Text-to-image generation on Pick-a-Pic v2 (test)
92.9PickScorecPGD
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| cPGDBase Model=SPO-SDXL, Inference Strategy=cPGD2026.02 | 92.9 | 78.8 | 88.4 | 53.8 | 84.4 | — | 82.2 | 96.7 | — | — | — | |
| PGDBase Model=SPO-SDXL, Inference Strategy=PGD2026.02 | 92.2 | 82.1 | 86.1 | 42.5 | 81.4 | — | 80.6 | 96.5 | — | — | — | |
| SPO-SDXLBase Model=SPO-SDXL, Inference Strategy=-2026.02 | 89.4 | 81.8 | 83 | 33.3 | 78.8 | — | 76.9 | 96 | — | — | — | |
| PGD-mergeapproach=Taylor Linearization Merged Checkpoint, guidance_source=PGD2026.02 | 87 | 59.9 | 88.9 | 63.7 | 77.6 | — | 76.9 | 84.4 | — | — | — | |
| MCDPOBase Model=SD1.52025.12 | 86.2 | 91.88 | 93.44 | 57.64 | 82.92 | 76.75 | 81.47 | — | — | — | — | |
| PGDBase Model=DPO-SDXL, Inference Strategy=PGD2026.02 | 83.3 | 59.7 | 85.4 | 62.3 | 73.6 | — | 74.8 | 85.6 | — | — | — | |
| cPGDBase Model=SPO-SD1.5, Inference Strategy=cPGD2026.02 | 82.3 | 71.2 | 81.8 | 60.6 | 76.2 | — | 71.7 | 75.5 | — | — | — | |
| PGDBase Model=KTO-SD1.5, Inference Strategy=PGD2026.02 | 81.6 | 70.3 | 83.3 | 61.1 | 77.4 | — | 73.4 | 80.2 | — | — | — | |
| cPGDBase Model=DPO-SDXL, Inference Strategy=cPGD2026.02 | 80.9 | 63.9 | 77.6 | 58.7 | 64.9 | — | 71.2 | 84.7 | — | — | — | |
| PGDBase Model=MaPO-SDXL, Inference Strategy=PGD2026.02 | 80.4 | 75.7 | 81.6 | 51.4 | 72.2 | — | 73.6 | 81.6 | — | — | — | |
| cPGDBase Model=SDXL, Inference Strategy=cPGD2026.02 | 80 | 50.9 | 80.2 | 64.9 | 69.8 | — | 70.8 | 77.1 | — | — | — | |
| PGDBase Model=SPO-SD1.5, Inference Strategy=PGD2026.02 | 79.7 | 69.8 | 70.3 | 44.3 | 67.9 | — | 66.2 | 69.6 | — | — | — | |
| PGDBase Model=DPO-SD1.5, Inference Strategy=PGD2026.02 | 79.2 | 66.3 | 70.3 | 59.4 | 63.2 | — | 66.1 | 66.3 | — | — | — | |
| cPGDBase Model=DPO-SD1.5, Inference Strategy=cPGD2026.02 | 79 | 71.7 | 81.8 | 62.3 | 76.2 | — | 71.6 | 75.5 | — | — | — | |
| PGDBase Model=SDXL, Inference Strategy=PGD2026.02 | 78.8 | 51.9 | 79 | 63.7 | 69.3 | — | 70.3 | 73.6 | — | — | — | |
| PGDBase Model=SD1.5, Inference Strategy=PGD2026.02 | 78.3 | 62.3 | 71.2 | 58.5 | 63.7 | — | 63.7 | 67.9 | — | — | — | |
| MCSFTBase Model=SD1.52025.12 | 78.24 | 78.16 | 89.6 | 57.36 | 79.2 | 69.53 | 75.35 | — | — | — | — | |
| cPGDBase Model=MaPO-SDXL, Inference Strategy=cPGD2026.02 | 77.4 | 69.1 | 78.8 | 59.4 | 72.4 | — | 71.9 | 72.9 | — | — | — | |
| cPGDBase Model=SD1.5, Inference Strategy=cPGD2026.02 | 76.9 | 63.2 | 71.7 | 59.9 | 72.2 | — | 68.8 | 71.9 | — | — | — | |
| NPOBase Model=DPO-SDXL, Inference Strategy=NPO2026.02 | 76.9 | 53.8 | 81.8 | 57.8 | 70.8 | — | 70 | 81.4 | — | — | — | |
| cPGDBase Model=KTO-SD1.5, Inference Strategy=cPGD2026.02 | 76.7 | 70.5 | 80.2 | 60.4 | 74.3 | — | 70.6 | 75.9 | — | — | — | |
| DPO-SD1.5Inference Strategy=Baseline2026.02 | 76.4 | 65.1 | 67.7 | 55.9 | 60.6 | — | 63.8 | 66.3 | — | — | — | |
| DSPOBase Model=SD1.52025.12 | 76.36 | 75.88 | 81.92 | 59.16 | 77.36 | 68.8 | 73.24 | — | — | — | — | |
| cPGD-mergeapproach=Taylor Linearization Merged Checkpoint, guidance_source=contrastive Preference-Guided Diffusion2026.02 | 76.2 | 45 | 84.2 | 68.9 | 71.7 | — | 69.7 | 72.4 | — | — | — | |
| Diffusion-DPOBase Model=SD1.5, Released checkpoint=true2025.12 | 75.52 | 65.08 | 70.28 | 57.8 | 64.44 | 67.33 | 66.74 | — | — | — | — | |
| MCDPOBase Model=SDXL2025.12 | 74.92 | 63.4 | 85.84 | 58.16 | 77.72 | 64.31 | 70.72 | — | — | — | — | |
| Diffusion-KTOBase Model=SD1.5, Released checkpoint=true2025.12 | 73.4 | 71.24 | 82.24 | 56.64 | 76.04 | 69.3 | 71.48 | — | — | — | — | |
| KTO-SD1.5Inference Strategy=Baseline2026.02 | 72.6 | 68.6 | 78.1 | 58.5 | 75 | — | 69.9 | 76.2 | — | — | — | |
| DPOtype=Optimization Method2026.02 | 71.7 | 53.3 | 77.6 | 61.6 | 65.8 | — | 66.3 | 67.9 | — | — | — | |
| DPO-SDXLBase Model=DPO-SDXL, Inference Strategy=-2026.02 | 71.7 | 53.3 | 77.6 | 61.6 | 65.8 | — | 65.2 | 67.9 | — | — | — | |
| Diffusion-DPOBase Model=SDXL, Released Checkpoint=true2025.12 | 71.6 | 49.2 | 72.92 | 61.24 | 68.64 | 60.36 | 63.99 | — | — | — | — | |
| SPO-SD1.5Inference Strategy=Baseline2026.02 | 71.2 | 68.2 | 63 | 38.7 | 61.1 | — | 61 | 64.9 | — | — | — | |
| DSPOBase Model=SDXL2025.12 | 68.8 | 44.76 | 71.8 | 60.6 | 74.24 | 44.77 | 60.82 | — | — | — | — | |
| NPOtype=Optimization Method2026.02 | 58.7 | 52.1 | 59.2 | 37.5 | 53.5 | — | 55 | 69.1 | — | — | — | |
| NPOBase Model=SDXL, Inference Strategy=NPO2026.02 | 58.7 | 52.1 | 59.2 | 37.5 | 53.5 | — | 54.1 | 69.1 | — | — | — | |
| MaPOtype=Optimization Method2026.02 | 55.9 | 68.2 | 65.3 | 50.2 | 68.2 | — | 61.6 | 61.8 | — | — | — | |
| MaPO-SDXLBase Model=MaPO-SDXL, Inference Strategy=-2026.02 | 55.9 | 68.2 | 65.3 | 50.2 | 68.2 | — | 60.8 | 61.8 | — | — | — | |
| MCSFTBase Model=SDXL2025.12 | 50.72 | 37.24 | 72.36 | 59.12 | 66.2 | 45.96 | 55.26 | — | — | — | — | |
| MAPOBase Model=SDXL, Released Checkpoint=true2025.12 | 50.65 | 56 | 80 | 59.15 | 74.19 | 52.26 | 62.02 | — | — | — | — | |
| SDXLrole=Reference Model2026.02 | 50 | 50 | 50 | 50 | 50 | — | 50 | 50 | — | — | — | |
| SD1.5Inference Strategy=Baseline2026.02 | 50 | 50 | 50 | 50 | 50 | — | 50 | 50 | — | — | — | |
| SDXLBase Model=SDXL, Inference Strategy=-2026.02 | 50 | 50 | 50 | 50 | 50 | — | 50 | 50 | — | — | — | |
| DRaFTInference Steps=1, Reward Grad=Yes2026.06 | 24.69 | 6.82 | — | — | 9.63 | — | — | — | — | — | — | |
| DRaFTInference Steps=1, Reward Grad=Yes2026.06 | 24.45 | 6.712 | — | — | 12.7 | — | — | — | — | — | — | |
| VGG-Flow1-stepInference Steps=1, Reward Grad=Yes2026.06 | 24.27 | 6.49 | — | — | 12.19 | — | — | — | — | — | — | |
| VGG-Flow1-stepInference Steps=1, Reward Grad=Yes2026.06 | 23.73 | 6.378 | — | — | 7.74 | — | — | — | — | — | — | |
| DrPOInference Steps=1, Reward Grad=No2026.06 | 23.66 | 6.717 | — | — | 12.46 | — | — | — | — | — | — | |
| DrPOInference Steps=1, Reward Grad=No2026.06 | 23.49 | 6.485 | — | — | 9.54 | — | — | — | — | — | — | |
| DPO1-stepInference Steps=1, Reward Grad=No2026.06 | 22.77 | 6.227 | — | — | 10.44 | — | — | — | — | — | — | |
| SDXL-DPOInference Steps=502026.06 | 22.57 | 6.076 | — | — | 9.38 | — | — | — | — | — | — | |
| PSOInference Steps=1, Reward Grad=No2026.06 | 22.56 | 6.092 | — | — | 8.97 | — | — | — | — | — | — | |
| GRPO1-stepInference Steps=1, Reward Grad=No2026.06 | 22.5 | 6.077 | — | — | 9.57 | — | — | — | — | — | — | |
| SDXL-Turbo (base)Inference Steps=12026.06 | 22.45 | 6.059 | — | — | 9.36 | — | — | — | — | — | — | |
| SDXL baseInference Steps=502026.06 | 22.15 | 6.104 | — | — | 6.85 | — | — | — | — | — | — | |
| DPO1-stepInference Steps=1, Reward Grad=No2026.06 | 22.02 | 6.08 | — | — | 5.95 | — | — | — | — | — | — | |
| GRPO1-stepInference Steps=1, Reward Grad=No2026.06 | 21.98 | 6.077 | — | — | 6.08 | — | — | — | — | — | — | |
| SD-Turbo (base)Inference Steps=1, Reward Grad=–2026.06 | 21.88 | 6.054 | — | — | 5.75 | — | — | — | — | — | — | |
| PSOInference Steps=1, Reward Grad=No2026.06 | 21.88 | 6.059 | — | — | 5.8 | — | — | — | — | — | — | |
| Poly-DPOParadigm=Off-Policy2026.04 | 21.48 | 5.67 | — | — | 0.679 | — | — | — | 28.3 | — | — | |
| SD2.1Inference Steps=50, Reward Grad=–2026.06 | 21.09 | 5.645 | — | — | 2.49 | — | — | — | — | — | — | |
| Diffusion-KTOParadigm=Off-Policy2026.04 | 21.06 | 5.66 | — | — | 0.628 | — | — | — | 28.06 | — | — | |
| RealAlignTrain Data=512 pairs2026.05 | 21.04 | 5.64 | — | — | 0.38 | — | — | 7.33 | — | 3.06 | 3.96 | |
| Diffusion-DPOTrain Data=851k pairs2026.05 | 21.03 | 5.59 | — | — | 0.33 | — | — | 6.8 | — | 3.03 | 3.78 | |
| Diffusion-DPOParadigm=Off-Policy2026.04 | 20.95 | 5.55 | — | — | 0.297 | — | — | — | 26.12 | — | — | |
| SD1.5Inference Steps=50, Reward Grad=–2026.06 | 20.79 | 5.455 | — | — | 1.22 | — | — | — | — | — | — | |
| SD-1.5Train Data=/2026.05 | 20.65 | 5.48 | — | — | 0.16 | — | — | 5.98 | — | 2.92 | 3.7 | |
| SD1.5Paradigm=-2026.04 | 20.57 | 5.42 | — | — | 0.085 | — | — | — | 25.02 | — | — | |
| LCM-SD1.5Inference Steps=4, Reward Grad=–2026.06 | 20.5 | 5.497 | — | — | -3.08 | — | — | — | — | — | — | |
| Semi-DPOBase=SD1.5, Compared Against=SD1.52026.04 | 0.852 | 0.807 | 0.873 | 0.606 | 0.791 | 0.77 | — | — | — | — | — | |
| Semi-DPOBase=SD1.5, Compared Against=Diff-DPO2026.04 | 0.763 | 0.726 | 0.821 | 0.535 | 0.729 | 0.69 | — | — | — | — | — | |
| Diff-DPOBase=SD1.5, Compared Against=SD1.52026.04 | 0.746 | 0.66 | 0.699 | 0.58 | 0.632 | 0.66 | — | — | — | — | — | |
| Diff-KTOBase=SD1.5, Compared Against=SD1.52026.04 | 0.734 | 0.718 | 0.858 | 0.609 | 0.755 | 0.627 | — | — | — | — | — | |
| Semi-DPOBase=SD1.5, Compared Against=Diff-KTO2026.04 | 0.73 | 0.645 | 0.595 | 0.504 | 0.602 | 0.702 | — | — | — | — | — | |
| Diff-DPOBase=SDXL, Compared Against=SDXL2026.04 | 0.712 | 0.488 | 0.778 | 0.633 | 0.68 | 0.614 | — | — | — | — | — | |
| Semi-DPOBase=SDXL, Compared Against=SDXL2026.04 | 0.65 | 0.653 | 0.82 | 0.572 | 0.74 | 0.7 | — | — | — | — | — | |
| Semi-DPOBase=SDXL, Compared Against=MaPO2026.04 | 0.629 | 0.509 | 0.712 | 0.538 | 0.639 | 0.693 | — | — | — | — | — | |
| MaPOBase=SDXL, Compared Against=SDXL2026.04 | 0.537 | 0.675 | 0.724 | 0.558 | 0.665 | 0.516 | — | — | — | — | — | |
| Semi-DPOBase=SDXL, Compared Against=Diff-DPO2026.04 | 0.429 | 0.687 | 0.552 | 0.44 | 0.562 | 0.605 | — | — | — | — | — | |
| ArenaPOBackbone=Stable Diffusion XL2026.05 | 0.228 | 6.0311 | 0.2889 | 0.3796 | 0.9257 | — | — | — | — | — | — | |
| SDPOBackbone=Stable Diffusion XL2026.05 | 0.2261 | 6.0295 | 0.2879 | 0.376 | 0.8983 | — | — | — | — | — | — | |
| Diff.-DPOBackbone=Stable Diffusion XL2026.05 | 0.2255 | 6.0273 | 0.2874 | 0.3739 | 0.8413 | — | — | — | — | — | — | |
| DSPOBackbone=Stable Diffusion XL2026.05 | 0.2248 | 6.0288 | 0.2873 | 0.3763 | 0.9032 | — | — | — | — | — | — | |
| Pre-TrainedBackbone=Stable Diffusion XL2026.05 | 0.2242 | 5.997 | 0.2846 | 0.3684 | 0.7382 | — | — | — | — | — | — | |
| MaPOBackbone=Stable Diffusion XL2026.05 | 0.2239 | 5.9862 | 0.2859 | 0.3679 | 0.7181 | — | — | — | — | — | — | |
| SFTBackbone=Stable Diffusion XL2026.05 | 0.2203 | 5.9062 | 0.2828 | 0.3668 | 0.6427 | — | — | — | — | — | — | |
| ArenaPOBackbone=Stable Diffusion 1.52026.05 | 0.2164 | 5.7716 | 0.2815 | 0.3522 | 0.6337 | — | — | — | — | — | — | |
| SDPOBackbone=Stable Diffusion 1.52026.05 | 0.2145 | 5.7356 | 0.2789 | 0.3482 | 0.5796 | — | — | — | — | — | — | |
| DSPOBackbone=Stable Diffusion 1.52026.05 | 0.214 | 5.7231 | 0.2782 | 0.3487 | 0.5825 | — | — | — | — | — | — | |
| Diff.-DPOBackbone=Stable Diffusion 1.52026.05 | 0.2127 | 5.6628 | 0.2777 | 0.3427 | 0.5089 | — | — | — | — | — | — | |
| SFTBackbone=Stable Diffusion 1.52026.05 | 0.2119 | 5.6673 | 0.2773 | 0.3419 | 0.5852 | — | — | — | — | — | — | |
| MaPOBackbone=Stable Diffusion 1.52026.05 | 0.2099 | 5.6523 | 0.276 | 0.3397 | 0.4894 | — | — | — | — | — | — | |
| Pre-TrainedBackbone=Stable Diffusion 1.52026.05 | 0.2073 | 5.3907 | 0.2651 | 0.3299 | -0.1376 | — | — | — | — | — | — |