Text-to-Image Generation on GenEval (Perceptual Metrics)
0.786DINOOurs (DINO)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Ours (DINO)Model=SDXL-Turbo, Noise initialization=pink noise2025.12 | 0.786 | 0.427 | 0.811 | 0.341 | |
| Ours (DINO)Backbone=SDXL-Turbo2025.12 | 0.784 | 0.411 | 0.767 | 0.349 | |
| Ours (DINO)Model=PixArt-α, Noise initialization=pink noise2025.12 | 0.764 | 0.388 | 0.75 | 0.335 | |
| Ours (DINO)Model=SANA-Sprint-1.6B, Noise initialization=pink noise2025.12 | 0.753 | 0.44 | 0.784 | 0.334 | |
| Parmar et al.Model=SDXL-Turbo, Noise initialization=pink noise2025.12 | 0.749 | 0.392 | 0.757 | 0.323 | |
| Ours (DINO)Backbone=SANA-Sprint-1.6B2025.12 | 0.744 | 0.438 | 0.781 | 0.335 | |
| Parmar et al.Model=SANA-Sprint-1.6B, Noise initialization=pink noise2025.12 | 0.737 | 0.369 | 0.767 | 0.341 | |
| Parmar et al.Backbone=SANA-Sprint-1.6B2025.12 | 0.714 | 0.354 | 0.741 | 0.342 | |
| Parmar et al.Backbone=SDXL-Turbo2025.12 | 0.705 | 0.331 | 0.682 | 0.333 | |
| Ours (DINO)Backbone=PixArt-α2025.12 | 0.695 | 0.335 | 0.664 | 0.337 | |
| Parmar et al.Model=PixArt-α, Noise initialization=pink noise2025.12 | 0.664 | 0.319 | 0.684 | 0.323 | |
| i.i.d.Model=SDXL-Turbo, Noise initialization=pink noise2025.12 | 0.642 | 0.305 | 0.729 | 0.328 | |
| i.i.d.Backbone=SDXL-Turbo2025.12 | 0.588 | 0.249 | 0.642 | 0.335 | |
| Parmar et al.Backbone=PixArt-α2025.12 | 0.559 | 0.246 | 0.569 | 0.327 | |
| i.i.d.Model=SANA-Sprint-1.6B, Noise initialization=pink noise2025.12 | 0.551 | 0.235 | 0.649 | 0.335 | |
| i.i.d.Model=PixArt-α, Noise initialization=pink noise2025.12 | 0.533 | 0.244 | 0.604 | 0.326 | |
| i.i.d.Backbone=SANA-Sprint-1.6B2025.12 | 0.526 | 0.229 | 0.635 | 0.336 | |
| i.i.d.Backbone=PixArt-α2025.12 | 0.431 | 0.182 | 0.474 | 0.326 |