Twenty ad variations from one idea

What it is good at

• Turns one approved hero creative into a full testing set within an afternoon.
• Keeps brand consistency because every variant descends from the same approved base.
• Makes it easy to isolate which single change actually improved performance.
• Covers every platform crop (square, story, landscape) from one working file.

Where it falls short

• Text overlays generated inside the image often need manual correction for accuracy.
• Too many variants without a testing plan wastes budget rather than saving it.
• Legal and compliance claims in ad copy still need a human check before launch.

How to use it for creative work

1. 01

   Lock one hero creative first

   Approve a single base image and headline before generating anything else. Every variation should be a controlled change from this base, not a fresh concept, or the set will not read as one campaign.

2. 02

   List the variables worth testing

   Pick three or four things actually worth testing, background colour, subject, headline angle, crop, and generate a version for each while holding everything else fixed. Testing five things at once tells you nothing.

3. 03

   Generate the image variants first

   Use an image tool to produce colour, background and crop variants from the approved photo. Keep the product or subject unchanged in every version so the test measures the variable, not a broken product render.

4. 04

   Write matching headline sets

   Use a text tool to write five headline angles (benefit, urgency, social proof, question, price) sized to fit each crop's character limit. Match tone across all of them so the set still feels like one brand.

5. 05

   Assemble and label for the ad platform

   Export at the exact aspect ratios the platform needs (1:1, 4:5, 9:16) and name files by variable changed, e.g. 'colour-navy_headline-urgency'. Clear naming is what makes the results analysable afterwards.

Ready-to-use prompts

Using the attached approved ad image, keep composition, subject and text placement identical. Produce four background colour variants: [colour 1], [colour 2], [colour 3], [colour 4]. Do not change the product or layout.

Using the attached approved image, produce three crops of the same scene: square 1:1, portrait 4:5, and vertical 9:16, recomposing the subject to stay centred and legible in each without changing the subject itself.

Write five headlines for [product/offer] aimed at [audience], each under [X] characters, one for each angle: direct benefit, urgency/scarcity, social proof, curiosity question, and price/value. Keep the same brand voice: [describe voice].

Same subject as attached, unchanged. Replace the background and setting with [new scene], keeping camera angle, light direction and product position identical to the original.

Given this list of variants I generated, [list], write a filename and one-line hypothesis for each, in the format 'variable-changed_expected-effect', so the results can be tracked against a single hypothesis per ad.

Questions people ask

How many variants should I actually launch?

Enough to test three or four variables cleanly, usually eight to twelve ads, not twenty unrelated ones. More variants without a real testing budget just splits your data too thin to learn anything.

Can AI write the final ad copy legally?

It can draft it, but claims about price, health, performance or comparisons need a human check against your actual policy and local advertising rules before anything goes live.

What is the fastest way to keep the set on-brand?

Generate every variant from the same approved base image and the same written voice guide, changing only one stated variable per version. Consistency comes from the discipline of changing one thing, not from a better tool.