Remove Duplicate Frames from a Video

Drop the frames that never changed

Screen recordings, slideshow exports, tutorial captures and animation renders spend most of their length showing the same picture. A 60 fps capture of you reading a page for five seconds is 300 frames of one image. Load the clip here and every consecutive repeat is detected and removed — the file gets smaller and the timeline stops being a wall of identical frames. Everything runs with ffmpeg inside your browser tab: nothing is uploaded, and the tool is free.

Detection uses ffmpeg's mpdecimate filter. It splits each frame into 8×8 blocks and compares it with the last frame it kept: if no block changed more than the hi threshold, and fewer than frac of the blocks changed more than lo, the frame is a duplicate and goes.

The controls

• Sensitivity (1–100) — how eagerly a frame counts as a duplicate. It scales the hi/lo thresholds around ffmpeg's own defaults, which sit at 50 (hi=768, lo=320).
  • 10–30 — only near-identical frames go. Safest for camera footage or anything with film grain you want to keep.
  • 50 (default) — the standard setting for screen and slideshow captures.
  • 70–100 — also removes frames that merely look the same: a blinking text cursor, dithering, video-compression shimmer. Great for slides, risky for footage with slow fades (a gradual fade can be swallowed).
• Timing — the dropped frames leave a hole in the timeline; this decides what happens to it.
  • Keep timing (default) — every remaining frame stays at its original timestamp, so the clip is exactly as long as before and plays identically. The result is a variable frame rate file.
  • Constant frame rate — the kept frames are re-held on an even grid, which is what video editors and NLEs prefer. The near-duplicates become exact repeats, which cost the encoder almost nothing, so the file still shrinks.
  • Compact — the gaps are closed as well, so a recording with long static stretches becomes a short clip of just the moments that changed. Audio is dropped in this mode, because a copied track would drift out of sync with a timeline that no longer matches it.
• Max frame rate (optional) — caps the rate before the duplicate scan, the classic "halve a 60 fps capture to 30" step. Capping first is deliberate: if your source is slower than the cap, the frames the cap inserts are removed again by the scan, so it can never invent frames.
• Output format — Auto keeps mp4/mov/m4v/mkv as-is and turns anything else (e.g. a WebM screen recording) into MP4; MP4 forces H.264 + AAC; WebM forces VP9 + Opus.
• Changed-area threshold (frac) — advanced. The fraction of a frame's blocks that must change for the frame to be kept (default 0.33). Lower it to ~0.05 when a small moving region matters (a mouse cursor, a subtitle line); raise it when only big changes count.

Worked example

demo.mp4 is a 3-second, 10 fps screen capture: 30 frames, but only three distinct pictures — one per second. Load it with the defaults (sensitivity 50, timing Keep timing) and you get demo-dedup.mp4 with 3 frames instead of 30, still playing over the original three seconds. Switch Timing to Compact and the same clip becomes 3 frames over 0.3 s — the pauses are gone. The URL /tools/video-dedup-frames/?sensitivity=70&timing=compact deep-links to that second setup with the fields pre-filled.

Notes and limits

• Only consecutive duplicates are removed. A frame that reappears later (a cut back to an earlier slide) is kept — this is a decimator, not a content-wide de-duplicator.
• Filtering forces a re-encode, so the output is not byte-identical to the source (H.264 crf 20, or VP9 crf 32 for WebM). Audio is stream-copied untouched whenever the container allows it, re-encoded to AAC/Opus when the container changes, and dropped in Compact mode.
• With Keep timing the very last frame has no following frame to hold it, so the video stream can end up to one duplicate-run shorter than the source. If the clip has audio, the file's overall length is unchanged.
• Variable-frame-rate output (Keep timing) plays fine in browsers and players, but some older editors dislike it — use Constant frame rate for those.
• Very noisy sources (heavy grain, low-bitrate camera footage) have no exact duplicates at all; raise the sensitivity or accept that there is nothing to drop.
• Input and output are each capped at 25 MB, since the file is processed in your browser's memory.

FAQ