Text-to-Image Generation on DrawBench (ImgRwd, Aes, HPSv2, Pick, CLIP, Mem, Time)
1.22ImageRewardDPS + StitchVM
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| DPS + StitchVMBackbone=SD3.5-Medium, Reward Model=HPSv2 Reward2026.05 | 1.22 | 5.43 | 0.348 | 23.03 | 28.95 | 26 | 16.5 | |
| DPS + StitchVMBackbone=SD3.5-Large, Reward Model=HPSv2 Reward2026.05 | 1.2 | 5.45 | 0.356 | 23.13 | 28.97 | 43.2 | 36.3 | |
| Flow baselineBackbone=SD3.5-Large2026.05 | 1.07 | 5.44 | 0.294 | 22.82 | 29.07 | — | — | |
| DPS + StitchVMBackbone=SD3.5-Large, Reward Model=CLIP Reward2026.05 | 1.05 | 5.2 | 0.279 | 22.44 | 32.95 | 41.5 | 36.3 | |
| DPS + StitchVMBackbone=SD3.5-Large, Reward Model=Aesthetic Reward2026.05 | 1.02 | 5.76 | 0.294 | 22.79 | 28.8 | 40.7 | 36.7 | |
| DPSBackbone=SD3.5-Large, Reward Model=HPSv2 Reward2026.05 | 1.01 | 5.38 | 0.344 | 23.31 | 28.68 | 89.3 | 84.6 | |
| DPS + StitchVMBackbone=SD3.5-Medium, Reward Model=Aesthetic Reward2026.05 | 0.98 | 5.87 | 0.282 | 22.44 | 28.43 | 23.4 | 14.7 | |
| DPSBackbone=SD3.5-Medium, Reward Model=HPSv2 Reward2026.05 | 0.96 | 5.27 | 0.335 | 23.02 | 28.94 | 56.4 | 52.8 | |
| Flow baselineBackbone=SD3.5-Medium2026.05 | 0.95 | 5.31 | 0.283 | 22.56 | 28.86 | — | — | |
| DPSBackbone=SD3.5-Large, Reward Model=Aesthetic Reward2026.05 | 0.91 | 5.71 | 0.28 | 22.41 | 27.95 | 87.3 | 84.8 | |
| DPSBackbone=SD3.5-Medium, Reward Model=Aesthetic Reward2026.05 | 0.82 | 5.73 | 0.28 | 22.47 | 28.16 | 54.3 | 54.2 | |
| DPSBackbone=SD3.5-Large, Reward Model=CLIP Reward2026.05 | 0.72 | 5.04 | 0.263 | 22.03 | 32.55 | 87.5 | 83.1 | |
| DPS + StitchVMBackbone=SD3.5-Medium, Reward Model=CLIP Reward2026.05 | 0.68 | 4.82 | 0.249 | 21.63 | 33.12 | 23.9 | 14.6 | |
| DPSBackbone=SD3.5-Medium, Reward Model=CLIP Reward2026.05 | 0.5 | 4.8 | 0.246 | 21.54 | 33.01 | 54.6 | 52.2 |