Class-Conditional Image Generation on ImageNet 256x256 (train val)
1.06FIDSFD (XL)
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SFD (XL)Epochs=800, #Params=676M2026.02 | 1.06 | — | — | 267 | 78 | 67 | |
| GroupDiff-4*Semantic feature distillation=true, Epoch=800 + 100, Backbone=SiT-XL/2, Classifier-free guidance=true, Fine-tuned from checkpoint=true2025.12 | 1.14 | — | — | 315.3 | 77 | 66 | |
| DDT-XLSemantic feature distillation=true, Epoch=800, Classifier-free guidance=true2025.12 | 1.26 | — | — | 310.6 | 79 | 65 | |
| SiT-XL/2 + REPA-ESemantic feature distillation=true, Epoch=800, Classifier-free guidance=true2025.12 | 1.26 | — | — | 314.9 | 79 | 66 | |
| REPA-EEpochs=800, #Params=675M2026.02 | 1.26 | — | — | 314.9 | 79 | 66 | |
| DDTEpochs=800, #Params=675M2026.02 | 1.26 | — | — | 310.6 | 79 | 65 | |
| RAE (DiTDH)Epochs=800, #Params=839M2026.02 | 1.28 | — | — | 262.9 | 78 | 67 | |
| SFD (XL)Epochs=80, #Params=675M2026.02 | 1.3 | — | — | 233.4 | 78 | 64 | |
| LightningDiTEpochs=800, #Params=675M2026.02 | 1.35 | — | — | 295.3 | 79 | 65 | |
| SiD2Epochs=12802026.02 | 1.38 | — | — | — | — | — | |
| GroupDiff-4*Semantic feature distillation=false, Epoch=1400 + 100, Backbone=SiT-XL/2, Classifier-free guidance=true, Fine-tuned from checkpoint=true2025.12 | 1.4 | — | — | 290.7 | 79 | 64 | |
| REGEpochs=480, #Params=677M2026.02 | 1.4 | — | — | 296.9 | 77 | 66 | |
| SiT-XL/2 + REPASemantic feature distillation=true, Epoch=800, Classifier-free guidance=true2025.12 | 1.42 | — | — | 305.7 | 80 | 64 | |
| SiT-REPA-XL/2epochs=800, params=675M, NFE=250x22025.12 | 1.42 | — | — | — | — | — | |
| GroupDiff-4*Semantic feature distillation=false, Epoch=1400 + 100, Backbone=DiT-XL/2, Classifier-free guidance=true, Fine-tuned from checkpoint=true2025.12 | 1.55 | — | — | 285.4 | 80 | 63 | |
| MARepochs=800, params=943M, NFE=642025.12 | 1.55 | — | — | — | — | — | |
| MDTv2-GCost(Iter×BS)=4600k×256, classifier-free guidance=true2023.03 | 1.58 | — | 4.52 | 314.73 | 79 | 65 | |
| MDTv2-XL/2Semantic feature distillation=false, Epoch=900, Classifier-free guidance=true2025.12 | 1.58 | — | — | 314.7 | 79 | 65 | |
| SRASemantic feature distillation=false, Epoch=800, Backbone=SiT-XL/2, Classifier-free guidance=true2025.12 | 1.58 | — | — | 311.4 | 80 | 63 | |
| MDTv2Epochs=1080, #Params=675M2026.02 | 1.58 | — | — | 314.7 | 79 | 65 | |
| RAE (DiTDH)Epochs=800, #Params=839M2026.02 | 1.6 | — | — | 242.7 | 79 | 65 | |
| MDTv2-GCost(Iter×BS)=3500k×256, classifier-free guidance=true2023.03 | 1.63 | — | 4.45 | 311.73 | 79 | 64 | |
| GroupDiff-4Semantic feature distillation=false, Epoch=800, Backbone=SiT-XL/2, Classifier-free guidance=true2025.12 | 1.63 | — | — | 283.2 | 81 | 64 | |
| MDTv2epochs=700, params=676M, NFE=250x22025.12 | 1.63 | — | — | — | — | — | |
| GroupDiff-4Semantic feature distillation=false, Epoch=800, Backbone=DiT-XL/2, Classifier-free guidance=true2025.12 | 1.66 | — | — | 279.4 | 83 | 57 | |
| REPA-EEpochs=80, #Params=675M2026.02 | 1.67 | — | — | 266.3 | 80 | 63 | |
| VAR-d30 + SSGType=VAR, #Para=2.0B, #Step=10, Time=1.02026.02 | 1.68 | — | — | 313.2 | 81 | 62 | |
| TREADEpochs=740, #Params=675M2026.02 | 1.69 | — | — | 292.7 | 81 | 63 | |
| DiffiTType=Diff., #Para=561M, #Step=2502026.02 | 1.73 | — | — | 276.5 | 80 | 62 | |
| VARType=AR, #Params=2.0B, Steps=102026.05 | 1.73 | — | — | — | — | — | |
| MAR-HType=Mask., #Para=479M, #Step=64, Time=13.42026.02 | 1.78 | — | — | 296 | — | — | |
| MagViT-v2epochs=1080, params=307M, NFE=642025.12 | 1.78 | — | — | — | — | — | |
| MDTCost (Iter x BS)=6500k x 256, Params.=676M, Steps=250, Classifier-free guidance=true2023.12 | 1.79 | — | — | 283 | — | — | |
| MDT-GCost(Iter×BS)=6500k×256, classifier-free guidance=true2023.03 | 1.79 | — | 4.57 | 283.01 | 81 | 61 | |
| VARepochs=350, params=2.0B, NFE=102025.12 | 1.8 | — | — | — | — | — | |
| REPA-EEpochs=800, #Params=675M2026.02 | 1.83 | — | — | 217.3 | — | — | |
| REGEpochs=80, #Params=677M2026.02 | 1.86 | — | — | 321.4 | 76 | 63 | |
| VAR-d30Semantic feature distillation=false, Epoch=350, Classifier-free guidance=true2025.12 | 1.92 | — | — | 323.1 | 82 | 59 | |
| VAR-CoDe N=9Type=VAR, #Para=2.3B, #Step=102026.02 | 1.94 | — | — | 296 | 80 | 61 | |
| HMAR-d30Type=VAR, #Para=2.4B, #Step=142026.02 | 1.95 | — | — | 334.5 | 82 | 62 | |
| Dispersive LossSemantic feature distillation=false, Epoch=1200, Backbone=SiT-XL/2, Classifier-free guidance=true2025.12 | 1.97 | — | — | — | — | — | |
| PixelFlowEpochs=320, #Params=677M2026.02 | 1.98 | — | — | 282.1 | 81 | 60 | |
| FlowDCN-XL/2epochs=400, params=675M, NFE=250x22025.12 | 2 | — | — | — | — | — | |
| VAR-d30Type=VAR, #Para=2.0B, #Step=10, Time=1.02026.02 | 2.02 | — | — | 302.9 | 82 | 60 | |
| SiT-XL/2Semantic feature distillation=false, Epoch=1400, Classifier-free guidance=true2025.12 | 2.06 | — | — | 270.3 | 82 | 59 | |
| SiTEpochs=1400, #Params=675M2026.02 | 2.06 | — | — | 270.3 | 82 | 59 | |
| SiT-XL/2epochs=1400, params=675M, NFE=250x22025.12 | 2.06 | — | — | — | — | — | |
| SiT-XLType=DM, #Params=675M, Steps=2502026.05 | 2.06 | — | — | — | — | — | |
| SiT-CFG + R-predResolution=256x256, Sampling method=ODE sampling, Guidance strategy=CFG + R-pred2026.01 | 2.08 | — | 4.53 | — | 82 | 60 | |
| SiT-CFG + RepGResolution=256x256, Sampling method=ODE sampling, Guidance strategy=CFG + RepG2026.01 | 2.1 | — | 4.61 | — | 80 | 60 | |
| LightningDiTEpochs=64, #Params=675M2026.02 | 2.11 | — | — | 252.3 | 81 | 58 | |
| VDM++Params.=2B, Steps=512, Classifier-free guidance=true2023.12 | 2.12 | — | — | 267.7 | — | — | |
| VDM++Semantic feature distillation=false, Classifier-free guidance=true2025.12 | 2.12 | — | — | 267.7 | — | — | |
| MDT-GCost(Iter×BS)=2500k×256, classifier-free guidance=true2023.03 | 2.15 | — | 4.52 | 249.27 | 82 | 58 | |
| RandAR-XXLSemantic feature distillation=false, Epoch=300, Classifier-free guidance=true2025.12 | 2.15 | — | — | 322 | 79 | 62 | |
| PixNerdEpochs=160, #Params=700M2026.02 | 2.15 | — | — | 297 | 79 | 59 | |
| DiT-CFG + RepGResolution=256x256, Guidance strategy=CFG + RepG2026.01 | 2.17 | — | 4.59 | — | 80 | 60 | |
| LightningDiTEpochs=800, #Params=675M2026.02 | 2.17 | — | — | 205.6 | 77 | 65 | |
| LlamaGen-3BType=AR, #Para=3.1B, #Step=12026.02 | 2.18 | — | — | 267.7 | 84 | 54 | |
| LlamaGen-3BSemantic feature distillation=false, Classifier-free guidance=true2025.12 | 2.18 | — | — | 263.3 | 81 | 58 | |
| LlamaGenType=AR, #Params=3.1B, Steps=2562026.05 | 2.18 | — | — | — | — | — | |
| REGEpochs=480, #Params=677M2026.02 | 2.2 | — | — | 219.1 | 77 | 66 | |
| RandAR-XLepochs=300, params=775M, NFE=2562025.12 | 2.22 | — | — | — | — | — | |
| DiT-XL/2-GClassifier-free guidance (cfg)=1.502022.12 | 2.27 | — | 4.6 | 278.24 | 83 | 57 | |
| DiT-XL/2Cost (Iter x BS)=7000k x 256, Params.=675M, Steps=250, Classifier-free guidance=true2023.12 | 2.27 | — | — | 278.2 | — | — | |
| DiT-XL/2-GCost(Iter×BS)=7000k×256, classifier-free guidance=true2023.03 | 2.27 | — | 4.6 | 278.24 | 83 | 57 | |
| DiT-XL/2Type=Diff., #Para=675M, #Step=250, Time=452026.02 | 2.27 | — | — | 278.2 | 83 | 57 | |
| DiT-CFGResolution=256x256, Guidance strategy=Classifier-free guidance2026.01 | 2.27 | — | 4.8 | — | 82 | 58 | |
| DiT-XL/2Semantic feature distillation=false, Epoch=1400, Classifier-free guidance=true2025.12 | 2.27 | — | — | 278.2 | 83 | 57 | |
| DiTEpochs=1400, #Params=675M2026.02 | 2.27 | — | — | 278.2 | 83 | 57 | |
| DiT-XL/2epochs=1400, params=675M, NFE=250x22025.12 | 2.27 | — | — | — | — | — | |
| DiT-XL/2Type=DM, #Params=675M, Steps=2502026.05 | 2.27 | — | — | — | — | — | |
| MaskDiTCost (Iter x BS)=1200k x 1024, Params.=736M, Steps=40, Classifier-free guidance=true2023.12 | 2.28 | — | — | 276.6 | — | — | |
| L-DiT-7BType=Diff., #Para=7.0B, #Step=250, Time=> 452026.02 | 2.28 | — | — | 316.2 | 83 | 58 | |
| DIFFUSSM-XL-GType=Diff., #Para=660M, #Step=2502026.02 | 2.28 | — | — | 259.1 | 86 | 56 | |
| MaskDiTSemantic feature distillation=false, Epoch=1600, Classifier-free guidance=true2025.12 | 2.28 | — | — | 276.6 | 89 | 61 | |
| MaskDiTEpochs=1600, #Params=675M2026.02 | 2.28 | — | — | 276.6 | 80 | 61 | |
| StyleGAN-XL2022.12 | 2.3 | — | 4.02 | 265.12 | 78 | 53 | |
| StyleGAN2023.03 | 2.3 | — | 4.02 | 265.12 | 78 | 53 | |
| StyleGAN-XLType=GAN, #Para=166M, #Step=1, Time=0.32026.02 | 2.3 | — | — | 265.1 | 78 | 53 | |
| StyleGAN-XLparams=166M, NFE=12025.12 | 2.3 | — | — | — | — | — | |
| MAR-BType=Mask., #Para=208M, #Step=64, Time=10.02026.02 | 2.31 | — | — | 281.7 | — | — | |
| W.A.L.T-LCost (Iter x BS)=437k x 1024, Params.=460M, Steps=50, Classifier-free guidance=true2023.12 | 2.4 | — | — | 290.5 | — | — | |
| STARFlow (VAE)Type=NF, #Params=1.4B, Steps=10242026.05 | 2.4 | — | — | — | — | — | |
| simple diffusionCost (Iter x BS)=500K x 2048, Params.=2B, Steps=512, Classifier-free guidance=true2023.12 | 2.44 | — | — | 256.3 | — | — | |
| SFD (XL)Epochs=800, #Params=676M2026.02 | 2.54 | — | — | — | — | — | |
| GIVT-Causal-L+AType=AR, #Para=304M, #Step=2562026.02 | 2.59 | — | — | — | 81 | 57 | |
| LlamaGen-XLepochs=300, params=775M, NFE=5762025.12 | 2.62 | — | — | — | — | — | |
| SoFlow-XL/2epochs=240, params=676M, NFE=22025.12 | 2.66 | — | — | — | — | — | |
| STARFlow (FAE)Type=NF, #Params=1.4B, Steps=2562026.05 | 2.67 | — | — | — | — | — | |
| NTMType=NF, #Params=1.4B, Steps=162026.05 | 2.8 | — | — | — | — | — | |
| DiT-XL(DiNOv2-B) + RJFEpochs=80, #Params=677M2026.02 | 2.81 | — | — | 201.22 | 82 | 56 | |
| SoFlow-L/2epochs=240, params=459M, NFE=22025.12 | 2.9 | — | — | — | — | — | |
| MeanFlow-XL/2epochs=240, params=676M, NFE=22025.12 | 2.93 | — | — | — | — | — | |
| SoFlow-XL/2epochs=240, params=676M, NFE=12025.12 | 2.96 | — | — | — | — | — | |
| SoFlow-L/2epochs=240, params=459M, NFE=12025.12 | 3.2 | — | — | — | — | — | |
| DiT-XL/2-GClassifier-free guidance (cfg)=1.252022.12 | 3.22 | — | 5.28 | 201.77 | 76 | 62 | |
| NTMType=NF, #Params=1.4B, Steps=82026.05 | 3.24 | — | — | — | — | — | |
| U-VIT-GCost(Iter×BS)=300k×1024, classifier-free guidance=true2023.03 | 3.4 | — | — | — | — | — | |
| REGEpochs=80, #Params=677M2026.02 | 3.4 | — | — | 184.1 | 77 | 63 |