An app that finds viral moments in long videos

How does the engine pick moments?

In four steps: transcribe, scan and critique, edit each clip, then reframe and caption. Every step works from the words in the video, so the engine is looking for complete thoughts that a stranger could watch without context, not for loud or fast stretches.

1. Transcribe. The audio is transcribed with word-level timings, which also drive the subtitles.
2. Scan and critique. A language model reads the transcript for candidate passages: a joke with its punchline, a claim with its reason, a story with its ending. A second pass, the critic, keeps those that make sense without the rest of the video.
3. Edit each clip. A per-clip editor checks the chosen boundaries and fixes or drops them, so a clip keeps the setup, the payoff and the reaction, and does not start mid-sentence or stop before the punchline. This editor has been live since 2026-10-03.
4. Reframe and caption. The video is cropped to 9:16 around the speaker and subtitles are burned in.

What is a self-contained moment?

A passage a viewer can follow with no earlier context: it names its subject, reaches a point and ends where the point ends. A funny line that refers to something said ten minutes earlier fails this test The critic drops candidates that fail it, and so does the editor when it cannot find a clean start.

What does it not do?

It does not predict reach, because it looks at what is said and not at your audience, your channel or the platform. It can miss a good moment or pick an average one, and it reads speech only, so a moment that is purely visual is harder for it to see. Review every clip.

• Reach depends on the platform, the account and the timing, none of which the engine sees.
• Streams are harder than podcasts, because the best moment is often a reaction to something on screen.
• A clip you dislike can be re-cut: three re-cuts per video on paid plans.

Which videos does it work best on?

Videos where speech carries the point: podcasts, interviews, talks, tutorials, commentary and streams with a talking streamer. It is weaker on music videos, silent footage and videos where the meaning sits in the picture. For a song or a montage, a manual editor will do a better job than any transcript reader.

What does it cost to run on a long video?

Sources can be up to 3 hours on a paid plan, and you can pick a part of a longer video. Billing counts the minutes of the video you send, so a one-hour podcast uses 60 minutes of your balance, whether the engine returns two clips or ten.