Text-to-Image Generation on Parti Prompt
0.2906HPSv2Linear-DPO
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Linear-DPOBase Model=SDXL2026.05 | 0.2906 | — | 0.2297 | — | — | — | — | 6.9953 | |
| Diffusion-DPOBase Model=SDXL2026.05 | 0.2893 | — | 0.2295 | — | — | — | — | 6.9971 | |
| MaPOBase Model=SDXL2026.05 | 0.2864 | — | 0.2268 | — | — | — | — | 6.6485 | |
| SDXLBase Model=SDXL2026.05 | 0.2839 | — | 0.2266 | — | — | — | — | 6.4288 | |
| SFTBase Model=SDXL2026.05 | 0.2813 | — | 0.2214 | — | — | — | — | 5.8542 | |
| AlignTokTokenizer=AlignTok, CFG=true, Resolution=256x256, Training steps=100K2025.09 | 0.246 | 0.443 | 0.609 | 0.594 | 5.389 | 32.56 | 78.2 | — | |
| FLUX VAETokenizer=FLUX VAE, CFG=true, Resolution=256x256, Training steps=100K2025.09 | 0.239 | 0.102 | 0.391 | 0.235 | 5.292 | 31.49 | 70.5 | — | |
| AlignTokTokenizer=AlignTok, CFG=Without, Resolution=256x256, Training steps=100K, Model parameters=2B2025.09 | 0.206 | 0.443 | 0.622 | -0.654 | 4.751 | 28.59 | 61.8 | — | |
| FLUX VAETokenizer=FLUX VAE, CFG=Without, Resolution=256x256, Training steps=100K, Model parameters=2B2025.09 | 0.186 | 0.102 | 0.378 | -1.176 | 4.003 | 26.85 | 53.8 | — |