The same prompt, every model

Can the model spell multi-line, specific text correctly instead of gibberish?

"A cafe chalkboard sign reading 'ESPRESSO $4 · LATTE $5 · MUFFIN $3' in neat handwritten chalk lettering, hanging on a brick wall"

Exact counts, colors, and a 'nothing else in frame' constraint — easy to fudge.

"Exactly 3 red apples and 2 green pears in a wicker basket on a wooden table, shot from a 45-degree angle, soft window light, nothing else in frame"

Open-ended and metaphorical — rewards an interesting interpretation over a literal one.

"An imaginative visual metaphor for a small business growing from an idea into a thriving storefront — no literal charts, graphs, or arrows"

Hands, faces, and two-person interaction — the classic AI image failure mode.

"Two people shaking hands in front of a storefront, both faces and both sets of hands clearly visible and in focus, natural daylight"

Same source photo, same instruction, every edit-capable model — tests how well each preserves the product while changing the background.

"Replace the background with a clean white studio backdrop while keeping the product's shape, color, and lighting exactly the same"

Use the same source product photo across every model for a fair comparison — pick something with a busy or colored background.