Text-to-Image on GenEval 11 (test)
100Accuracy (Single Obj)SD 3.5 M
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| SD 3.5 MBackbone=SD 3.5 M, Reward=None2025.11 | 100 | 86 | 55 | 82 | 23 | 62 | 68 | |
| Individual Expert (Flow-GRPO)Backbone=SD 3.5 M, Reward=GenEval [11]2025.11 | 100 | 96 | 93 | 91 | 93 | 84 | 93 | |
| Individual Expert (Flow-GRPO)Backbone=SD 3.5 M, Reward=PickScore [15]2025.11 | 100 | 94 | 76 | 88 | 36 | 59 | 76 | |
| Individual Expert (Flow-GRPO)Backbone=SD 3.5 M, Reward=OCR [6]2025.11 | 100 | 87 | 56 | 82 | 25 | 59 | 68 | |
| Rewarded SoupBackbone=SD 3.5 M, Reward=[11] + [15] + [6]2025.11 | 100 | 94 | 76 | 86 | 43 | 72 | 79 | |
| MORL-DBackbone=SD 3.5 M, Reward=[11] + [15] + [6]2025.11 | 100 | 99 | 94 | 89 | 88 | 78 | 91 | |
| MORL-DRBackbone=SD 3.5 M, Reward=[11] + [15] + [6]2025.11 | 100 | 82 | 61 | 81 | 21 | 59 | 67 | |
| MapReduce LoRABackbone=SD 3.5 M, Reward=[11] + [15] + [6]2025.11 | 100 | 98 | 93 | 88 | 81 | 78 | 90 | |
| MapReduce LoRA + RaTEBackbone=SD 3.5 M, Reward=[11] + [15] + [6]2025.11 | 100 | 98 | 93 | 90 | 95 | 79 | 92 | |
| Individual Expert (Flow-GRPO)Backbone=FLUX.1-dev, Reward=GenEval [11]2025.11 | 100 | 100 | 90 | 89 | 90 | 73 | 90 | |
| Individual Expert (Flow-GRPO)Backbone=FLUX.1-dev, Reward=PickScore [15]2025.11 | 100 | 95 | 61 | 85 | 24 | 57 | 70 | |
| Individual Expert (Flow-GRPO)Backbone=FLUX.1-dev, Reward=OCR [6]2025.11 | 100 | 84 | 76 | 83 | 20 | 46 | 68 | |
| MORL-DBackbone=FLUX.1-dev, Reward=[11] + [15] + [6]2025.11 | 100 | 99 | 89 | 90 | 78 | 73 | 88 | |
| MORL-DRBackbone=FLUX.1-dev, Reward=[11] + [15] + [6]2025.11 | 100 | 95 | 90 | 94 | 73 | 79 | 88 | |
| MapReduce LoRABackbone=FLUX.1-dev, Reward=[11] + [15] + [6]2025.11 | 100 | 96 | 88 | 88 | 81 | 79 | 89 | |
| SD 3 MBackbone=SD 3 M, Reward=None2025.11 | 99 | 84 | 56 | 84 | 32 | 52 | 68 | |
| SD 3 M + CaPOBackbone=SD 3 M, Reward=VQAScore [21] + MPS [43] + VILA [14]2025.11 | 99 | 87 | 63 | 86 | 31 | 59 | 71 | |
| FLUX.1-devBackbone=FLUX.1-dev, Reward=None2025.11 | 99 | 87 | 71 | 82 | 18 | 44 | 67 | |
| Rewarded SoupBackbone=FLUX.1-dev, Reward=[11] + [15] + [6]2025.11 | 99 | 96 | 81 | 88 | 31 | 66 | 77 |