Image Editing on ProductConsistency (test)
0.1204Character Error Rate (CER)Flux.1-Kontext-dev + SFT + Cyclic Reward
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Flux.1-Kontext-dev + SFT + Cyclic RewardTraining Protocol=SFT + Cyclic Reward2026.06 | 0.1204 | 0.9224 | 0.8115 | 8.7996 | 7.9901 | 9.074 | 8.6216 | |
| Flux.1-Kontext-dev + SFTTraining Protocol=SFT2026.06 | 0.1293 | 0.9216 | 0.799 | 8.7283 | 7.9467 | 8.9797 | 8.5519 | |
| Flux.1-Kontext-devTraining Protocol=Base2026.06 | 0.149 | 0.921 | 0.811 | 8.7111 | 7.9506 | 8.9096 | 8.524 | |
| Edit-R1-Flux2026.06 | 0.155 | 0.9195 | 0.7966 | 8.7015 | 8.0226 | 9.0142 | 8.5798 | |
| Qwen-Image-Edit-2511 + SFT + Cyclic RewardTraining Protocol=SFT + Cyclic Reward2026.06 | 0.208 | 0.9245 | 0.799 | 8.8866 | 8.3373 | 8.8923 | 8.7055 | |
| RePlan-Flux2026.06 | 0.2914 | 0.9174 | 0.8085 | 8.4118 | 6.8085 | 8.8727 | 8.0311 | |
| GPT-Image-1 High2026.06 | 0.3315 | 0.908 | 0.78 | 9.0134 | 8.5598 | 8.4077 | 8.63 | |
| Edit-R1-Qwen2026.06 | 0.443 | 0.9046 | 0.7597 | 8.5834 | 8.3314 | 8.1542 | 8.3565 | |
| RePlan-Qwen2026.06 | 0.5164 | 0.901 | 0.7419 | 7.6963 | 6.3391 | 7.6542 | 7.2298 | |
| Qwen-Image-Lightning2026.06 | 0.6073 | 0.892 | 0.768 | 8.5188 | 8.1941 | 8.1977 | 8.2738 | |
| Qwen-Image-Edit-2511 + SFTTraining Protocol=SFT2026.06 | 0.7803 | 0.8785 | 0.695 | 8.2763 | 8.1571 | 7.5885 | 8.0062 | |
| Qwen-Image-Edit-2511Training Protocol=Base2026.06 | 1.0682 | 0.8728 | 0.708 | 8.4578 | 8.2467 | 7.5958 | 8.1003 | |
| Nano Banana2026.06 | 1.1868 | 0.886 | 0.702 | 8.8256 | 8.3552 | 8.0839 | 8.4167 | |
| Step1x-edit-v1p22026.06 | 1.1909 | 0.8626 | 0.7157 | 7.5812 | 7.4636 | 7.0414 | 7.3621 | |
| BAGEL2026.06 | 1.681 | 0.8767 | 0.7515 | 7.726 | 7.8088 | 6.2203 | 7.252 | |
| OmniGen22026.06 | 1.7094 | 0.8858 | 0.779 | 7.6739 | 7.8613 | 4.9908 | 6.8422 | |
| HiDream-E1-12026.06 | 3.8774 | 0.839 | 0.724 | 6.5828 | 7.4134 | 3.4477 | 5.8146 |