Find the moments people will remember.
Upload a recording and get ranked clips, with the transcript, timecode and selection reason for each one in the result manifest.
How it works
- Add a video or audio source.
- Let the Agent transcribe and rank moments.
- Preview the clips; to move a boundary, load the source into the browser editor and trim it by hand.
- Render the selected clip with the format you need.
How the clipper chooses moments
ToClips first transcribes the recording with word-level timestamps. A language model then reads the whole transcript and proposes clips. Each candidate comes back with start and end times, a title, an opening hook, a suggested caption and hashtags, cover text, a score, and a reason that quotes the line from the transcript that makes the moment worth cutting.
Candidates are sorted by score, so the strongest moments are at the top of the review list. You can open any of them in the editor, move the in and out points, fix the caption text and render again before anything is published.
Selection is driven by what is said, not by what is on screen. That makes the clipper strong on interviews, podcasts, webinars, lectures and commentary, and weak on footage where the story is visual — a silent product demo, a sports highlight or a music video will not produce meaningful candidates.
Inputs and outputs
- Video: MP4, MOV, WebM and MPEG. Audio files (MP3, WAV, M4A) can be uploaded, but clips are rendered from a video track.
- Up to 5 GB per file and up to 20 files in one batch job.
- Up to 200 clips per job. With several sources, the target count is shared out in proportion to each file’s duration.
- Every clip: an H.264 MP4 with AAC audio, an SRT subtitle file and a cover image.
- Every source: a full transcript as SRT and VTT. The whole batch also downloads as one zip, with a JSON manifest of every clip decision.
Getting better candidates
- Describe the result in one sentence: who it is for, where it will be posted and what kind of moment you want — for example “30-second hot takes for YouTube Shorts” or “practical tips a beginner can use today”.
- Set the transcript language to English or Chinese when you know it. Auto-detect works, but mixed-language recordings are transcribed more reliably with an explicit choice.
- Check names, brands and jargon in the transcript before rendering. Speech recognition can mishear proper nouns, and burned-in captions inherit those mistakes.
- Treat the score as a ranking, not a verdict. Watch the top few candidates first; a lower-ranked moment that fits your channel is often the better post.
Frequently asked questions
What sources work best?
Interviews, podcasts, webinars and lessons with a clear spoken narrative are the best starting point.
Can I change the timecode?
Yes, in the browser editor: load the source, set the in and out points and render. In batch mode the boundaries are chosen automatically.
Does ToClips look at the picture when it picks clips?
No. Candidates come from the transcript, so moments that only work visually will be missed. You can still cut them by hand in the editor.
Can I try it without uploading anything?
Yes. The sample page replays a finished sample job — ranked candidates, transcript excerpts and rendered clips — without using credits or uploading a file.