Text Rendering Evaluation on Visual text scenarios
88.93NEDZ-Image (Our GRPO)
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Z-Image (Our GRPO)Backbone=Z-Image, Alignment=GRPO2026.05 | 88.93 | 91.05 | 89.24 | 88.76 | 56.48 | |
| Z-Image (Our DPO)Backbone=Z-Image, Alignment=DPO2026.05 | 88.3 | 89.94 | 87.9 | 87.62 | 53.52 | |
| Z-Image2026.05 | 87.39 | 88.6 | 87.26 | 86.57 | 52.78 | |
| Qwen-Image2026.05 | 86.15 | 85.63 | 83.52 | 83.18 | 47.22 | |
| FLUX (Our GRPO)Backbone=FLUX, Alignment=GRPO2026.05 | 60.75 | 60.7 | 57.58 | 58.29 | 20.37 | |
| FLUX (Our DPO)Backbone=FLUX, Alignment=DPO2026.05 | 59.25 | 58.72 | 53.36 | 53.96 | 17.78 | |
| SD3.52026.05 | 58.76 | 52.42 | 51.46 | 50.45 | 10.19 | |
| FLUX2026.05 | 57.68 | 57.46 | 53.01 | 53.15 | 16.67 | |
| TextDiffuser2026.05 | 37.66 | 57.7 | 30.68 | 36.33 | 8.52 | |
| AnyText2026.05 | 28.57 | 0.14 | 0.08 | 0.08 | 0 |