How 5 AI image models render: “At dusk, a narrow flower shop with rain-dark brick, overflowing window boxes, and warm interior light; the invented name "EMBER VAULT" is rendered in red neon tubing above the entrance, reflected in puddles.”

Same prompt, sent verbatim, at each model's default settings and a fixed seed. Judge: is the text exactly right, spelled correctly, well set?

Fresh prompt from the daily stream · voting in progress — model names and win rates are revealed when this prompt retires from the rotation

Expected text: EMBER VAULT

5 text-to-image models have rendered this typography prompt with the identical text at their default settings. Voting is in progress: the images are shown without attribution so votes stay blind, and the model behind each one, with its win rate, is revealed when the prompt retires from the rotation.

What voters look for: correct spelling, placement and legibility, and type that looks printed or carved rather than pasted on.

FAQ

Why are the model names hidden?

This prompt is still being voted on. The arena is blind: voters compare two images without knowing which model made them. Names and win rates appear here once the prompt retires from the rotation.

Can I vote on this prompt?

Yes — pairs from every live prompt are served on the arena vote page. You will not be shown a prompt you wrote yourself.

Were the prompts adjusted per model?

No. Every model received the exact same prompt, unmodified, at its default settings, with no negative prompt or style suffix.