Text-to-Image Generation on GENEVAL (Compositional Attributes Subset) benchmark leaderboard · SOTA2 Research

Text-to-Image Generation on GENEVAL (Compositional Attributes Subset)

91Overall ScoreRewardFlow

Evaluation Results

Method	Links
RewardFlow2026.04		91	99	99	97	98	47	84
ReNO2026.04		85	99	98	94	95	35	75
Qwen2026.04		83	99	98	92	92	27	71
RewardFlow2026.04		81	99	97	90	95	39	72
ReNO2026.04		72	99	90	79	87	21	56
SD3 (8B)2026.04		68	98	84	66	74	40	43
DALL-E 32026.04		67	96	87	47	83	43	45
RewardFlow2026.04		65	99	77	65	89	21	33
Flux2026.04		64	98	80	64	78	18	43
IF-XL2026.04		61	97	74	66	81	13	35
ReNO2026.04		59	98	72	58	85	15	27
SDXL2026.04		55	98	74	39	85	15	23
DALL-E 22026.04		52	94	66	49	77	10	19
SD v2.12026.04		50	98	51	44	85	7	17
PixArt-α2026.04		48	98	50	44	80	8	7
PixArt-α DMD2026.04		45	95	38	46	76	5	9