Video to prompt, shot by shot.
Drop a clip. Get a prompt sheet with the measured camera, light, palette and rhythm already filled in — plus a first frame for image-to-video. No upload, no API key, no account.
The whole job, in one tab.
Drop a clip. The workbench finds the cuts, pulls the frames, measures the light and motion, and writes the prompt sheet. No timeline to learn, no project file, no render queue. Nothing to install, nothing to upload.
Nothing is uploaded. Everything runs here.
The brief
Makes the prompts yoursThe workbench never invents a subject. Measured fields fill camera, light, palette and rhythm; you supply the idea. Changing the brief regenerates all three variants.
Analyze a clip to write the prompt sheet.
What a video-to-prompt pack contains
The ZIP is built for the handoff between “I like this reference” and “I am generating a new version.” It contains:
prompts.md— a human-readable sheet with three variants per shot.prompts.json— the same prompts as structured data, for a workflow or agent.shots.csv— timecodes, duration, luma key, contrast, motion, palette and cut confidence.frames/— a first and mid frame per shot, named by shot number.storyboard.jpg— a contact sheet with timecodes burned in.breakdown.jsonandreport.html— the full measurements and a human-readable report.
The frames are references, not upscaled stills. The browser extracts them at a maximum side of 1280 px from the source video.
How a prompt is built
The workbench never invents a subject. It measures the things a model cannot infer from a single still — shot length, aspect ratio, light key, contrast, color temperature, dominant palette, motion direction and rhythm — and then joins them to the brief you typed:
[camera move], [shot size] of [subject] in [setting].
[light key], [contrast] contrast, [color temp] palette.
Color palette: #hex, #hex, #hex.
[style]. [duration] s, [aspect].
No text, no logos, no watermarks, no extra limbs.That is why the prompt reads like a shot description instead of a pile of adjectives. The camera phrase comes from the measured motion, or from the camera override you pick. The light and palette come from the frame. The subject and setting come from you.
Three variants, three jobs
| Variant | Use it when | What changes |
|---|---|---|
| Literal | You want the same look as the reference. | Camera, light, palette and pacing are kept. |
| Same camera, new subject | You are swapping the product or the person. | The grammar stays; the subject is replaced. |
| Stylized | You want a different medium or treatment. | The camera and palette stay; the style is reinterpreted. |
Browser limits
The workbench accepts clips up to 3 minutes and 300 MB, with up to 120 detected shots. MP4/H.264 and WebM are the safest formats; HEVC support depends on the browser. The analysis runs locally: the file is read with the browser’s File API and a Blob URL, not uploaded.
The prompt sheet is a scaffold, not a semantic description. It does not know that the subject is a person, a product or a landscape unless you tell it. For a description from the pixels, use a vision model or the open-source rea-video CLI.
Questions
Does Rea Video send my clip to an AI model?
No. The browser workbench measures the video locally and writes prompts from those measurements plus your brief. It does not call a vision or video model. The prompts are scaffolds for you to paste into the model you already use.
Do I need an API key?
No. There is no account and no key. If you want a semantic description of the subject, you add it in the brief; the measured camera, light, color and motion fields are filled automatically.
Can I use the first frame as an image-to-video reference?
Yes. Every shot in the pack has a first frame and a mid frame. Use the first frame as the image input when your model supports image-to-video, and paste the literal prompt as the motion and camera instruction.
What video formats and sizes work?
MP4/H.264 and WebM are the safest formats. HEVC/H.265 depends on the browser and operating system. The clip must be 3 minutes or shorter and no larger than 300 MB, with up to 120 detected shots.
Is this better than writing a prompt by hand?
It is faster and more consistent for the measurable half: shot length, aspect ratio, light key, contrast, palette, motion direction and rhythm. You still supply the idea, the subject and the taste. The prompt sheet is a starting point, not a guarantee of a good generation.
Turn a clip into a shot list.
A measured scaffold for the next generation — in the browser, in about a minute.