Image Generation on ImageNet 256x256 (test val)
1.78FIDMAR-L
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| MAR-LType=Mask., Latent=Cont., KV-Cache=false, Params=479M2024.12 | 1.78 | — | 81 | 60 | 296 | — | — | — | |
| DiT-XL/2Type=Diff., Latent=Cont., KV-Cache=false, Params=675M2024.12 | 2.27 | — | 83 | 57 | 278.2 | — | — | — | |
| LlamaGen-XXLType=AR, Latent=Disc., KV-Cache=true, Params=1.4B2024.12 | 2.34 | — | 80 | 59 | 253.9 | — | — | — | |
| ACDIT-HType=AR+Diff, Latent=Cont., KV-Cache=true, Params=954M2024.12 | 2.37 | — | 82 | 57 | 273.3 | — | — | — | |
| ACDIT-XLType=AR+Diff, Latent=Cont., KV-Cache=true, Params=677M2024.12 | 2.45 | — | 82 | 57 | 267.4 | — | — | — | |
| ACDIT-LType=AR+Diff, Latent=Cont., KV-Cache=true, Params=460M2024.12 | 2.53 | — | 82 | 55 | 262.9 | — | — | — | |
| LlamaGen+NPPType=AR, Latent=Cont., KV-Cache=true, Params=1.4B2024.12 | 2.55 | — | 84 | 56 | 282 | — | — | — | |
| VAR-d20Type=VAR, Latent=Disc., KV-Cache=true, Params=600M2024.12 | 2.57 | — | 83 | 56 | 302.6 | — | — | — | |
| MonoFormerType=AR+Diff, Latent=Cont., KV-Cache=false, Params=1.1B2024.12 | 2.57 | — | 84 | 56 | 272.6 | — | — | — | |
| ImageFolderType=AR, Latent=Disc., KV-Cache=true, Params=362M2024.12 | 2.6 | — | 75 | 63 | 295 | — | — | — | |
| LlamaGen-XLType=AR, Latent=Disc., KV-Cache=true, Params=775M2024.12 | 2.62 | — | 80 | 57 | 244.1 | — | — | — | |
| LlamaGen-LType=AR, Latent=Disc., KV-Cache=true, Params=343M2024.12 | 3.07 | — | 83 | 52 | 256.1 | — | — | — | |
| VAR-d16Type=VAR, Latent=Disc., KV-Cache=true, Params=310M2024.12 | 3.3 | — | 84 | 51 | 274.4 | — | — | — | |
| LDM-4-GType=Diff., Latent=Cont., KV-Cache=false, Params=400M2024.12 | 3.6 | — | — | — | 247.7 | — | — | — | |
| Guided Diffusion w/ BIGRoC_GTArchitecture=Guided Diffusion, Resolution=256 x 256, Refinement=BIGRoC_GT2021.08 | 3.63 | — | — | — | 260.02 | — | — | — | |
| Guided Diffusion w/ BIGRoC_PLArchitecture=Guided Diffusion, Resolution=256 x 256, Refinement=BIGRoC_PL2021.08 | 3.69 | — | — | — | 249.91 | — | — | — | |
| DartType=AR+Diff, Latent=Cont., KV-Cache=false, Params=820M2024.12 | 3.82 | — | — | — | 263.8 | — | — | — | |
| ADM-G, ADM-UADM-G Iterations=540K, ADM-U Iterations=500K, Generator Compute (V100-days)=329, Classifier Compute (V100-days)=30, Total Compute (V100-days)=3592021.05 | 3.85 | 5.86 | 84 | 53 | — | — | — | — | |
| Guided DiffusionArchitecture=Guided Diffusion, Resolution=256 x 2562021.08 | 3.94 | — | — | — | 215.84 | — | — | — | |
| CBS (SiT-XL/2)Classifier-Free Guidance=Enabled, # Act. Par.=675, GFLOPs=118.64, Boundaries=(0.4, 0.77)2026.06 | 4.03 | — | 80 | 56 | 195.72 | — | — | — | |
| MAR-L-CausalType=AR, Latent=Cont., KV-Cache=false, Params=479M2024.12 | 4.07 | — | — | — | 232.4 | — | — | — | |
| ADM-G, ADM-UADM-G Iterations=540K, ADM-U Iterations=150K, Generator Compute (V100-days)=219, Classifier Compute (V100-days)=30, Total Compute (V100-days)=2492021.05 | 4.15 | 6.14 | 82 | 54 | — | — | — | — | |
| ADM-GIterations=1980K, Generator Compute (V100-days)=916, Classifier Compute (V100-days)=46, Total Compute (V100-days)=9622021.05 | 4.59 | 5.25 | 82 | 52 | — | — | — | — | |
| ADM-G, ADM-UADM-G Iterations=200K, ADM-U Iterations=150K, Generator Compute (V100-days)=110, Classifier Compute (V100-days)=16, Total Compute (V100-days)=126, Classifier iterations=150K2021.05 | 4.93 | 5.82 | 82 | 52 | — | — | — | — | |
| DiT-L/2Type=Diff., Latent=Cont., KV-Cache=false, Params=458M2024.12 | 5.02 | — | 75 | 57 | 167.2 | — | — | — | |
| MaskGITType=Mask., Latent=Disc., KV-Cache=false, Params=227M2024.12 | 6.18 | — | 83 | 58 | 316.2 | — | — | — | |
| SiT-XL/2 (Uniform Partitioned)Classifier-Free Guidance=Enabled, # Act. Par.=675, GFLOPs=118.64, Boundaries=(0.33, 0.66)2026.06 | 6.24 | — | 74 | 56 | 165.29 | — | — | — | |
| ADM-GIterations=750K, Generator Compute (V100-days)=347, Classifier Compute (V100-days)=46, Total Compute (V100-days)=3932021.05 | 6.49 | 5.39 | 81 | 50 | — | — | — | — | |
| SiT-XL/2 (Monolithic)Classifier-Free Guidance=Enabled, # Act. Par.=675, GFLOPs=118.64, Partitioning=Monolithic2026.06 | 6.53 | — | 73 | 56 | 162.23 | — | — | — | |
| ADM-GIterations=750K, Generator Compute (V100-days)=347, Classifier Compute (V100-days)=14, Total Compute (V100-days)=361, Classifier iterations=150K2021.05 | 6.68 | 5.34 | 81 | 51 | — | — | — | — | |
| BigGAN-deep w/ BIGRoC_GTArchitecture=BigGAN-deep, Resolution=256 x 256, Refinement=BIGRoC_GT2021.08 | 6.84 | — | — | — | 228.23 | — | — | — | |
| BigGAN-deep w/ BIGRoC_PLArchitecture=BigGAN-deep, Resolution=256 x 256, Refinement=BIGRoC_PL2021.08 | 6.93 | — | — | — | 221.78 | — | — | — | |
| MAGEType=Mask., Latent=Disc., KV-Cache=false, Params=230M2024.12 | 6.93 | — | — | — | 195.8 | — | — | — | |
| BigGAN-deepTotal Compute (V100-days)=128-2562021.05 | 6.95 | 7.36 | 87 | 28 | — | — | — | — | |
| BigGAN-deepArchitecture=BigGAN-deep, Resolution=256 x 2562021.08 | 7.03 | — | — | — | 202.65 | — | — | — | |
| RQTranType=AR, Latent=Disc., KV-Cache=true, Params=3.8B2024.12 | 7.55 | — | — | — | 134 | — | — | — | |
| CBS (SiT-B/2)Classifier-Free Guidance=Enabled, # Act. Par.=130, GFLOPs=23.01, Boundaries=(0.4, 0.77)2026.06 | 10.72 | — | 72 | 56 | 121.09 | — | — | — | |
| ADMType=Diff., KV-Cache=false, Params=554M2024.12 | 10.94 | — | 69 | 63 | 101 | — | — | — | |
| VQGANType=AR, Latent=Disc., KV-Cache=true, Params=1.4B2024.12 | 15.78 | — | — | — | 78.3 | — | — | — | |
| CBS (SiT-XL/2)Classifier-Free Guidance=Disabled, # Act. Par.=675, GFLOPs=118.64, Boundaries=(0.4, 0.77)2026.06 | 15.81 | — | 64 | 66 | 86.43 | — | — | — | |
| SiT-B/2 (Uniform Partitioned)Classifier-Free Guidance=Enabled, # Act. Par.=130, GFLOPs=23.01, Boundaries=(0.33, 0.66)2026.06 | 16.51 | — | 66 | 56 | 87.21 | — | — | — | |
| SiT-B/2 (Monolithic)Classifier-Free Guidance=Enabled, # Act. Par.=130, GFLOPs=23.01, Partitioning=Monolithic2026.06 | 16.79 | — | 66 | 55 | 84.75 | — | — | — | |
| SiT-XL/2 (Uniform Partitioned)Classifier-Free Guidance=Disabled, # Act. Par.=675, GFLOPs=118.64, Boundaries=(0.33, 0.66)2026.06 | 17.97 | — | 63 | 64 | 73.85 | — | — | — | |
| SiT-XL/2 (Monolithic)Classifier-Free Guidance=Disabled, # Act. Par.=675, GFLOPs=118.64, Partitioning=Monolithic2026.06 | 18.04 | — | 63 | 64 | 73.9 | — | — | — | |
| CBS (SiT-S/2)Classifier-Free Guidance=Enabled, # Act. Par.=33, GFLOPs=6.06, Boundaries=(0.4, 0.77)2026.06 | 18.61 | — | 62 | 51 | 72.73 | — | — | — | |
| SiT-S/2 (Uniform Partitioned)Classifier-Free Guidance=Enabled, # Act. Par.=33, GFLOPs=6.06, Boundaries=(0.33, 0.66)2026.06 | 29.83 | — | 57 | 51 | 52.17 | — | — | — | |
| SiT-S/2 (Monolithic)Classifier-Free Guidance=Enabled, # Act. Par.=33, GFLOPs=6.06, Partitioning=Monolithic2026.06 | 30.1 | — | 57 | 50 | 49.82 | — | — | — | |
| CBS (SiT-B/2)Classifier-Free Guidance=Disabled, # Act. Par.=130, GFLOPs=23.01, Boundaries=(0.4, 0.77)2026.06 | 30.51 | — | 55 | 65 | 52.2 | — | — | — | |
| SiT-B/2 (Uniform Partitioned)Classifier-Free Guidance=Disabled, # Act. Par.=130, GFLOPs=23.01, Boundaries=(0.33, 0.66)2026.06 | 34.52 | — | 52 | 64 | 41.45 | — | — | — | |
| SiT-B/2 (Monolithic)Classifier-Free Guidance=Disabled, # Act. Par.=130, GFLOPs=23.01, Partitioning=Monolithic2026.06 | 34.84 | — | 52 | 64 | 41.53 | — | — | — | |
| CBS (SiT-S/2)Classifier-Free Guidance=Disabled, # Act. Par.=33, GFLOPs=6.06, Boundaries=(0.4, 0.77)2026.06 | 50.87 | — | 44 | 62 | 29.57 | — | — | — | |
| SiT-S/2 (Uniform Partitioned)Classifier-Free Guidance=Disabled, # Act. Par.=33, GFLOPs=6.06, Boundaries=(0.33, 0.66)2026.06 | 58.82 | — | 40 | 59 | 23.23 | — | — | — | |
| SiT-S/2 (Monolithic)Classifier-Free Guidance=Disabled, # Act. Par.=33, GFLOPs=6.06, Partitioning=Monolithic2026.06 | 58.97 | — | 40 | 59 | 23.34 | — | — | — | |
| ADMPARAMS=554M, DEC. PARAMS=0, EPOCHS=4002026.02 | — | — | — | — | — | — | 10.94 | 3.94 | |
| DiT-XL/2+SD-VAEPARAMS=675M, DEC. PARAMS=49M, EPOCHS=14002026.02 | — | — | — | — | — | 23.4 | 9.62 | 3.04 | |
| JiT-LPARAMS=459M, DEC. PARAMS=0, EPOCHS=2002026.02 | — | — | — | — | — | — | 16.21 | 2.79 | |
| LF-DiT-LPARAMS=465M, DEC. PARAMS=0, EPOCHS=2002026.02 | — | — | — | — | — | — | 7.2 | 2.48 | |
| LightningDiTPARAMS=675M, DEC. PARAMS=41M, EPOCHS=8002026.02 | — | — | — | — | — | 25.29 | 2.17 | 1.35 | |
| MAR + DetokPARAMS=479M, DEC. PARAMS=86M, EPOCHS=8002026.02 | — | — | — | — | — | 24.06 | 1.86 | 1.35 | |
| MAR + UniflowPARAMS=479M, DEC. PARAMS=300M, EPOCHS=4002026.02 | — | — | — | — | — | 32.48 | 2.45 | 1.85 | |
| RAEPARAMS=839M, DEC. PARAMS=415M, EPOCHS=8002026.02 | — | — | — | — | — | 18.09 | 1.51 | 1.13 | |
| REPA-EPARAMS=675M, DEC. PARAMS=41M, EPOCHS=14802026.02 | — | — | — | — | — | 26.25 | 1.69 | 1.12 | |
| REPA+SD-VAEPARAMS=675M, DEC. PARAMS=49M, EPOCHS=8002026.02 | — | — | — | — | — | 23.4 | 5.9 | 1.42 | |
| SiD UViT/2PARAMS=2B, DEC. PARAMS=02026.02 | — | — | — | — | — | — | 2.77 | 2.44 | |
| SiD2 UViT/2DEC. PARAMS=02026.02 | — | — | — | — | — | — | — | 1.73 | |
| SiT-XL/2+SD-VAEPARAMS=675M, DEC. PARAMS=49M, EPOCHS=14002026.02 | — | — | — | — | — | 23.4 | 8.3 | 2.62 |