What it does
Plans a vertical install ad for a mobile app — concept, character consistency, segmenting, voice and music, safe-zone crop, captions and delivery — as instructions precise enough to follow without re-learning each trap. It is built on one rule that outranks the rest: the ad may only claim what the product actually did. App-store categories that run heavily one-star usually run one-star for a single reason, and it is ad-to-product mismatch — a creative that lowers cost-per-install and craters day-one retention has not lowered cost-per-install. So the prompt shown on screen is the prompt that produced the footage shown, a result screen comes from a real run on a real account rather than a seeded record, and a loading wait is compressed rather than cut. The method it carries is the part that is expensive to learn twice: build a character sheet by EDITING one approved frame instead of describing the person four times, because four descriptions produce four siblings that the video model averages into a fifth. Judge a music bed by its loudness envelope rather than by one listen, since a bed that climbs makes every cut audible. Keep the product interface inside the platform's safe zone, not just the text, or the ad's own proof sits behind the caption and the button. It plans and measures; it does not generate, spend or publish. It asks for the evidence behind a claim, and where it was not given any it says the claim is unsupported rather than writing the line anyway. Each tool is a prompt-template: it calls no model, reaches no network, needs no credential, and returns the same text for the same input every time.
Plans an install ad that only claims what the product actually did.
Example prompts
- Plan a 30-second install ad for our photo app. Here's what the product actually does.
- I need four consistent shots of the same character — what's the method that actually holds?
- Should this be one 30s generation or separate segments? There's dialogue and a screen recording.