZIP. This free video to image converter runs in your browser; your video is not uploaded. Capture one still, sample at an interval, or remove consecutive duplicates. Start with MP4 or MOV; codec support differs by capture mode.
Up to 600 samples per run. Repeated images are kept. These modes use browser seeking, not the HDR/software decoder. Timestamps are requested positions, not exact frame times.
Ready.
Error or unexpected result? See troubleshooting
We made a 20-second test screen recording: 1280 × 720 H.264 MP4 at 30 fps. It shows a cursor moving for 3 seconds, a 5-second idle pause, a 4-second upload progress bar, a 6-second pause, and a final 2-second cursor move. We ran it through this tool in desktop Chrome with the default settings: Remove consecutive duplicates, threshold 2, JPEG quality 0.92, original width.
The cursor movement in the first 3 seconds and the last 2 seconds did not produce extra images. A small cursor changes the perceptual hash very little, so threshold 2 treated those frames as duplicates. Use One frame every N seconds when you need images at regular times, whatever changes on screen. Results can differ by browser and decoder.
Need all samples in one image? Use the Video Contact Sheet Generator. For a shorter playable video, use the Video Condenser. Read the processing and privacy limits before you start.
Technical reference: WebCodecs API. Browser codec support varies by device.
We made six synthetic screen recordings with FFmpeg: 1280 × 720 H.264 MP4 at 30 fps. Each one tests one kind of change. We ran them on 31 October 2026 in desktop Chrome 156 on macOS, with the default settings: Remove consecutive duplicates, threshold 2, JPEG. These are six designed cases, not a measure of how often the tool works on real recordings.
| Recording | What changes | Frames decoded | Images kept | What we saw |
|---|---|---|---|---|
| cursor-move (10 s) | A small cursor moves for 6 seconds | 300 of 300 | 1 | No image shows the cursor movement. |
| small-change (10 s) | A 40 × 12 pixel status badge changes colour every 2 seconds | 300 of 300 | 1 | No badge change was kept. |
| colour-change (10 s) | A large card changes fill colour every 2 seconds; the layout stays the same | 300 of 300 | 2 | Only the first of four colour changes was kept, at 2.0 s. |
| fast-motion (10 s) | A large panel moves across the screen the whole time | 300 of 300 | 37 | About one image every 0.27 seconds. |
| scroll (10 s) | A page scrolls for 6 seconds, then stays still | 300 of 300 | 80 | Images through the scroll; none after it stopped at 6.0 s. |
| slides-2min (120 s) | A new slide layout every 15 seconds | 3,600 of 3,600 | 8 | One image per slide, at 0, 15, 30 … 105 seconds. |
Read the results. Duplicate removal keeps images when the layout changes: slides, scrolling, and large movement. It can drop small changes and colour-only changes, because it compares the overall pattern of light and dark areas. Use One frame every N seconds when every moment matters. On cursor-move, interval mode at 1 second gave 10 images, and 7 of them show the cursor. On slides-2min, interval mode at 5 seconds gave 24 images, with repeated images kept.
Speed. In a foreground Chrome tab, each 10-second recording took less than 1 second, and the 2-minute recording took about 4.4 seconds. A 30-second, 60 fps, 1180 × 2556 phone-size recording took about 9 seconds as H.264 and 11 seconds as HEVC, with all 1,800 frames decoded. Before 31 October 2026, this mode played the video in real time instead; the table above shows the same images kept either way. Duplicate removal still drops identical frames, even at threshold 0. Use the local FFmpeg method if you need every frame as an image.
Download the test recordings and try the same settings. Results can differ by browser, device, and decoder.
Video decoding and image encoding run in your browser. Your video is not uploaded.
Drops consecutive frames that look similar to the last kept frame. This can reduce repeated images. Small changes, such as a cursor, can also be dropped.
Export frames as PNG, JPEG or WebP in a ZIP with a per-frame manifest.
Uses the browser's own video decoders through WebCodecs, in a background worker. Speed depends on the file, codec, and device.
Uses an FFmpeg decoding path to tone-map HDR sources to SDR images. This path needs an extra decoder download on first use.
A perceptual hash compares each frame with the last kept frame. The threshold sets how much visual change is needed to keep a new image. It measures overall similarity, not which moment is important.
Decoding, hashing and encoding run on your device. Some paths download a decoder on first use, so load it while online before you work offline. Your video is not uploaded.
In consecutive-duplicate mode, filtering remains active at threshold 0: equal perceptual hashes can still be removed. Timestamp and interval modes keep repeated samples. They require browser-playable video and do not use the HDR or software-decoding paths.
Use this video to photo converter to save a still for a thumbnail, a reference, or a closer look at a moment. It captures images from a local file; it does not generate new detail or design a thumbnail cover.
JPG and JPEG are the same image format. JPEG uses the Quality setting to balance detail and file size. PNG cannot restore detail lost in the original video. Browser seeking chooses the displayed frame, so a timestamp is a requested position rather than a frame-accurate measurement. These two sampling modes require browser-playable SDR video and reject detected PQ/HLG HDR.
Want a video contact sheet instead of separate photos? Compare individual frames and a contact sheet to choose the right output.
Need every decoded frame or prefer a command line? Compare local browser extraction, FFmpeg, and screenshots, with tested commands and image-quality limits.
Not sure which interval to choose? Read how to extract video frames every N seconds for image-count examples, output settings, and sampling limits.
Choose One frame every N seconds to sample a video at a fixed interval. Use 1 second for a quick overview, or 5 seconds for a smaller set. Set a start and optional end time to limit the range. The end is excluded. A synthetic 60-second selection at a 5-second interval produces 12 images, starting at 0 seconds.
Choose One frame at a timestamp to save a still from a requested position. Both sampling modes export a ZIP with a manifest, keep repeated images, and use your chosen image format. Browser seeking selects the displayed frame; these are not frame-accurate extraction modes. Runs are limited to 600 images. Detected PQ/HLG HDR is rejected in sampling mode.
For a single grid instead of separate images, create a timestamped contact sheet. To check a supported file's codec and resolution, use the Video Information Viewer.
Use the recording to find the moment an error appears, then export a still that shows the message and the surrounding interface.
Download the 20-second sample MP4. It shows a mock upload with a progress bar. Treat the in-progress state as the screenshot you want to attach.
Compare the five images with the recording. The progress states are visible, but the first and last cursor movements did not get their own images in the reference run. If a pointer position or a brief message matters to your report, capture that moment by timestamp.
For a sequence of screenshots, sample a short time range around the error. Use a smaller interval when the state changes quickly.
Select a local video, choose a capture mode, and open Options. Set Format to JPEG, then select Extract Frames ZIP. Unzip the download to get the JPG images and manifest. JPG and JPEG are the same image format.
Yes. Choose One frame at a timestamp, enter a position in seconds before the video ends, and select an image format in Options. Extract Frames ZIP downloads one image and a manifest. This mode requires browser-playable SDR video and uses browser seeking, not frame-accurate capture.
No. Extraction runs on your device. The original video is not sent to a server.
Processing runs locally after the required decoder has loaded. HDR and software-decoding paths need an extra decoder download on first use. Load it while online before using those paths offline.
Use a current browser over HTTPS or localhost. Duplicate-removal mode decodes the file with WebCodecs, usually faster than real time. If that decoder is unavailable, the tool tries a software decoder. If decoding fails, it plays the video at normal speed and captures the frames that the browser shows; that fallback takes about as long as the video and can skip frames if you switch tabs. Codec support varies by browser and device.
Start with MP4 or MOV using H.264 or HEVC. WebM files use a software-decoding path when needed. Support depends on the codec, container, and available decoder; not every video file will work.
PNG, JPEG and WebP. Frames download as a ZIP with a manifest describing each frame.
Consecutive frames are often near-identical. Dropping them can reduce the number of repeated images to review or store. The comparison measures overall visual similarity, so small changes such as a cursor can also be dropped. Use interval mode when you need images at regular times.
Videos up to 1 GB are supported. Because processing is local, very large files depend on your device's available memory.
Yes. It is completely free with no account, no sign-up and no upload.
| Feature | Frame Extractor | Video Condenser |
|---|---|---|
| Output | ZIP of PNG, JPEG or WebP images | Shorter MP4 without audio |
| Removal method | Consecutive duplicate frames | Long static sections |
| Timing | Individual still images | Kept sections keep their original timing; small motion can be shortened |
| HDR | Tone-mapping path available | Not supported |
| Video upload | No | No |
Choosing an output? Compare Frame Extractor vs Contact Sheet.