Compositional Image Generation on GenEval
60.8Overall ScoreSDv1.4
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SDv1.4Description=Base model2026.07 | 60.8 | 98.1 | 34.9 | 36.9 | 73.4 | — | — | — | — | — | — | — | |
| Poly-DPORL Paradigm=Off-Policy, Base Model=SDXL2026.04 | 60.34 | 98.75 | 82.83 | 46.25 | 87.23 | 16 | 31 | — | — | — | — | — | |
| LPORL Paradigm=On-Policy, Base Model=SDXL2026.04 | 59.85 | 99.69 | 84.34 | 43.13 | 90.43 | 13.75 | 27.75 | — | — | — | — | — | |
| SAGEBase Model=SD-v1.4, Reference name in table=Ours2026.07 | 59.8 | 97.2 | 34.4 | 34.4 | 73.4 | — | — | — | — | — | — | 1.6 | |
| SLDBase Model=SD-v1.42026.07 | 58.4 | 97.5 | 25.8 | 40 | 70.2 | — | — | — | — | — | — | 3.9 | |
| Diffusion-DPORL Paradigm=Off-Policy, Base Model=SDXL2026.04 | 58.02 | 99.38 | 82.58 | 49.06 | 85.11 | 13.05 | 18.5 | — | — | — | — | — | |
| DESBase Model=SD-v1.42026.07 | 56.7 | 96.6 | 30.1 | 31.3 | 68.9 | — | — | — | — | — | — | 6.8 | |
| RECEBase Model=SD-v1.42026.07 | 56.2 | 96.6 | 27.5 | 30.9 | 70 | — | — | — | — | — | — | 7.6 | |
| SDXLRL Paradigm=-, Base Model=SDXL2026.04 | 55.63 | 98.12 | 75.25 | 43.75 | 89.63 | 11.25 | 15.75 | — | — | — | — | — | |
| Adv-UnlearnBase Model=SD-v1.42026.07 | 54.2 | 96.3 | 20.5 | 27.2 | 72.9 | — | — | — | — | — | — | 10.9 | |
| MAPORL Paradigm=Off-Policy, Base Model=SDXL2026.04 | 52.8 | 96.56 | 66.41 | 40 | 84.31 | 10.75 | 18.75 | — | — | — | — | — | |
| SPORL Paradigm=On-Policy, Base Model=SDXL2026.04 | 52.75 | 96.88 | 69.7 | 37.19 | 83.51 | 9.5 | 19.75 | — | — | — | — | — | |
| Poly-DPORL Paradigm=Off-Policy, Base Model=SD1.52026.04 | 49.87 | 96.25 | 46.46 | 51.25 | 87.23 | 4 | 14 | — | — | — | — | — | |
| LPORL Paradigm=On-Policy, Base Model=SD1.52026.04 | 48.77 | 97.81 | 55.3 | 42.19 | 80.59 | 6.75 | 10 | — | — | — | — | — | |
| SafeDiffusion-R1Training setup=GenEval+Nudity2026.05 | 47.83 | 99.06 | 61.36 | 30 | 76.33 | 9.75 | 10.5 | — | — | — | — | — | |
| STEREOBase Model=SD-v1.42026.07 | 46.7 | 88.4 | 17.2 | 20.9 | 60.4 | — | — | — | — | — | — | 23.2 | |
| SafeDiffusion-R1Training setup=Nudity Only2026.05 | 44.12 | 96.88 | 43.94 | 35 | 78.19 | 4 | 6.75 | — | — | — | — | — | |
| Diffusion-KTORL Paradigm=Off-Policy, Base Model=SD1.52026.04 | 43.65 | 97.5 | 35.35 | 36.25 | 79.79 | 7 | 6 | — | — | — | — | — | |
| Diffusion-DPORL Paradigm=Off-Policy, Base Model=SD1.52026.04 | 43 | 96.88 | 39.9 | 38.75 | 75.53 | 3.25 | 3.75 | — | — | — | — | — | |
| SD-Safe2026.05 | 42.55 | 97.19 | 38.64 | 34.38 | 77.13 | 3 | 5 | — | — | — | — | — | |
| SD1.5RL Paradigm=-, Base Model=SD1.52026.04 | 42.34 | 95.62 | 37.63 | 37.81 | 74.73 | 3.5 | 4.57 | — | — | — | — | — | |
| SD1.42026.05 | 42.08 | 97.81 | 39.65 | 31.56 | 74.73 | 3 | 5.75 | — | — | — | — | — | |
| SPORL Paradigm=On-Policy, Base Model=SD1.52026.04 | 41.53 | 95.63 | 36.62 | 34.83 | 72.34 | 3.75 | 6.5 | — | — | — | — | — | |
| SafeRCLIPBase Model=SD-v1.42026.07 | 39.5 | 76.6 | 9.3 | 21.3 | 50.8 | — | — | — | — | — | — | 35 | |
| SafeCLIPBase Model=SD-v1.42026.07 | 39.1 | 76.4 | 9.2 | 21.2 | 49.7 | — | — | — | — | — | — | 35.7 | |
| RECE2026.05 | 38.36 | 94.69 | 27.02 | 29.69 | 71.01 | 4 | 3.75 | — | — | — | — | — | |
| MACEBase Model=SD-v1.42026.07 | 36.2 | 80.6 | 8.8 | 18.8 | 36.4 | — | — | — | — | — | — | 40.5 | |
| Seedream 3.02026.03 | 0.99 | 0.96 | 0.91 | 0.93 | 0.47 | 0.8 | — | — | — | — | — | — | |
| DGPONFE=80, Target=GenEval2026.03 | 0.97 | — | — | — | — | — | — | — | — | — | — | — | |
| FlowGRPO#Iter=>5k, Resolution=512x5122025.09 | 0.95 | — | — | — | — | — | — | — | — | — | — | — | |
| Flow-GRPONFE=80, Target=GenEval2026.03 | 0.95 | — | — | — | — | — | — | — | — | — | — | — | |
| Flow-GRPOKL constraint=w/o KL2026.03 | 0.95 | — | — | — | — | — | — | — | — | — | — | — | |
| Flow-GRPOKL constraint=w/ KL2026.03 | 0.95 | — | — | — | — | — | — | — | — | — | — | — | |
| DiffusionNFT#Iter=1.7k, Resolution=512x5122025.09 | 0.94 | — | — | — | — | — | — | — | — | — | — | — | |
| GRPO-GenevalTeacher Model=True2026.05 | 0.94 | — | — | — | — | — | — | — | — | — | — | — | |
| E2POK=12, G=42026.05 | 0.932 | — | — | — | — | — | — | 0.048 | 0.127 | 0.467 | 0.322 | — | |
| CRD2026.03 | 0.93 | — | — | — | — | — | — | — | — | — | — | — | |
| CRDCFG sampling scale=1.52026.03 | 0.93 | — | — | — | — | — | — | — | — | — | — | — | |
| DiffusionNFTK=12, G=242026.05 | 0.922 | — | — | — | — | — | — | 0.054 | 0.118 | 0.418 | 0.259 | — | |
| TDM-R1 (Ours)NFE=4, Context=Compositional Image Generation2026.03 | 0.92 | — | — | — | — | — | — | — | — | — | — | — | |
| DiffusionNFT2026.03 | 0.92 | — | — | — | — | — | — | — | — | — | — | — | |
| CRDCFG sampling scale=3.02026.03 | 0.92 | — | — | — | — | — | — | — | — | — | — | — | |
| Flow-OPDCold-start=Model Merging2026.05 | 0.92 | — | — | — | — | — | — | — | — | — | — | — | |
| DiffusionNFTK=12, G=482026.05 | 0.917 | — | — | — | — | — | — | 0.051 | 0.109 | 0.463 | 0.196 | — | |
| E2POK=4, G=22026.05 | 0.917 | — | — | — | — | — | — | 0.075 | 0.136 | 0.453 | 0.195 | — | |
| DiffusionNFTK=4, G=82026.05 | 0.915 | — | — | — | — | — | — | 0.093 | 0.128 | 0.417 | 0.188 | — | |
| AWMCFG=w/o CFG2026.03 | 0.91 | — | — | — | — | — | — | — | — | — | — | — | |
| Flow-OPDCold-start=Supervised Fine-Tuning (SFT)2026.05 | 0.91 | — | — | — | — | — | — | — | — | — | — | — | |
| p_ϕ in Surrogate RewardNFE=80, Context=Compositional Image Generation2026.03 | 0.89 | — | — | — | — | — | — | — | — | — | — | — | |
| DiffusionNFTK=12, G=42026.05 | 0.868 | — | — | — | — | — | — | 0.077 | 0.126 | 0.458 | 0.26 | — | |
| AWMCFG=w/ CFG2026.03 | 0.86 | — | — | — | — | — | — | — | — | — | — | — | |
| SFT+GRPO-MixInitialization=SFT2026.05 | 0.85 | — | — | — | — | — | — | — | — | — | — | — | |
| Merge+GRPO-MixInitialization=Model Merging2026.05 | 0.84 | — | — | — | — | — | — | — | — | — | — | — | |
| DiffusionNFTK=4, G=22026.05 | 0.822 | — | — | — | — | — | — | 0.076 | 0.042 | 0.45 | 0.194 | — | |
| SuperFlowModel Category=Flow Matching Models2025.12 | 0.8 | 0.99 | 0.87 | 0.72 | 0.87 | 0.33 | 0.69 | — | — | — | — | — | |
| Flow-GRPOModel Category=Flow Matching Models2025.12 | 0.78 | 0.99 | 0.85 | 0.69 | 0.85 | 0.31 | 0.66 | — | — | — | — | — | |
| Flow-GRPOK=12, G=242026.05 | 0.776 | — | — | — | — | — | — | 0.064 | 0.123 | 0.422 | 0.318 | — | |
| Flow-GRPOK=4, G=82026.05 | 0.769 | — | — | — | — | — | — | 0.081 | 0.125 | 0.447 | 0.19 | — | |
| Flow-SPOModel Category=Flow Matching Models2025.12 | 0.75 | 0.99 | 0.84 | 0.66 | 0.84 | 0.3 | 0.64 | — | — | — | — | — | |
| GRPO-Mix2026.05 | 0.73 | — | — | — | — | — | — | — | — | — | — | — | |
| Flow-GRPOK=12, G=42026.05 | 0.724 | — | — | — | — | — | — | 0.051 | 0.126 | 0.462 | 0.321 | — | |
| SD3.5-LResolution=1024x10242025.09 | 0.71 | — | — | — | — | — | — | — | — | — | — | — | |
| Flow-GRPOK=4, G=22026.05 | 0.678 | — | — | — | — | — | — | 0.079 | 0.118 | 0.452 | 0.195 | — | |
| DALLE-3Model Category=Diffusion Models2025.12 | 0.67 | 0.96 | 0.87 | 0.47 | 0.83 | 0.43 | 0.45 | — | — | — | — | — | |
| TDM-R1 (Ours)NFE=4, Context=Visual Text Rendering2026.03 | 0.67 | — | — | — | — | — | — | — | — | — | — | — | |
| FLUX.1-Dev2025.09 | 0.66 | — | — | — | — | — | — | — | — | — | — | — | |
| FlowGRPO#Iter=2k, Resolution=512x5122025.09 | 0.66 | — | — | — | — | — | — | — | — | — | — | — | |
| p_ϕ in Surrogate RewardNFE=80, Context=Visual Text Rendering2026.03 | 0.64 | — | — | — | — | — | — | — | — | — | — | — | |
| GRPO-OCRTeacher Model=True2026.05 | 0.64 | — | — | — | — | — | — | — | — | — | — | — | |
| GRPO-deqaTeacher Model=True2026.05 | 0.64 | — | — | — | — | — | — | — | — | — | — | — | |
| JanusFlowModel Category=Autoregressive Models2025.12 | 0.63 | 0.97 | 0.59 | 0.45 | 0.83 | 0.53 | 0.42 | — | — | — | — | — | |
| SD3.5-M + CFGCFG=true, Resolution=512x5122025.09 | 0.63 | — | — | — | — | — | — | — | — | — | — | — | |
| SD3.5-MNFE=802026.03 | 0.63 | — | — | — | — | — | — | — | — | — | — | — | |
| SD3.5-M2026.03 | 0.63 | — | — | — | — | — | — | — | — | — | — | — | |
| SD-3.5-M2026.05 | 0.63 | — | — | — | — | — | — | — | — | — | — | — | |
| TDM-SD3.5-MNFE=42026.03 | 0.61 | — | — | — | — | — | — | — | — | — | — | — | |
| SD3.5-MModel Category=Flow Matching Models2025.12 | 0.58 | 0.98 | 0.78 | 0.5 | 0.81 | 0.24 | 0.52 | — | — | — | — | — | |
| Delta-KBase Model=Stable Diffusion XL2026.03 | 0.58 | 0.98 | 0.79 | 0.42 | 0.88 | 0.15 | — | — | — | — | — | — | |
| SD-XLModel Category=Diffusion Models2025.12 | 0.55 | 0.98 | 0.74 | 0.39 | 0.85 | 0.15 | 0.23 | — | — | — | — | — | |
| SD-XLResolution=1024x10242025.09 | 0.55 | — | — | — | — | — | — | — | — | — | — | — | |
| Stable Diffusion XL2026.03 | 0.55 | 0.98 | 0.74 | 0.39 | 0.85 | 0.15 | — | — | — | — | — | — | |
| Emu3-GenModel Category=Autoregressive Models2025.12 | 0.54 | 0.98 | 0.71 | 0.34 | 0.81 | 0.17 | 0.21 | — | — | — | — | — | |
| FlowGRPO#Iter=4k, Resolution=512x5122025.09 | 0.54 | — | — | — | — | — | — | — | — | — | — | — | |
| Show-oModel Category=Autoregressive Models2025.12 | 0.53 | 0.95 | 0.52 | 0.49 | 0.82 | 0.11 | 0.28 | — | — | — | — | — | |
| DALLE-2Model Category=Diffusion Models2025.12 | 0.52 | 0.94 | 0.66 | 0.49 | 0.77 | 0.1 | 0.19 | — | — | — | — | — | |
| GRPO-PickscoreTeacher Model=True2026.05 | 0.51 | — | — | — | — | — | — | — | — | — | — | — | |
| SD2.1Model Category=Diffusion Models2025.12 | 0.5 | 0.98 | 0.51 | 0.44 | 0.85 | 0.07 | 0.17 | — | — | — | — | — | |
| Stable Diffusion v2.12026.03 | 0.5 | 0.98 | 0.51 | 0.44 | 0.85 | 0.07 | — | — | — | — | — | — | |
| SD1.5Model Category=Diffusion Models2025.12 | 0.43 | 0.97 | 0.38 | 0.35 | 0.76 | 0.04 | 0.06 | — | — | — | — | — | |
| LDMModel Category=Diffusion Models2025.12 | 0.37 | 0.92 | 0.29 | 0.23 | 0.7 | 0.02 | 0.05 | — | — | — | — | — | |
| SD3.5-MK=12, G=—2026.05 | 0.263 | — | — | — | — | — | — | 0.044 | 0.143 | 0.458 | 0.392 | — | |
| SD3.5-MK=4, G=—2026.05 | 0.263 | — | — | — | — | — | — | 0.044 | 0.143 | 0.458 | 0.392 | — | |
| SD3.5-MCFG=false, Resolution=512x5122025.09 | 0.24 | — | — | — | — | — | — | — | — | — | — | — | |
| minDALL-E2026.03 | 0.23 | 0.73 | 0.11 | 0.12 | 0.37 | 0.02 | — | — | — | — | — | — |