List your company

Compositional Text-to-Image Generation on T2I-CompBench++ (test)

85.08ColorMILR
62.730468.532774.33580.1373Feb 2, 2026Feb 12, 2026Feb 22, 2026Mar 5, 2026Mar 15, 2026Mar 25, 2026Apr 5, 2026

Evaluation results

MethodLinks
MILR2026.0285.0851.1769.4946.1338.0830.7860.2136.84—
OP-GRPOBackbone=SD3.5-M, Checkpoint=Best2026.0484.561.4—5546.131.968.2—57.9
LatentMorph2026.0284.0469.4679.8950.934739.5662.3363.6—
Flow-GRPOBackbone=SD3.5-M, Checkpoint=Best2026.0483.861.3—54.544.73267.5—57.3
TwiG-RLLearning Strategy=Reinforcement Learning2026.0282.4961.2873.1934.0638.8731.9961.9354.45—
OP-GRPOBackbone=SD3.5-M, Checkpoint=Intermediate2026.0482.460.3—49.843.731.866—56.7
T2I-R12026.0281.358.5272.4133.7834.2330.945.3239.93—
SD3.5-MCheckpoint=Base Model2026.0479.656.9—28.337.631.259.5—48.9
Flow-GRPOBackbone=SD3.5-M, Checkpoint=Intermediate2026.0476.957.8—41.240.531.264.4—52
FLUX.1 DevCheckpoint=Base Model2026.0473.857.4—28.939.131.562.1—48.8
TwiG-ZSLearning Strategy=Zero-shot2026.0273.1141.5564.7721.9833.6830.936.5848.16—
TIR2026.0268.9249.1260.121.7735.4831.2141.240.12—
Self-CoT2026.0268.1937.8954.121.934.093039.8944.01—
GRPO2026.0267.936.3152.1323.0131.7831.2942.0139.04—
T2I-Copilot2026.0267.4247.8261.3422.1234.0230.4143.4441.85—
SFT2026.0264.2334.5649.4620.9832.931.5541.0935.8—
Vanilla2026.0263.5935.2849.3620.6132.9430.8541.5235.59—