Video Reverser.

The one that does not make you upload the file, does not cap you at five seconds, and tells you what it is about to do before it does it. This runs entirely on your own machine using the browser's own video decoder and encoder, which is why there is no upload step, no queue page, and no size limit beyond what your disk holds. It handles ten minutes at a time by working in passes rather than holding every frame at once, keeps phone rotation flags so sideways footage lands upright, and gives you three separate answers for the audio instead of silently stripping it. The output is an H.264 MP4 with AAC sound, which plays in every feed and every editor without a second conversion.

Because reversing a clip properly means reversing everything in it, and audio played back to front is what a reversed moment sounds like: the splash arrives before the dive, the word ends before it starts. That is the point when the sound is a splash and a disaster when it is a sentence, so the tool gives you three chips rather than one behaviour. Reverse the sound is the default and turns the audio at exactly the frame the picture turns, so nothing drifts. Keep the sound forward leaves the audio untouched under the reversed picture, which is the right answer for a music bed or a voiceover recorded separately. No sound strips the track, which is what you want if you are laying your own music over it and do not want a claim on it.

Ten minutes per export, and no limit on the file you start from. The ceiling is about patience rather than memory: reversing decodes and encodes every frame, so ten minutes of 1080p is a real wait, and past that people close the tab. Memory is not the constraint, because the tool never holds the whole clip. It works in passes from the end of your marks backwards, decoding a few seconds at a time into a cache of compressed stills, writing that pass out in reverse, then emptying the cache and stepping back. If your marks cover more than ten minutes the in mark moves later rather than the out mark moving earlier, because a long take handed to a reverser is almost always a take whose ending is the interesting part.

There is one generation of re-encode, and it is unavoidable: reversed frames are not the frames that were in the file, so the picture has to be decoded, reordered, and encoded again. The tool spends generously to make that generation invisible, at a bitrate above what a compressor would pick for the same size, and it will never encode above the source's own bitrate, so a lightly compressed clip is not inflated on the way through. Between the decode and the encode each frame is parked as a high quality still, which is one more generation on paper and not one you will see at normal playback. If you want the sharpest possible result, mark the segment you actually need rather than reversing a long take and trimming afterwards.

Use the loop and boomerang maker at /tools/boomerang-video, which is a different tool because it is a different job. A boomerang plays a short moment forward and then straight back through itself, over and over, so what matters is the turn: the frame either side of it has to be the same frame or the bounce reads as a stutter. That means the whole segment has to be in memory at once, which is why the boomerang is capped at thirty seconds. A reverse only ever runs one direction, so it can work in passes and does not need the cap. Rule of thumb: if you want the moment to come back, that is a boomerang. If you want the moment undone, that is this.