Script Generator
Turn a document into a video script an AI actually wrote.
Not your prose chopped into chunks. The model decides what to leave out, moves the finding to scene one, and rewrites the rest for the ear.
- Cuts the document down — methodology and appendices do not make the video.
- Rewrites written prose into narration that works spoken aloud.
- Every scene comes back timed at a real narration pace.
The cut is the job
A twenty-page report narrated end to end is roughly ninety minutes of video, which nobody watches. So the useful work is not segmentation, it is selection — and that is what a model can do here that a splitter cannot.
It is instructed to keep the finding, the evidence for it, and what it means, and to drop related work, detailed methodology, appendices, and anything the document itself treats as supporting material. It also reorders: documents build to their conclusion, and video loses the viewer before it arrives, so the finding goes in scene one.
- Short, standard, and long, from 5–7 scenes up to 14–20.
- A hard ceiling on narration words per scene, so no scene outruns its visuals.
- Runtime shown per scene and in total at 150 words per minute.
Written prose is not spoken narration
A sentence with two subordinate clauses reads fine and collapses when spoken. Passive constructions hide the actor. Phrases like 'as noted above' and 'see Section 4' mean nothing in a linear medium. Acronyms a reader can scan back for are just noise to a listener.
The model is told to fix all of that: split clauses, prefer active voice, replace document navigation with a restatement of the actual content, expand acronyms on first use, and add the spoken signposts that written headings used to provide.
What to check before you record
Read the narration against your document, focusing on numbers and on any sentence that got shorter. Shortening is where meaning goes missing.
The model is instructed to preserve hedging and copy figures exactly, and it is good at it. It is not perfect at it, and a script is cheap to check compared with a video you have to re-record.
Frequently Asked Questions
How is this different from pasting my document into ChatGPT?
The extraction is the same one our video pipeline uses, so tables and structure survive; the prompt is tuned specifically for the cut and for spoken register; and the output comes back as timed scenes rather than prose you then have to break up.
Does it keep my document's exact wording?
No, deliberately — written prose does not work read aloud. It keeps your figures and qualifiers exact while rewriting the sentences around them.
How accurate is the runtime estimate?
It assumes 150 words per minute with no pauses. Real narration with breaths and beats runs five to ten per cent longer, so treat it as a floor.
Why do I need an account?
Each script is a language model reading your full document on our servers. A free Google account covers it — no card, and the script is yours to copy or download.