Every AI image generator we've benchmarked

The one to beat, and it is not close. Point it at a label, a logo, or a shop sign and the words come out right, the single hardest thing for these models, and its product shots scored a flawless ten.

The most complete tool in the test: it never dropped below strong in any category, and it is free. Where GPT Image 2 wins on text, this wins on breadth, and it reads Japanese signage correctly, which almost nothing else did.

Elite product photography and gallery-grade photorealism, with the best wet-street reflections and neon of any tool we tested, and the cleanest primitive geometry on spatial prompts.

The text specialist that is elite at product too. It rendered a full retail label set, spine text included, where most tools garble on every surface.

A strikingly consistent all-rounder with the best hands in the benchmark, a genuine two-person cup exchange with correct finger counts and clean proportions.

Strong anime, photoreal and spatial layouts, with a genuine polished cel-shaded illustration and a true square-based pyramid on the shapes prompt.

A video tool whose image mode excels at precision, with the best split-scene render in the test and a serum drop caught suspended mid-fall.

A specialist with the best anime in the whole benchmark, genuine soft cel shading and an authentic uniform, plus the only image to render actual falling rain.

A precise photoreal all-rounder that leads on raw instruction-following, and the only tool to render an actual Tokyo crossing with a real crosswalk.

Beautiful, mood-driven images on a genuinely free tier, undone by garbled text and broken precision, no pyramid on the shapes prompt and both halves snowy on the split scene.