Plan a 20-ad test: four hooks times five visuals

A 20-ad test from one product is a grid of 4 hooks by 5 visuals, so each hook appears in 5 ads and each visual in 4, and you can read which hook and which visual did better. Higgsfield's changelog, read 2026-10-02, says Ads Studio produces up to 20 ads per product; the count is only useful if the 20 are planned.

Image requests in the grid use Sume's Image API fields; planning advice is general.

How do I build the grid?

Pick 4 headlines and 5 visual treatments, such as a plain background, a lifestyle scene, a close-up, a gift setting, and a flat lay. Each cell is one ad. Keep the logo, ratio, and product photo the same in all 20.

How do I generate it on Sume?

One request per hook, with n: 4 or the model's own ceiling, only gives the same prompt several times. For a visual grid, send one prompt per cell, or one prompt per hook and let n supply variety. 20 ad variations from one product image shows the first pattern.

How should I name files?

Name each by cell, such as A-3.png, and record the prompt next to it. When results come back, you can sort by hook or by visual and see which row or column carries the signal.

What would I not read into a result?

Small differences on a small spend are noise. Run the winner against a fresh set before you scale; ad variations for video ads covers the same discipline for video.

Decide before you generate what a win is, such as a click rate or a saved share, and how many impressions each cell needs. Without that, a grid of 20 produces 20 opinions. Keep the grid sheet next to the results so that, later, you can say which hook and which visual carried the lift.