Script → Scene Splitter
Paste a script and get it broken into scenes — each with its narration, an estimated start and end time, a visual direction, the b-roll search terms to find footage for it, and an image prompt if you are generating visuals.
Paste a script and press Split Into Scenes.
Nothing is sent anywhere until you press it.
Breaking your script into visual scenes…
This usually takes ten to thirty seconds.
Sourcing the footage for these scenes
Each scene above comes with search keywords. These are the three libraries those keywords are worth pasting into — one free, one free with a larger catalogue, one paid with commercial clearance you do not have to think about.
- PexelsFree
Most faceless b-roll, most of the time
Free tierEverything
Visit Pexels → - PixabayFree
Filling gaps when the first library comes up short
Free tierEverything
Visit Pixabay → - StoryblocksSubscription, from roughly $15/mo annually
Channels publishing often enough that footage sameness becomes a problem
Free tierNone
Visit Storyblocks →
How it works
You paste finished narration. The model reads it and decides where the visual should change — which is rarely where a paragraph ends — and for each of those segments it writes what should be on screen, the search phrases to find that footage, and a prompt if you are generating the visuals instead. Your sentences come back unchanged and in order.
The timings are calculated here, not by the model. Each scene's narration is counted in words and divided by the speaking rate for the pacing you chose, then accumulated down the list. That is the difference between a timeline you can edit against and a set of plausible-looking numbers that stop adding up by scene six.
An example
Take a ten-minute documentary script about data centres. Run it through at balanced pacing with stock footage selected, and it comes back as a scene list covering the whole script — how many scenes depends on your script, your pacing and how often the subject actually changes. One of them might cover the sentences about power consumption, with a visual direction of "wide interior of a server hall, rows receding", keywords like data center servers rack and power station cooling towers aerial, and a prompt you could paste into an image generator. You then spend your afternoon downloading exactly the clips on that list instead of re-reading the script over and over to work out what to download.
Why scene planning matters for faceless video
A faceless video has no presenter to hold attention, so the visual has to change often enough to keep the eye busy while the narration does the work. Most weak faceless videos are not badly written — they are one long stock loop under four minutes of narration. Planning the visual beats before you start downloading is what stops that, and it is also what stops you buying footage you never use.
It matters for monetization too. A video assembled from clearly chosen, purposeful visuals is a different thing from a template with a voice over it — which is precisely the distinction the readiness checker is built around.
What it will not do
It will not write your script, fix your script, or add anything to it. If a scene's narration reads oddly, that text came from your draft — this tool deliberately has no licence to improve it. It also will not name artists or studios in the visual prompts, so what comes back is safe to paste into a generator without inviting a copyright problem.
Common questions
Does it rewrite my script?
No. The narration in each scene is your own sentences, in your order — the model is instructed to copy them verbatim and is explicitly told not to add facts, claims or a call to action you did not write. It decides where the scene breaks fall, not what the video says.
Where do the timestamps come from?
From arithmetic, not the model. Each scene's narration is counted in words and divided by the speaking rate for your chosen pacing — 120, 150 or 180 words a minute — then accumulated. That is why the timeline is always sequential with no gaps or overlaps. A language model asked for timestamps produces plausible numbers that quietly drift.
Why is this one limited when the other tools are not?
Because this one costs money per run. Everything that works in your browser — the calculators, the thumbnail tools, the converters — is unlimited and always will be. Anything that calls a model has a daily cap so the free tools can stay free.
Is my script stored?
No. It is sent to the model to be segmented and is not written to any database, not logged, and not retained for training. What is recorded is operational only: which tool ran, whether it worked, how long it took and roughly how many tokens it used.
How long can the script be?
Up to 12,000 characters, which is roughly 2,000 words or about thirteen minutes of narration. Longer scripts are refused rather than silently cut — split them and run the parts separately.
Can I use the b-roll keywords directly?
That is what they are for. They are written as literal search phrases for a stock library, naming a subject and usually a shot — "data center servers rack" rather than "inspiring technology".