Product / features
Everything between upload and publish.
Transcription, moment selection, clip cutting, captions, framing and export all run off the same source file and the same word-timed transcript.
A transcript that stays attached to the timeline.
The audio is pulled out as 16 kHz mono, split into chunks and transcribed with word-level and segment-level timings. Those timings are what place the cuts and time the captions. While the source is retained with READY status, the clip transcript is yours to correct word by word in the editor - a fix changes the text and leaves its timing alone. It carries no speaker labels.
Review strong moments before you spend time cutting.
A model reads the transcript and returns each candidate clip as one to four ranges, with a title, a description, hashtags and a score from 0 to 10. Clips come back sorted by score, so the strongest moment is the first one in the list.
The first two seconds decide the rest.
Optional rules the model has to satisfy while it picks: open ON the payoff, so the first thing heard is the claim itself rather than "so today I want to talk about", and make every clip self-contained, so a viewer who never saw the source still follows it. Alongside them the clip's own title can hold at headline size over the opening two seconds before dropping to its normal size. The rules steer selection; they do not guarantee a hook exists in the material.
You set the length. The cut lands on a sentence.
Before the run you choose one of five length ranges - 10 to 35, 15 to 60, 30 to 90, 60 to 120 or 90 to 180 seconds, as far as your plan reaches. The AI returns every strong qualifying moment up to your plan's limit; ten to thirty-five is the recommended length. On the plans that carry it you can drop pauses over 1.5, 2 or 3 seconds, or tighten: long pauses and hesitation sounds go, and the clip plays 8% above conversation. The word timings keep each cut on a sentence. The model still chooses the original range; while the source is retained with READY status, you can shorten or re-extend the generated clip only inside its original generated timeline.
05 · Reframe
Wide source, vertical output.
Every clip renders vertical at 9:16, 1080 x 1920 or 720 x 1280. Two framings: blurred pad, the default, which pads the frame with a blurred copy of itself so nothing is cut off, or fill-frame, which crops in. While the source is retained with READY status, switch between them after the run and slide the crop left or right until the speaker sits where you want them. The crop holds that position for the whole clip - it does not track the speaker.
Captions built from the same transcript.
Captions are burned in word by word from the transcript that placed the cut, and the word being spoken flips to a highlight colour. 12 presets are the starting points; behind them the font, size, weight, uppercase, base and highlight colours, outline or background box, vertical position, words per line and the entrance each line arrives on are yours to set before the run or afterwards while the source is retained with READY status.
Change your mind after the render.
While the source is retained with READY status, correct a caption, restyle it, reframe the picture, change the title, or shorten and re-extend the generated clip only inside its original generated timeline, then re-render without spending credits. The model-selected source moment cannot be replaced. Platform safe-zone guides are preview-only and are never burned into the MP4. When the source expires, is deleted, or leaves READY status, the finished MP4 stays downloadable but editing and re-rendering end.
Download the finished MP4.
Each finished clip is a 9:16 MP4 at 1080 x 1920 or 720 x 1280, 30 FPS or the source's own rate, whichever is higher, with H.264 video and AAC audio. It appears in your clip list as soon as the job is done. The recording it was cut from stays in your uploads for the window your plan sets, so a second batch of clips skips both the upload and the transcription.
One run, end to end.
| Stage | Input | Analysis | Edit | Format | Output |
|---|---|---|---|---|---|
| Upload to download | |||||
| What happens | Your file is uploaded and its length is verified. | The audio is transcribed with word timings, then a model scores candidate clips. | Each clip is cut on the ranges the model returned. | The frame is padded or cropped to vertical and the captions are burned in. | The finished file lands in your clip list. |
| What you set | The source file. | Nothing. | Number of clips, length range, pause removal or tightening, hook rules. | Framing, caption preset and style, output size, handle, logo. | Nothing. |
| What is fixed | MP4, MOV or WebM. Your plan sets the size and length ceilings. | Score 0 to 10, up to four ranges per clip. | No arbitrary source-range picker. Generated clip boundaries stay adjustable only inside their original generated timeline while the source is retained with READY status. | 9:16. There is no square or landscape output. | MP4, H.264 video, AAC audio, 30 FPS floor. |
Upload to download
- What happens
- InputYour file is uploaded and its length is verified.
- AnalysisThe audio is transcribed with word timings, then a model scores candidate clips.
- EditEach clip is cut on the ranges the model returned.
- FormatThe frame is padded or cropped to vertical and the captions are burned in.
- OutputThe finished file lands in your clip list.
- What you set
- InputThe source file.
- AnalysisNothing.
- EditNumber of clips, length range, pause removal or tightening, hook rules.
- FormatFraming, caption preset and style, output size, handle, logo.
- OutputNothing.
- What is fixed
- InputMP4, MOV or WebM. Your plan sets the size and length ceilings.
- AnalysisScore 0 to 10, up to four ranges per clip.
- EditNo arbitrary source-range picker. Generated clip boundaries stay adjustable only inside their original generated timeline while the source is retained with READY status.
- Format9:16. There is no square or landscape output.
- OutputMP4, H.264 video, AAC audio, 30 FPS floor.
One recording, start to finish.
Upload, processing, the clip list, playback and the download. Cut between real states of the running product, at the speed it actually runs.
Start clippingReady when the recording is