Image Generation on ImageNet-1K
2.78FIDD2C
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| D2CArchitecture=SiT XL-2, Training Set=0.05M, Iter.=180k2025.07 | 2.78 | — | — | — | — | — | — | — | — | — | — | |
| D2CArchitecture=DiT L-2, Training Set=0.01M, Iter.=10k2025.07 | 4.2 | — | — | — | — | — | — | — | — | — | — | |
| D2CArchitecture=SiT XL-2, Training Set=0.01M, Iter.=40k2025.07 | 4.3 | — | — | — | — | — | — | — | — | — | — | |
| REGArchitecture=SiT XL-2, Training Set=1.28M, Iter.=200k2025.07 | 5 | — | — | — | — | — | — | — | — | — | — | |
| REPAArchitecture=SiT XL-2, Training Set=1.28M, Iter.=4M2025.07 | 5.9 | — | — | — | — | — | — | — | — | — | — | |
| REPA-EArchitecture=SiT XL-2, Training Set=1.28M, Iter.=235k2025.07 | 5.9 | — | — | — | — | — | — | — | — | — | — | |
| MDLMSampling steps=16, CFG scale=22026.02 | 6.725 | 172.664 | — | — | — | — | — | — | — | — | — | |
| D2CArchitecture=SiT L-2, Training Set=0.01M, Iter.=80k2025.07 | 7.07 | — | — | — | — | — | — | — | — | — | — | |
| GIDDSampling steps=16, CFG scale=22026.02 | 7.076 | 148.148 | — | — | — | — | — | — | — | — | — | |
| SiT XL-2Training Set=1.28M, Iter.=7M2025.07 | 8.3 | — | — | — | — | — | — | — | — | — | — | |
| REPAArchitecture=SiT L-2, Training Set=1.28M, Iter.=700k2025.07 | 8.4 | — | — | — | — | — | — | — | — | — | — | |
| XDLMSampling steps=16, CFG scale=22026.02 | 8.625 | 165.916 | — | — | — | — | — | — | — | — | — | |
| XDLMSampling steps=8, CFG scale=22026.02 | 8.956 | 148.723 | — | — | — | — | — | — | — | — | — | |
| UDLMSampling steps=16, CFG scale=22026.02 | 8.98 | 132.099 | — | — | — | — | — | — | — | — | — | |
| UDLMSampling steps=8, CFG scale=22026.02 | 9.718 | 123.582 | — | — | — | — | — | — | — | — | — | |
| MDLMSampling steps=8, CFG scale=22026.02 | 11.144 | 119.15 | — | — | — | — | — | — | — | — | — | |
| XDLMSampling steps=4, CFG scale=22026.02 | 13.55 | 107.403 | — | — | — | — | — | — | — | — | — | |
| UDLMSampling steps=4, CFG scale=22026.02 | 14.055 | 97.859 | — | — | — | — | — | — | — | — | — | |
| SWEETArchitecture=L6H12, Parameters=64.3M, FLOPs=23.8G, Training Epochs=502026.01 | 14.41 | 38.86 | — | — | — | — | — | — | — | — | — | |
| D2CArchitecture=DiT L-2, Training Set=0.05M, Iter.=10k2025.07 | 14.81 | — | — | — | — | — | — | — | — | — | — | |
| GIDDSampling steps=8, CFG scale=22026.02 | 15.151 | 95.789 | — | — | — | — | — | — | — | — | — | |
| DMAEArchitecture=L6H12, Parameters=64.3M, FLOPs=23.8G, Training Epochs=502026.01 | 15.5 | 37.3 | — | — | — | — | — | — | — | — | — | |
| REPAArchitecture=DiT L-2, Training Set=1.28M, Iter.=400k2025.07 | 15.6 | — | — | — | — | — | — | — | — | — | — | |
| WAVEArchitecture=L6H12, Parameters=64.3M, FLOPs=23.8G, Training Epochs=502026.01 | 16.61 | 34.37 | — | — | — | — | — | — | — | — | — | |
| WT-SelectArchitecture=L6H12, Parameters=64.3M, FLOPs=23.8G, Training Epochs=502026.01 | 17.47 | 33.61 | — | — | — | — | — | — | — | — | — | |
| Iso. PruningArchitecture=L6H12, Parameters=64.3M, FLOPs=23.8G, Training Epochs=502026.01 | 18.68 | 31.53 | — | — | — | — | — | — | — | — | — | |
| SiT L-2Training Set=1.28M, Iter.=400k2025.07 | 18.8 | — | — | — | — | — | — | — | — | — | — | |
| SWEETArchitecture=L12H6, Parameters=31.8M, FLOPs=11.9G, Training Epochs=502026.01 | 19.04 | 30.97 | — | — | — | — | — | — | — | — | — | |
| WAVEArchitecture=L12H6, Parameters=31.8M, FLOPs=11.9G, Training Epochs=502026.01 | 20.37 | 28.17 | — | — | — | — | — | — | — | — | — | |
| WT-SelectArchitecture=L12H6, Parameters=31.8M, FLOPs=11.9G, Training Epochs=502026.01 | 21.68 | 27.5 | — | — | — | — | — | — | — | — | — | |
| DMAEArchitecture=L12H6, Parameters=31.8M, FLOPs=11.9G, Training Epochs=502026.01 | 21.99 | 26.68 | — | — | — | — | — | — | — | — | — | |
| DiT L-2Training Set=1.28M, Iter.=400k2025.07 | 23.3 | — | — | — | — | — | — | — | — | — | — | |
| Iso. PruningArchitecture=L12H6, Parameters=31.8M, FLOPs=11.9G, Training Epochs=502026.01 | 23.83 | 24.92 | — | — | — | — | — | — | — | — | — | |
| SWEETArchitecture=Average, Training Epochs=502026.01 | 26.67 | 26.07 | — | — | — | — | — | — | — | — | — | |
| SWEETArchitecture=L3H12, Parameters=33.6M, FLOPs=11.9G, Training Epochs=502026.01 | 27.6 | 24.04 | — | — | — | — | — | — | — | — | — | |
| DMAEArchitecture=Average, Training Epochs=502026.01 | 28.86 | 24.32 | — | — | — | — | — | — | — | — | — | |
| DMAEArchitecture=L3H12, Parameters=33.6M, FLOPs=11.9G, Training Epochs=502026.01 | 29.09 | 23.18 | — | — | — | — | — | — | — | — | — | |
| WAVEArchitecture=Average, Training Epochs=502026.01 | 29.69 | 23.47 | — | — | — | — | — | — | — | — | — | |
| WT-SelectArchitecture=Average, Training Epochs=502026.01 | 30.64 | 23.06 | — | — | — | — | — | — | — | — | — | |
| SWEETArchitecture=L6H6, Parameters=16.4M, FLOPs=6.0G, Training Epochs=502026.01 | 31.1 | 20.91 | — | — | — | — | — | — | — | — | — | |
| WAVEArchitecture=L3H12, Parameters=33.6M, FLOPs=11.9G, Training Epochs=502026.01 | 31.54 | 21.3 | — | — | — | — | — | — | — | — | — | |
| WT-SelectArchitecture=L3H12, Parameters=33.6M, FLOPs=11.9G, Training Epochs=502026.01 | 31.64 | 21.42 | — | — | — | — | — | — | — | — | — | |
| Iso. PruningArchitecture=Average, Training Epochs=502026.01 | 31.89 | 21.84 | — | — | — | — | — | — | — | — | — | |
| Iso. PruningArchitecture=L3H12, Parameters=33.6M, FLOPs=11.9G, Training Epochs=502026.01 | 32.29 | 21.23 | — | — | — | — | — | — | — | — | — | |
| MDLMSampling steps=4, CFG scale=22026.02 | 33.468 | 54.74 | — | — | — | — | — | — | — | — | — | |
| DMAEArchitecture=L6H6, Parameters=16.4M, FLOPs=6.0G, Training Epochs=502026.01 | 33.5 | 19.47 | — | — | — | — | — | — | — | — | — | |
| WAVEArchitecture=L6H6, Parameters=16.4M, FLOPs=6.0G, Training Epochs=502026.01 | 34.58 | 18.85 | — | — | — | — | — | — | — | — | — | |
| WT-SelectArchitecture=L6H6, Parameters=16.4M, FLOPs=6.0G, Training Epochs=502026.01 | 35.09 | 18.91 | — | — | — | — | — | — | — | — | — | |
| Iso. PruningArchitecture=L6H6, Parameters=16.4M, FLOPs=6.0G, Training Epochs=502026.01 | 37.6 | 17.44 | — | — | — | — | — | — | — | — | — | |
| GIDDSampling steps=4, CFG scale=22026.02 | 41 | 44.084 | — | — | — | — | — | — | — | — | — | |
| SWEETArchitecture=L12H3, Parameters=8.1M, FLOPs=3.0G, Training Epochs=502026.01 | 41.2 | 15.58 | — | — | — | — | — | — | — | — | — | |
| DMAEArchitecture=L12H3, Parameters=8.1M, FLOPs=3.0G, Training Epochs=502026.01 | 44.23 | 14.96 | — | — | — | — | — | — | — | — | — | |
| WAVEArchitecture=L12H3, Parameters=8.1M, FLOPs=3.0G, Training Epochs=502026.01 | 45.35 | 14.66 | — | — | — | — | — | — | — | — | — | |
| Iso. PruningArchitecture=L12H3, Parameters=8.1M, FLOPs=3.0G, Training Epochs=502026.01 | 47.06 | 14.09 | — | — | — | — | — | — | — | — | — | |
| WT-SelectArchitecture=L12H3, Parameters=8.1M, FLOPs=3.0G, Training Epochs=502026.01 | 47.31 | 13.86 | — | — | — | — | — | — | — | — | — | |
| Classifier GuidanceGuidance Method=CG, Guidance Scale (lambda)=12025.10 | — | 28.52 | — | — | — | — | — | 47 | 62 | 57 | 25 | |
| Classifier GuidanceGuidance Method=CG, Guidance Scale (lambda)=22025.10 | — | 38.39 | — | — | — | — | — | 55 | 64 | 63 | 27 | |
| Classifier GuidanceGuidance Method=CG, Guidance Scale (lambda)=102025.10 | — | 83.19 | — | — | — | — | — | 69 | 55 | 74 | 27 | |
| Classifier GuidanceGuidance Method=CG, Guidance Scale (lambda)=202025.10 | — | 75.84 | — | — | — | — | — | 62 | 55 | 70 | 23 | |
| DINObackbone=ViT-B/16, Steps=100K, CFG=w/o, generative_model=SiT-B/22026.06 | — | 33.47 | — | — | 41.15 | 50.51 | 60.7 | — | — | — | — | |
| Dynamic GuidanceGuidance Method=DG, Guidance Scale (lambda)=12025.10 | — | 49.63 | — | — | — | — | — | 62 | 59 | 66 | 27 | |
| Dynamic GuidanceGuidance Method=DG, Guidance Scale (lambda)=22025.10 | — | 59.31 | — | — | — | — | — | 70 | 56 | 70 | 26 | |
| Dynamic GuidanceGuidance Method=DG, Guidance Scale (lambda)=102025.10 | — | 88.49 | — | — | — | — | — | 77 | 52 | 77 | 26 | |
| Dynamic GuidanceGuidance Method=DG, Guidance Scale (lambda)=202025.10 | — | 80.99 | — | — | — | — | — | 70 | 52 | 75 | 23 | |
| RobuQBase Model=α-Flow-XL/2, W/A=4/42025.09 | — | — | 2.59 | 2.23 | — | — | — | — | — | — | — | |
| RobuQBase Model=α-Flow-XL/2, W/A=1.58/42025.09 | — | — | 6.01 | 5.89 | — | — | — | — | — | — | — | |
| Uncond. ADMGuidance Method=None2025.10 | — | 34.24 | — | — | — | — | — | 53 | 61 | 60 | 26 | |
| VISRegbackbone=ViT-B/16, Steps=100K, CFG=w/o, generative_model=SiT-B/22026.06 | — | 33.48 | — | — | 40.36 | 51.38 | 61.26 | — | — | — | — | |
| α-Flow-XL/2 (FP)W/A=32/322025.09 | — | — | 2.58 | 2.15 | — | — | — | — | — | — | — |