Setup required. Requires setup: Clip Finder runs on a Deepgram transcription key (DEEPGRAM_API_KEY) that an admin must configure. Until that key is set, the page shows an honest "not set up yet" state and uploads are blocked. Limited to about 10 runs per week per account.
What it does
Upload one long-form file — a podcast episode, webinar, interview, livestream recording, or talking-head video — and Clip Finder transcribes the spoken audio with Deepgram, then has Claude read the timestamped transcript and pick the moments most worth turning into a short. For each pick you get the start and end time, a one-line reason it's clip-worthy, a suggested on-screen hook, and a caption to post it with — written in your brand voice when you've filled out your brief. This saves you from scrubbing through an hour of footage hunting for the good 30 seconds. You can optionally set a target clip length (30, 45, 60, 90, 120, or 180 seconds) before uploading so the picks are sized for the platform you're posting to — the page also shows each platform's approximate short-form length limit for reference. Once a run finishes, the uploaded file plays right in the browser and clicking any suggested clip plays just that segment, so you can hear or watch exactly what was picked before you cut. Important: the output is suggestions plus an in-browser preview — Clip Finder does not edit, cut, trim, or export video, so you take the timestamps and hooks into your own editor (or hand them to an editor) to make the actual clips.
How to use it
- 1
Open Clip Finder
Go to /dashboard/clip-finder; if it shows a "not set up yet" notice, the transcription key hasn't been configured yet and an admin needs to add it before you can upload.
- 2
Upload your file
Add one audio file (MP3, M4A, WAV, and similar) or video file (MP4, MOV, WebM), up to 200MB — for a long video that exceeds the limit, export the audio-only version first. Optionally pick a target clip length first so the suggestions match the platform you're posting to.
- 3
Let it transcribe and analyze
Deepgram transcribes the speech and Claude reads the timestamped transcript to pick the standout moments; this runs in one pass while you wait and is typically faster than real time.
- 4
Review the suggested moments
Each result shows a start/end timestamp, why the moment was chosen, a suggested hook, and a suggested caption — each with a copy button.
- 5
Preview the file and each segment
After the run, click Load preview (or any clip's timestamp) to play the uploaded file in the browser — clicking a clip plays just that segment, from its start to its end, so you can verify the pick before cutting.
- 6
Make the clips yourself
Use the timestamps to cut each clip in your own video editor (Clip Finder does not render video), then post it with the suggested hook and caption.
- 7
Find past runs in your history
Each analyzed file stays in your Clip Finder history with its clip list (timestamps, reasons, hooks, and captions to copy). You can delete a job and its uploaded file when you're done with it.
Tips
- Fill out your brand brief first — your business, audience, and tone feed the suggested hooks and captions so they sound like you instead of generic.
- Clear, single-speaker audio (podcasts, talks, interviews) produces the sharpest moment picks; heavy background noise or crosstalk makes the transcript messier.
- Runs are capped at about 10 per week, so pick your longest, highest-value recordings rather than re-running the same file.
- Treat the timestamps and hooks as a starting shortlist — trim a few seconds of lead-in before you cut, since the exact in/out points are estimates.