Human Preference Alignment on DrawBench Held-out (test)
23.67PickScore (Training Reward)RAM
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| RAM# Steps=3002026.05 | 23.67 | 6.11 | 4.17 | 1.36 | 0.32 | 23.95 | |
| AWM†# Steps=1k2026.05 | 23.39 | 6.31 | 4.1 | 1.27 | 0.31 | 23.76 | |
| Flow-GRPO# Steps=> 3k2026.05 | 23.31 | 5.92 | 4.22 | 1.28 | 0.32 | 23.53 | |
| DiffusionNFT†# Steps=1.4k2026.05 | 23.29 | 6.16 | 4.13 | 1.23 | 0.31 | 23.65 | |
| SD3.5M2026.05 | 21.79 | 5.41 | 4.08 | 0.82 | 0.28 | 22.4 |