What you need
- Claude Code in a terminal, pointed at a folder with your footage.
- ffmpeg and ffprobe on your PATH.
- A local Whisper model for word-level transcripts.
- Python with numpy, soundfile and OpenCV for the checks.
Pick the moments from the transcript
The best clips open on a first-person line with a loss, a number or a confession, finish a thought by about 4 seconds and carry 20 or more words in the first 6.
That rule comes from Sprites' own YouTube retention data. The clip that opened on "how one ad mistake cost us $12K" held 62% of viewers at 30 seconds. Clips that opened on a reference the viewer did not have, or on "so" or "another common objection", averaged 18 to 22 seconds watched.
Cut, tighten and pace each clip
Each clip gets the full tightening pass, then a per-speaker speed-up toward 185 words a minute with pitch preserved, capped so it never sounds pushed.
Tighten
Fillers, stutters, restatements and pauses over 0.4 s come out, cut in energy valleys. Zero fillers in the first 10 seconds.
Pace per speaker
Measure each speaker's words per minute and speed them toward 185, capped around x1.15 to x1.25. Speed the picture by the same factor in the same chain.
Reframe to 9:16
Face on screen from the first frame, centred with headroom, measured per shot. See reframe to 9:16.
Caption and hook
Burned-in captions rebuilt from the cut word timings, and a hook line of 8 words or fewer in the first 7 seconds. See auto captions.
End clean
About 0.8 s of room tone after the last word, fading out. No black frame, so the clip loops.
Deliver both shapes
9:16 and 16:9 masters, an SRT, a 16:9 thumbnail and a 9:16 cover. See thumbnails.
ffmpeg -i seg.mp4 -filter_complex "[0:v]setpts=PTS/1.12[v];[0:a]atempo=1.12[a]" -map "[v]" -map "[a]" seg_fast.mp4Claude vs an auto clip generator
Auto clippers score moments with their own model and hand you dozens of drafts. Claude works from your rules and your transcript, so you can tell it exactly what a good opening is and check why it picked each one.
The trade is speed for control. A clipper is faster to start. Claude is better when you know what retains for your audience and want every clip held to it. The full comparison is on Opus Clip alternative.
Every Claude video editing guide
Frequently asked questions
How do I turn a podcast into clips with AI?
Transcribe the full episode with word timestamps, have Claude pick 40 to 65 second moments that open on a first-person line with a stake, then tighten, pace, reframe to 9:16 and caption each one with ffmpeg. Approve the picks before anything renders.
How long should a podcast short be?
Sprites cuts its own shorts to 40 to 65 seconds, with a complete thought by about 4 seconds and roughly 20 words in the first 6.
Can Claude turn a long video into shorts?
Yes. Claude Code reads the transcript of the long video, picks the moments, and cuts, reframes and captions each one with ffmpeg on your machine.
Want finished ads, not editing sessions?
Director by Sprites is a full creative team for your startup. Every Monday you get finished, on-brand video ads, reviewed by a human. Strategy, hooks, scripts, shooting and editing, end to end. You never write a prompt. You pick the winners. Sprites ships the campaigns. In private beta.