Can it identify celebrities?

When the broadcast context makes it unambiguous (recognized show, on-screen talent in their known program), yes. For ambiguous individuals it falls back to specific physical descriptors so the output stays useful for downstream T2V models.

What about TOS — is downloading legal?

yt-dlp is a widely-used open-source tool. We delete the source video immediately after frame extraction and never cache it. For commercial use, respect each platform's terms and use the output for inspiration, not direct re-publish.

How long does it take?

Most clips finish in 30-60 seconds. Longer or higher-res sources take a bit more for the download + Opus vision pass.

Covers the vision API call (Opus is meaningfully pricier than gpt-4o-mini) plus the bandwidth for yt-dlp + frame extraction. Same as roughly 1 Wan 2.2 render.

reverse-engineer

Get the shot list from any viral video.

Paste a TikTok, YouTube, or Reel — or upload a clip — and Claude Opus 4.7 extracts a structured timeline: Time Code, Camera Move, Scene, Characters, Actions. Study what's working, then write your own version.

See sample output

cost

2 credits per extraction

powered by

Claude Opus 4.7 · GPT-4o vision

live preview

● rec

why it matters

Built for the way creators actually work.

benefit 01

Time-coded shot list

Every shot gets its own block with MM:SS-MM:SS timing, camera setup, and what's actually happening on screen — not vague summaries.

benefit 02

Identifies on-screen talent

Recognizes broadcast personalities in context (NBA panel, news shows, podcast hosts) instead of falling back to 'host in grey suit'.

benefit 03

Export as plain text

Copy to clipboard or download a .txt file. Drop it straight into Multi-Scene UGC, your favorite T2V model, or your script doc.

benefit 04

Works on any clip up to 90s

We sample ~24 keyframes evenly across the clip. Longer videos get the first 90s analyzed; the header tells you what was actually covered.

how it works

4 steps.
Most of it is the model, not you.

Paste a link or drop a file

Works with TikTok, YouTube, Instagram, X, and direct mp4. Uploads supported up to your plan's file size limit.

We extract keyframes

yt-dlp pulls the source, ffmpeg samples 24 evenly-spaced 1024px frames. Source video and frames are deleted from our servers after the run — we never keep your input.

Vision pass writes the timeline

Claude Opus 4.7 (or GPT-4o on fallback) sees all frames at once with embedded timestamps. Returns strict JSON segments.

Use it anywhere

Copy as text, export .txt, or save to your Prompt Studio library. Feed it to Multi-Scene UGC to render your own version with continuity locks.

built for

Made for the people shipping shorts.

Short-form creators

Reverse-engineer your favorite viral video. Get the prompt skeleton, then change the wardrobe, setting, or hook for your own take.

Ad strategists

Audit competitor spots. Pull the exact shot rhythm and camera language they use, then beat it with better hooks.

Cinematographers in training

Drop any clip and see how a pro would describe each shot. Learn the language of camera moves, lenses, and grade.

Briefs + treatments

Generate a starting timeline for a director's treatment in 30 seconds instead of an hour.

models in the loop

Claude Opus 4.7GPT-4o vision

briefing notes

Questions worth answering.

play in the same band

Pairs well with.

/prompt studio

Cinematic Shot Lists

Type a rough idea, get one fully-spec'd cinematic shot list with 16 fields per shot. Replaces 'write 5 prompt variants.'

/narrative

Multi-Scene UGC

Up to 5 cinematic scenes per video — each with its own prompt, avatar, and voice. Wardrobe + character locks travel across cuts.

no card · cancel anytime

Try it free in 60 seconds.

Free plan includes 50 credits and the workspace, projects, brand kit, and saved prompts. Upgrade when you outgrow it.

See pricing