Compositional Text-to-Image Generation on T2I-CompBench Shape
27.9CLIP ScoreSDXL
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SDXL2026.04 | 27.9 | 95.3 | 0.6 | |
| MultiDiffusion*fall back to vanilla SDXL=true2026.04 | 27.9 | 94.6 | 0.4 | |
| Golden Noise2026.04 | 27.6 | 96.1 | 0.7 | |
| Attend&Excite*fall back to vanilla SDXL=true2026.04 | 27.3 | 94.8 | 0.3 | |
| Golden RPG2026.04 | 26.7 | 95.9 | 0.3 | |
| RPG2026.04 | 26.6 | 94.9 | 0.4 |