Human Preference Alignment on PickScore
86.851PickScoreSuperFlow
Evaluation Results
| Method | Links | |
|---|---|---|
| SuperFlowEvaluation Steps=40, Base Model=SD3.5-M2025.12 | 86.851 | |
| Flow-GRPOEvaluation Steps=40, Base Model=SD3.5-M2025.12 | 85.364 | |
| Flow-SPOEvaluation Steps=40, Base Model=SD3.5-M2025.12 | 85.16 | |
| SD3.5-MEvaluation Steps=402025.12 | 83.039 | |
| UDM-GRPO2026.04 | 23.81 | |
| GRPO-PickscoreTeacher Model=True2026.05 | 23.19 | |
| Flow-OPDCold-start=Model Merging2026.05 | 23.08 | |
| GRPO-deqaTeacher Model=True2026.05 | 23.02 | |
| SD3.5-L2026.04 | 22.91 | |
| FLUX.1-Dev2026.04 | 22.84 | |
| SDXL2026.04 | 22.42 | |
| Merge+GRPO-MixInitialization=Model Merging2026.05 | 21.87 | |
| GRPO-Mix2026.05 | 21.84 | |
| Flow-OPDCold-start=Supervised Fine-Tuning (SFT)2026.05 | 21.83 | |
| URSAClassifier-Free Guidance (CFG)=true2026.04 | 21.79 | |
| SFT+GRPO-MixInitialization=SFT2026.05 | 21.79 | |
| GRPO-OCRTeacher Model=True2026.05 | 21.69 | |
| SD-3.5-M2026.05 | 21.64 | |
| GRPO-GenevalTeacher Model=True2026.05 | 21.53 | |
| URSA (w/o CFG)Classifier-Free Guidance (CFG)=false2026.04 | 20.46 |