Gemini reads the transcript
HookWise sends Gemini the transcript with a start and end time on every line, plus your settings. Gemini answers with reels: a hook line, the clips that follow it, a title, hook text, a reason and a score. Videos longer than 25 minutes go out in 25-minute windows, with 30 seconds of overlap between them.
Before the list appears, HookWise checks every proposal. Each start and end is snapped to the nearest word boundary, ranges that fall outside the video are dropped, and the reels are sorted by score. When two proposals open on the same hook, only the higher-scoring one stays.
Getting better proposals from your footage
AI Clipping suits talk-heavy video: podcasts, interviews, classes, webinars and livestreams. Gemini only reads the words, so a clean transcript gives it more to work with. If Base gets names or terms wrong, transcribe again with Small or Medium before you run it.
The settings narrow the search. Pick 1 to 10 reels per source and a target length of 15, 30, 45, 60 or 90 seconds, 3 minutes, or free length for a story that needs room. A niche such as "personal finance for first jobbers" and an instruction like "focus on the part about pricing" point Gemini at the right stretch. Start with 3 reels, read the reasons, then run again with sharper instructions.
What it needs and where it falls short
AI Clipping needs the Pro or Studio plan and your own Gemini API key from Google AI Studio. Transcription happens offline, but this step sends the transcript text to Gemini over the internet, and Google bills the tokens to your key. HookWise shows an estimated token count and cost before each run. Google's pricing page has the final rates.
Gemini judges the words only. A visual moment with little dialogue, like a guest bursting into laughter or a product held up to the camera, can be missed. The score is Gemini's own rating of the moment and does not predict views. Watch every reel before you render, and move the cut points that land badly.
