Image generators

Leaderboard: traditional dress register

Blind reader pairs on identical prompts, fit as a Bradley-Terry rating on an Elo-style scale (1500 is the mean) with a 95% bootstrap range, refit on every read of this page; ties count half, calibration pairs and the votes of readers who failed one are left out. A generator is ranked in a category after 40 votes there and shows as collecting before. Beside it, measured here: the photographic Score (a different thing, never combined with the rating), median speed, the provider's price, and failures and refusals per attempt. Suite v1, as of November 7, 2026; 297 reader votes, 0 set aside from 2 readers who failed a calibration pair.

Rank	Generator	Reader rating (95% range)	Votes	Win rate	Photographic Score	Median speed	Price	Failures	Refusals
1	Stable Image Core Stability AI	1549 (1501 to 1609)	72	63%	-3.18	3.9 s	$0.040	0% of 356	13% of 356
2	Recraft V3 Recraft	1520 (1471 to 1569)	107	56%	-2.67	4.5 s	$0.040	8% of 3057	0% of 3057
3	FLUX 1.1 Pro Black Forest Labs	1513 (1467 to 1559)	111	60%	-3.57	3.5 s	$0.040	9% of 3071	0% of 3071
4	Ideogram 3.0 Turbo Ideogram	1460 (1408 to 1517)	64	39%	-3.15	6.5 s	$0.030	45% of 583	0% of 583
5	FLUX.1 schnell Black Forest Labs	1434 (1376 to 1486)	107	32%	-3.27	2.2 s	$0.003	8% of 3054	0% of 3054
	Stable Diffusion 3.5 Large Stability AI	collecting (15 of 40)	15	50%	-3.10	7 s	$0.080	0% of 376	13% of 376

Download: CSV · JSON (ratings and records, CC BY 4.0; renders are each provider's output under its terms).

Head to head

How the comparison works · The prompt suite · All models