How it works.
Five steps from raw footage to posted clip. Two of them you do; three the AI does. The two you do take under 15 minutes.
Upload your raw footage.
Drag in 2-6 hours of .mp4 / .mov / .mkv from your shoot. Files transfer to your private vault. Multiple cameras or angles get bundled into one session. While the upload finishes, transcription starts in parallel — you don't wait.
Deepgram transcribes with speaker diarization.
Nova-3 model. Word-level timestamps. Speaker labels (engineer, artist, host, guest). Average accuracy 96%+ on studio audio. Done in roughly 5% of the source duration — a 4h shoot transcribes in ~12 min.
Claude runs three passes on the transcript.
Pass 1 flags every candidate moment (jokes, reveals, bars, reactions, hooks). Pass 2 ranks the top 25 by viral potential against your brand voice. Pass 3 writes a click-worthy title for each. Cuts snap to word boundaries — no mid-sentence chops.
You approve in a swipe-friendly review UI.
Each clip shows: thumbnail strip, AI title, score, transcript, and a preview player. Tap a category (your brands, configurable), the clip moves to the approved queue. Reject just as fast. Built to clear 40 candidates in under 10 minutes.
Your editor downloads a categorized zip.
Share-link page shows just the approved clips, organized by category. One 'Download All' button → zip with filenames like 'lush-studio__001__the-vocal-take.mp4'. Editor opens Premiere and cuts. Mark each as 'edited' as they finish. You see the post-prod queue update in real time.
What it's built on.
No black boxes. Same stack we'd pick if we were building it just for ourselves — and we did.
Best-in-class speech-to-text with speaker diarization. Word-level timestamps drive frame-accurate cuts.
Three-pass prompt architecture. Prompt caching cuts cost. Claude's understanding of joke structure + dramatic arc beats other models on this.
S3-compatible blob storage with zero egress fees. Video downloads to editors are free at any scale.
Per-tenant data isolation. Branching enables zero-downtime migrations. Generous free tier scales with us.
Python ecosystem (ffmpeg, openai-whisper fallback, etc.) on serverless infrastructure. Scales to zero between sessions.
Edge-deployed dashboard. Server components for low TTFB. Same framework as this marketing site.
Drop-in JWT auth with hosted UI. Free for the first 10k MAUs. Frees us from rebuilding session management.
Checkout + Customer Portal. Webhooks gate access. Per-plan quotas tied to subscription tier.
Want to see it on your footage?
In the private beta we offer a free white-glove first session: send us 2-4 hours of footage, we run it, and you get the approved clips back to test the workflow with no commitment.