An AI script generator that starts from the runtime

Decide the length first, in words

A script that runs long is not a small problem in a 30-second slot; it is a different ad. Multiply the runtime by a speaking pace and put the result in the brief.

The three paces are planning assumptions, not measurements. Subtract words for any pause, music beat or on-screen-only moment you plan, and read the finished script aloud with a timer before you trust it.

Arithmetic on assumed paces, no external source. The 30-second example below uses the middle column.

The fit check on scene 2

A first draft is usually over. Here is the problem scene as first written, against its 9-second window.

Overhead lighting was designed to illuminate an entire room evenly, which is exactly why it does such a poor job of lighting the small area where you are actually reading, writing and working late at night. It flattens everything and leaves you squinting at your own notes.

First draft

47 wordsabout 18.8 s in a 9 s window

Trimmed

22 wordsabout 8.8 s in a 9 s window

What was cut: the explanation of why overhead light exists. The viewer already knows; the claim survives.

From script to paid copy

Once the scenes are settled, the claim in scene 3 is the line the written ad should carry. Take it to the ad copy generator to derive paid-social copy from the script.

Short ad, 15 to 60 seconds

One claim, one visual proof, one action. Put the word budget per scene in the brief, because the hook window is only a few seconds.

Shape: Hook, problem, proof, action

Explainer or YouTube video

Give the viewer’s question and your answer in one sentence, then ask for sections with time windows and a separate Visual column for the editor.

Shape: Promise, sections, recap, next step

Podcast intro or segment

Ask for talking points, not word-for-word text, and say how many voices there are. Read-aloud prose from a page sounds like a page.

Shape: Cold open, prompts, outro

Sales call guide

Give the persona, the three objections you actually hear and the one next step. The useful output is the questions to ask before you pitch.

Shape: Opener, discovery, objections, close

Paste into any chat

Write a [format] script, [duration] long.
Spoken pace: about [130, 150 or 170] words a minute, so [N] spoken words in total.
Audience: [who is watching or listening, and what they already know]
One claim to land: [a single sentence]
Structure: [for example hook, problem, proof, action], with a time window for each part
Layout: two columns, Visual and Voiceover, with a word count for each scene
Tone: [three adjectives and one thing to avoid]
Call to action: [one]

Then, one change per round

• Cut scene 2 to 22 words without losing the claim.
• Same script at 45 seconds: say what you removed.
• Rewrite the voiceover so each sentence can be said in one breath.

Ask for cuts by scene. “Make it shorter” lets the model choose what to lose.

Which model? Read two aloud

ZeroTwo offers 60+ models. A script is meant to be heard, so the test that matters is the ear, and we have not run a matched script test across models: no winner is named here. Compare them on your own brief.

• OpenAI
• Anthropic
• Google
• xAI
• Mistral
• DeepSeek

See the current model list.