Image Generation on ImageNet 512x512
334.7ISFREPix-XL
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| FREPix-XLParams=674M, Epochs=330, NFE=100×22026.05 | 334.7 | 2.38 | — | 80 | 59 | — | — | — | — | |
| PixelU-H/32Params=1152M, Epochs=600, NFE=100×2, GFLOPs=138, Diffusion Space=Pixel-space2026.06 | 322.1 | 1.92 | 5.98 | 80 | 58 | — | — | — | — | |
| DA-VAETraining Regime=Fine-tune, AutoEncoders=DA-VAE (f32c128p1), Token-nums=16×16, Training Epoches=802026.03 | 314.3 | — | — | — | — | — | 0.47 | 4.84 | 1.68 | |
| JiT-H/32Params=756M, Epochs=600, NFE=100×22026.05 | 309.1 | 1.94 | — | — | — | — | — | — | — | |
| JiT-H/32Params=682M, Epochs=600, NFE=100×2, GFLOPs=183, Diffusion Space=Pixel-space2026.06 | 309.1 | 1.94 | — | — | — | — | — | — | — | |
| JiT-G/32NFE=100x2, space=pixel, params=2B, Gflops=3842026.01 | 306.8 | 1.78 | — | — | — | — | — | — | — | |
| DDT-XL/2params=675+49M, Gflops=525, Diffusion Space=Latent, Architecture=XL/22026.01 | 305.1 | 1.28 | — | — | — | — | — | — | — | |
| DeCo-XL/16Params=682M, Epochs=340, NFE=100×22026.05 | 290 | 2.22 | — | 80 | 60 | — | — | — | — | |
| DeCo-XL/16Params=682M, Epochs=340, NFE=100×2, Diffusion Space=Pixel-space2026.06 | 290 | 2.22 | 4.67 | 80 | 60 | — | — | — | — | |
| pMF-H/32NFE=1, space=pixel, params=959M, Gflops=2722026.01 | 284.9 | 2.48 | — | — | — | — | — | — | — | |
| VDM++NFE=256x2, space=pixel, params=2.5B, Gflops=5552026.01 | 278.1 | 2.65 | — | — | — | — | — | — | — | |
| VDM++params=2B, Gflops=555, Diffusion Space=Pixel, Architecture=UViT/42026.01 | 278.1 | 2.65 | — | — | — | — | — | — | — | |
| VDM++CFG=true2026.03 | 278.1 | 2.65 | — | — | — | — | — | — | — | |
| VDM++Params=2B, Epochs=800, NFE=250×22026.05 | 278.1 | 2.65 | — | — | — | — | — | — | — | |
| VDM++CFG scale=w = 1.352026.05 | 278.1 | 2.65 | — | — | — | — | — | — | — | |
| VDM++Params=2B, Epochs=800, NFE=250×2, GFLOPs=555, Diffusion Space=Pixel-space2026.06 | 278.1 | 2.65 | — | — | — | — | — | — | — | |
| DA-VAETraining Regime=Fine-tune, AutoEncoders=DA-VAE (f32c128p1), Token-nums=16×16, Training Epoches=252026.03 | 277.6 | — | — | — | — | — | 0.47 | 6.04 | 2.07 | |
| k-Diffparams=133M, Gflops=26, Diffusion Space=Pixel, Architecture=JiT-B/322026.01 | 277.1 | 4.03 | — | — | — | — | — | — | — | |
| pMF-L/32NFE=1, space=pixel, params=413M, Gflops=1172026.01 | 276.8 | 2.75 | — | — | — | — | — | — | — | |
| SiT-XL/2 + REPANFE=250x2, space=latent, params=675M (49M), Gflops=525 (1230)2026.01 | 274.6 | 2.08 | — | — | — | — | — | — | — | |
| REPAparams=675+49M, Gflops=525, Diffusion Space=Latent, Architecture=SiT-XL/22026.01 | 274.6 | 2.08 | — | — | — | — | — | — | — | |
| REPAEpochs=200, CFG=true2026.03 | 274.6 | 2.08 | 4.19 | 83 | 58 | — | — | — | — | |
| REPATraining Regime=Scratch, Training Epoches=2002026.03 | 274.6 | — | — | — | — | — | — | — | 2.08 | |
| pMF-B/32NFE=1, space=pixel, params=120M, Gflops=342026.01 | 271.9 | 3.7 | — | — | — | — | — | — | — | |
| PixelDiT-XLParams=797M, Epochs=360, NFE=100×22026.05 | 271.1 | 2.21 | — | 78 | 65 | — | — | — | — | |
| JiT-B/32params=133M, Gflops=26, Diffusion Space=Pixel2026.01 | 271 | 4.02 | — | — | — | — | — | — | — | |
| StyleGAN-XLNFE=1, space=pixel, params=168M, Gflops=20612026.01 | 267.8 | 2.41 | — | — | — | — | — | — | — | |
| DiverseDiTEpochs=200, CFG=true2026.03 | 267.12 | 1.99 | 4.19 | 83 | 61 | — | — | — | — | |
| SiT-XL/2 + sREPAEpochs=100, CFG scale=w = 1.352026.05 | 260.3 | 2.29 | 4.18 | 84 | 57 | — | — | — | — | |
| RAE + DiTDH-XL/2NFE=50x2, space=latent, params=831M (415M), Gflops=642 (408)2026.01 | 259.6 | 1.13 | — | — | — | — | — | — | — | |
| RAEparams=839+415M, Gflops=642, Diffusion Space=Latent, Architecture=DiTDH-XL/22026.01 | 259.6 | 1.13 | — | — | — | — | — | — | — | |
| MakDiTEpochs=800, CFG=true2026.03 | 256.3 | 2.5 | 5.1 | 83 | 56 | — | — | — | — | |
| MaskDiTEpochs=800, CFG scale=w = 1.352026.05 | 256.3 | 2.5 | 5.1 | 83 | 56 | — | — | — | — | |
| SiT-XL/2 + REPAEpochs=100, CFG scale=w = 1.352026.05 | 255.7 | 2.32 | 4.16 | 84 | 56 | — | — | — | — | |
| DiT-XLTraining Regime=Scratch, AutoEncoders=SD-VAE (f8c4p2), Token-nums=32×32, Training Epoches=24002026.03 | 255.3 | — | — | — | — | — | 0.48 | 12.04 | 3.04 | |
| LightningDiT-XLTraining Regime=Fine-tune, AutoEncoders=VA-VAE (f16c32p2), Token-nums=16×16, Training Epoches=802026.03 | 254.5 | — | — | — | — | — | 0.5 | 11.31 | 3.12 | |
| SiT-XL/2 + sREPAEpochs=80, CFG scale=w = 1.352026.05 | 254.3 | 2.4 | 4.24 | 84 | 54 | — | — | — | — | |
| SiT-XL/2NFE=250x2, space=latent, params=675M (49M), Gflops=525 (1230)2026.01 | 252.2 | 2.62 | — | — | — | — | — | — | — | |
| SiT-XL/2params=675+49M, Gflops=525, Diffusion Space=Latent, Architecture=XL/22026.01 | 252.2 | 2.62 | — | — | — | — | — | — | — | |
| SiT-XL/2Epochs=600, CFG=true2026.03 | 252.2 | 2.62 | 4.18 | 84 | 57 | — | — | — | — | |
| SiT-XL/2Params=675M + 86M, Epochs=600, NFE=250×22026.05 | 252.2 | 2.62 | — | 84 | 57 | — | — | — | — | |
| SiT-XL/2Epochs=600, CFG scale=w = 1.352026.05 | 252.2 | 2.62 | 4.18 | 84 | 57 | — | — | — | — | |
| SiT-XL/2Params=675M + 86M, Epochs=600, NFE=250×2, GFLOPs=525, Diffusion Space=Latent-space2026.06 | 252.2 | 2.62 | 4.18 | 84 | 57 | — | — | — | — | |
| SiD, UViTNFE=1000x2, space=pixel, params=2.5B, Gflops=5552026.01 | 248.7 | 3.02 | — | — | — | — | — | — | — | |
| SiDparams=2B, Gflops=555, Diffusion Space=Pixel, Architecture=UViT/42026.01 | 248.7 | 3.02 | — | — | — | — | — | — | — | |
| SiT-XL/2 + REPAEpochs=80, CFG scale=w = 1.352026.05 | 247.3 | 2.44 | 4.21 | 84 | 56 | — | — | — | — | |
| PixNerdparams=700 M, Gflops=583, Diffusion Space=Pixel, Architecture=XL/162026.01 | 245.6 | 2.84 | — | — | — | — | — | — | — | |
| PixNerd-XL/16Params=700M, Epochs=340, NFE=100×22026.05 | 245.6 | 2.84 | — | 80 | 59 | — | — | — | — | |
| PixNerd-XL/16Params=700M, Epochs=340, NFE=100×2, GFLOPs=583, Diffusion Space=Pixel-space2026.06 | 245.6 | 2.84 | 5.95 | 80 | 59 | — | — | — | — | |
| DiT w/ Faster DiffusionSampling Steps (T)=250, Image Resolution=512x512, Latency (s/image)=17.352023.12 | 245.13 | 3.25 | 5.05 | 83 | 51 | — | — | — | — | |
| DiverseDiTEpochs=80, CFG=true2026.03 | 241.08 | 2.21 | 4.31 | 82 | 61 | — | — | — | — | |
| DiTSampling Steps (T)=250, Image Resolution=512x512, Latency (s/image)=26.252023.12 | 240.82 | 3.04 | 5.02 | 84 | 54 | — | — | — | — | |
| DiT-XL/2NFE=250x2, space=latent, params=675M (49M), Gflops=525 (1230)2026.01 | 240.8 | 3.04 | — | — | — | — | — | — | — | |
| DiT-XL/2params=675+49M, Gflops=525, Diffusion Space=Latent, Architecture=XL/22026.01 | 240.8 | 3.04 | — | — | — | — | — | — | — | |
| DiT-XL/2Epochs=600, CFG=true2026.03 | 240.8 | 3.04 | 5.02 | 84 | 54 | — | — | — | — | |
| DiT-XL/2Params=675M + 86M, Epochs=600, NFE=250×22026.05 | 240.8 | 3.04 | — | 84 | 54 | — | — | — | — | |
| DiT-XL/2Epochs=600, CFG scale=w = 1.352026.05 | 240.8 | 3.04 | 5.02 | 84 | 54 | — | — | — | — | |
| DiT-XL/2Params=675M + 86M, Epochs=600, NFE=250×2, GFLOPs=525, Diffusion Space=Latent-space2026.06 | 240.8 | 3.04 | 5.02 | 84 | 54 | — | — | — | — | |
| LightningDiT-XL*Training Regime=Scratch, AutoEncoders=VA-VAE (f16c32p2), Token-nums=16×16, Training Epoches=802026.03 | 229.7 | — | — | — | — | — | 0.5 | 21.79 | 3.98 | |
| ADM-GEpochs=400, CFG=true2026.03 | 221.7 | 2.85 | 5.86 | 84 | 53 | — | — | — | — | |
| ADM-G, ADM-UEpochs=400, CFG scale=w = 1.352026.05 | 221.7 | 2.85 | 5.86 | 84 | 53 | — | — | — | — | |
| RINparams=320M, Gflops=415, Diffusion Space=Pixel2026.01 | 216 | 3.95 | — | — | — | — | — | — | — | |
| RINParams=320M, NFE=250, GFLOPs=415, Diffusion Space=Pixel-space2026.06 | 216 | 3.95 | — | — | — | — | — | — | — | |
| RINParams=320M, NFE=2502026.05 | 210 | 3.95 | — | — | — | — | — | — | — | |
| Simple diffusion (U-ViT, L)Epochs=800, CFG scale=w = 1.352026.05 | 205.3 | 4.53 | — | — | — | — | — | — | — | |
| SimpleDiffusionParams=2B, Epochs=800, NFE=250×2, GFLOPs=555, Diffusion Space=Pixel-space2026.06 | 205 | 3.54 | — | — | — | — | — | — | — | |
| CubiDParams (B)=0.95, Tokenizer=DC-AE-f32c32, Token Dimension=322026.03 | 188.7 | — | — | — | — | 1.58 | — | — | — | |
| USiT-2BParams (B)=1.58, Tokenizer=DC-AE-f32c32, Token Dimension=322026.03 | 187.68 | — | — | — | — | 1.72 | — | — | — | |
| USiT-HParams (B)=0.50, Tokenizer=DC-AE-f32c32, Token Dimension=322026.03 | 174.58 | — | — | — | — | 1.89 | — | — | — | |
| ADM-GNFE=250x2, space=pixel, params=559M, Gflops=19832026.01 | 172.7 | 7.72 | — | — | — | — | — | — | — | |
| ADM-Gparams=559M, Gflops=1983, Diffusion Space=Pixel2026.01 | 172.7 | 7.72 | — | — | — | — | — | — | — | |
| ADM-GParams=554M, Epochs=400, NFE=2502026.05 | 172.7 | 7.72 | — | 87 | 53 | — | — | — | — | |
| ADM-GParams=554M, Epochs=400, NFE=250, GFLOPs=1983, Diffusion Space=Pixel-space2026.06 | 172.7 | 7.72 | 6.57 | 87 | 53 | — | — | — | — | |
| Simple diffusion (U-Net)Epochs=800, CFG scale=w = 1.352026.05 | 171 | 4.28 | — | — | — | — | — | — | — | |
| BigGAN-deepNFE=1, space=pixel, params=56M, Gflops=762026.01 | 152.8 | 7.5 | — | — | — | — | — | — | — | |
| DiverseDiTBase Model=REPA-XL, Iter.=400k, CFG=1.352026.03 | 137.09 | 7.18 | 5.38 | 78 | 64 | — | — | — | — | |
| SiT-XLParams (B)=0.67, Tokenizer=DC-AE-f32c32, Token Dimension=322026.03 | 131.37 | — | — | — | — | 2.41 | — | — | — | |
| REPA-XLIter.=400k, CFG=1.352026.03 | 127.83 | 7.91 | 5.41 | 79 | 65 | — | — | — | — | |
| DC-Gen-DiT-XLTraining Regime=Fine-tune, AutoEncoders=DC-AE (f32c32p1), Token-nums=16×16, Training Epoches=802026.03 | 122.5 | — | — | — | — | — | 0.66 | 8.21 | 2.22 | |
| DiT-XLTraining Regime=Scratch, AutoEncoders=DC-AE (f32c32p1), Token-nums=16×16, Training Epoches=24002026.03 | 117.5 | — | — | — | — | — | 0.66 | 9.56 | 2.84 | |
| DiverseDiTBase Model=REPA-L, Iter.=400k, CFG=1.352026.03 | 114 | 9.83 | 5.49 | 78 | 64 | — | — | — | — | |
| REPA-LIter.=400k, CFG=1.352026.03 | 106.43 | 10.82 | 5.52 | 78 | 63 | — | — | — | — | |
| DiC-HG FLOPs (Wino.)=817.2 (388.4), TP=53.3, Training Steps=400K2024.12 | 101.78 | 12.89 | — | — | — | — | — | — | — | |
| DiC-XLG FLOPs (Wino.)=464.3 (228.7), TP=84.2, Training Steps=400K2024.12 | 93.55 | 15.32 | — | — | — | — | — | — | — | |
| Subtraction-based Boundary RF ModelBackbone=DiT-B, Sampler=SDE2025.06 | 80.92 | 14.58 | — | — | — | — | — | — | — | |
| Mask-based Boundary RF ModelBackbone=DiT-B, Sampler=SDE2025.06 | 77.25 | 15.69 | — | — | — | — | — | — | — | |
| DiverseDiTBase Model=SiT-XL, Iter.=400k, CFG=1.352026.03 | 76.45 | 17.68 | 5.48 | 77 | 60 | — | — | — | — | |
| Subtraction-based Boundary RF ModelBackbone=DiT-B, Sampler=Euler2025.06 | 75.96 | 15.08 | — | — | — | — | — | — | — | |
| Vanilla RF ModelBackbone=DiT-B, Sampler=SDE2025.06 | 72.93 | 16.7 | — | — | — | — | — | — | — | |
| DiverseDiTBase Model=SiT-L, Iter.=400k, CFG=1.352026.03 | 71.78 | 19.19 | 5.74 | 76 | 61 | — | — | — | — | |
| SiT-XLIter.=400k, CFG=1.352026.03 | 71.57 | 19.65 | 5.55 | 75 | 60 | — | — | — | — | |
| Mask-based Boundary RF ModelBackbone=DiT-B, Sampler=Euler2025.06 | 70.35 | 15.21 | — | — | — | — | — | — | — | |
| Vanilla RF ModelBackbone=DiT-B, Sampler=Euler2025.06 | 69.92 | 16.35 | — | — | — | — | — | — | — | |
| DiT-XL/2G FLOPs (Wino.)=524.7, TP=16.2, Training Steps=400K2024.12 | 66.3 | 20.94 | — | — | — | — | — | — | — | |
| DiverseDiTBase Model=REPA-B, Iter.=400k, CFG=1.352026.03 | 64.62 | 23.82 | 7.76 | 70 | 63 | — | — | — | — | |
| SiT-LIter.=400k, CFG=1.352026.03 | 64.05 | 22.75 | 5.78 | 73 | 63 | — | — | — | — | |
| REPA-BIter.=400k, CFG=1.352026.03 | 53.92 | 30.13 | 7.79 | 68 | 64 | — | — | — | — | |
| DiverseDiTBase Model=SiT-B, Iter.=400k, CFG=1.352026.03 | 45.09 | 33.18 | 6.92 | 67 | 65 | — | — | — | — | |
| SiT-BIter.=400k, CFG=1.352026.03 | 36.8 | 43.46 | 7.53 | 60 | 64 | — | — | — | — | |
| HSOParadigm=Hierarchical Opt., Prep. Cost=~8s, NFE=42025.11 | — | 15.71 | — | — | — | — | — | — | — |