Text-to-Image Generation on TextGround-Bench (Hard)
20.21CSw/ Text Only†
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| w/ Text Only†Fine-tuning=TextGround4M, Resolution=512x512, Supervision Signal=Text2026.04 | 20.21 | 2.53 | 4.15 | 91.05 | 0.43 | 2.88 | |
| TextDiffuser-2Fine-tuning=None2026.04 | 20.06 | 4.36 | 7.05 | 94.87 | 0.15 | 1.18 | |
| w/ Bbox Only†Fine-tuning=TextGround4M, Resolution=512x512, Supervision Signal=Bbox2026.04 | 19.94 | 4.37 | 7.12 | 90.53 | 0.55 | 5.34 | |
| Janus Pro 1B†Fine-tuning=TextGround4M, Resolution=512x5122026.04 | 19.86 | 2.6 | 4.23 | 91.22 | 0.38 | 2.46 | |
| Janus Pro 1BFine-tuning=None2026.04 | 19.79 | 1.46 | 2.24 | 89.95 | 0 | 1.07 | |
| Layout-posting (Ours†)Fine-tuning=TextGround4M, Resolution=512x512, Supervision Signal=Text & Bbox2026.04 | 19.74 | 4.53 | 7.33 | 90.57 | 0.8 | 6.09 | |
| Pre-Image†Fine-tuning=TextGround4M, Resolution=512x5122026.04 | 19.71 | 4.72 | 7.43 | 89.78 | 0.43 | 6.3 | |
| DALL·E 3Fine-tuning=None2026.04 | 19.7 | 48.07 | 36.98 | 81.45 | 1.36 | 42.35 | |
| FLUX.1-devFine-tuning=None2026.04 | 19.27 | 51.85 | 37.46 | 76 | 2.82 | 60.76 | |
| Janus Pro 7BFine-tuning=None2026.04 | 19.17 | 4.87 | 7.22 | 86.94 | 0.19 | 5.98 | |
| GPT-4oFine-tuning=None2026.04 | 19.15 | 77.9 | 65.43 | 64.74 | 8.44 | 83.53 | |
| PixArt-ΣFine-tuning=None2026.04 | 19.12 | 0.78 | 0.68 | 81.86 | 0 | 0 | |
| SD-3.5 LargeFine-tuning=None2026.04 | 17.28 | 58.69 | 45.11 | 68.57 | 1.89 | 38.35 |