Class-to-image generation on ImageNet 256x256
1.42FIDREPA-XL/2
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| REPA-XL/2Params=675M + 86M, Epochs=800, NFE=250×22026.05 | 1.42 | 305.7 | 0.8 | 0.64 | — | — | |
| PixelDiT-XLParams=797M, Epochs=320, NFE=100×22026.05 | 1.61 | 292.7 | 0.78 | 0.64 | — | — | |
| JiT-G/16Params=2B, Epochs=600, NFE=100×22026.05 | 1.82 | 292.6 | — | — | — | — | |
| DeCo-XL/16Params=682M, Epochs=320, NFE=100×22026.05 | 1.9 | 303 | 0.8 | 0.61 | — | — | |
| FREPix-XLParams=674M, Epochs=320, NFE=100×22026.05 | 1.91 | 295.6 | 0.79 | 0.62 | — | — | |
| PixelFlow-XL/4Params=677M, Epochs=320, NFE=120×22026.05 | 1.98 | 282.1 | 0.81 | 0.6 | — | — | |
| RDMParams=553M + 553M, Epochs=400, NFE=2502026.05 | 1.99 | 260.4 | 0.81 | 0.58 | — | — | |
| SiT-XL/2Params=675M + 86M, Epochs=1400, NFE=250×22026.05 | 2.06 | 284 | 0.83 | 0.59 | — | — | |
| PixNerd-XL/16Params=700M, Epochs=320, NFE=100×22026.05 | 2.15 | 297 | 0.79 | 0.59 | — | — | |
| DiT-XL/2Params=675M + 86M, Epochs=1400, NFE=250×22026.05 | 2.27 | 278.2 | 0.83 | 0.57 | — | — | |
| FREPix-XLParams=674M, Epochs=80, NFE=100×22026.05 | 2.29 | 294.9 | 0.79 | 0.6 | — | — | |
| ADMParams=554M, Epochs=400, NFE=2502026.05 | 4.59 | 186.7 | 0.82 | 0.52 | — | — | |
| PixelGen-XL/16Epochs=80, Diffusion Space=Pixel, Sampler type=Heun, Inference steps=50, CFG setting=off2026.02 | 5.11 | 159.2 | 0.72 | 0.63 | — | — | |
| MaskDiTEpochs=1600, Diffusion Space=Latent, Sampler type=Euler, Inference steps=250, CFG setting=off2026.02 | 5.69 | 177.9 | 0.74 | 0.6 | — | — | |
| REPA-XL/2Epochs=800, Diffusion Space=Latent, Sampler type=Euler, Inference steps=250, CFG setting=off2026.02 | 5.9 | 157.8 | 0.7 | 0.69 | — | — | |
| FractalMAR-HParams=848M, Epochs=6002026.05 | 6.15 | 348.9 | 0.81 | 0.46 | — | — | |
| MDTEpochs=1300, Diffusion Space=Latent, Sampler type=Euler, Inference steps=250, CFG setting=off2026.02 | 6.23 | 143 | 0.71 | 0.65 | — | — | |
| DDT-XL/2Epochs=400, Diffusion Space=Latent, Sampler type=Euler, Inference steps=250, CFG setting=off2026.02 | 6.27 | 154.7 | 0.69 | 0.67 | — | — | |
| DDT-XL/2Epochs=80, Diffusion Space=Latent, Sampler type=Euler, Inference steps=250, CFG setting=off2026.02 | 6.62 | 135.2 | 0.69 | 0.67 | — | — | |
| JetFormerParams=2.8B2026.05 | 6.64 | — | 0.69 | 0.56 | — | — | |
| FasterDiTEpochs=400, Diffusion Space=Latent, Sampler type=Euler, Inference steps=250, CFG setting=off2026.02 | 7.91 | 131.3 | 0.67 | 0.69 | — | — | |
| FlowDCNEpochs=400, Diffusion Space=Latent, Sampler type=Euler, Inference steps=250, CFG setting=off2026.02 | 8.36 | 122.5 | 0.69 | 0.65 | — | — | |
| SiT-XL/2Epochs=1400, Diffusion Space=Latent, Sampler type=Euler, Inference steps=250, CFG setting=off2026.02 | 8.61 | 131.7 | 0.68 | 0.67 | — | — | |
| DiT-XL/2Epochs=1400, Diffusion Space=Latent, Sampler type=Euler, Inference steps=250, CFG setting=off2026.02 | 9.62 | 121.5 | 0.67 | 0.67 | — | — | |
| LDM-4Epochs=170, Diffusion Space=Latent, Sampler type=Euler, Inference steps=250, CFG setting=off2026.02 | 10.56 | 103.5 | 0.71 | 0.62 | — | — | |
| ADMEpochs=400, Diffusion Space=Pixel, CFG setting=off2026.02 | 10.94 | — | 0.69 | 0.63 | — | — | |
| PixelFlow-XL/4Epochs=320, Diffusion Space=Pixel, CFG setting=off2026.02 | 12.23 | 103.3 | 0.63 | 0.66 | — | — | |
| DeCo-XL/16Epochs=320, Diffusion Space=Pixel, CFG setting=off2026.02 | 14.88 | 88.2 | 0.6 | 0.68 | — | — | |
| PixNerd-XL/16Epochs=320, Diffusion Space=Pixel, CFG setting=off2026.02 | 15.61 | 88.9 | 0.59 | 0.68 | — | — | |
| DeCoParams=426M, Train Mem (GB)=27.5, Train Speed (s/it)=0.24, Inference Mem (GB)=2.4, Inference 1 image=0.46s, Inference 1 iter=0.009s2025.11 | 31.35 | 48.35 | — | — | 9.34 | 0.65 | |
| DeCo w/o LFreqFMParams=426M, Train Mem (GB)=27.5, Train Speed (s/it)=0.24, Inference Mem (GB)=2.4, Inference 1 image=0.46s, Inference 1 iter=0.009s2025.11 | 34.12 | 46.44 | — | — | 10.41 | 0.64 | |
| PixNerdParams=458M, Train Mem (GB)=29.2, Train Speed (s/it)=0.25, Inference Mem (GB)=2.6, Inference 1 image=0.48s, Inference 1 iter=0.010s2025.11 | 37.49 | 43.01 | — | — | 10.65 | 0.62 | |
| JiT+REPAParams=459M, Train Mem (GB)=24.8, Train Speed (s/it)=0.23, Inference Mem (GB)=2.5, Inference 1 image=0.46s, Inference 1 iter=0.009s2025.11 | 39.06 | 39.57 | — | — | 11.45 | 0.63 | |
| DiT-L/2Params=458M+ 86M, Train Mem (GB)=28.5, Train Speed (s/it)=0.43, Inference Mem (GB)=3.5, Inference 1 image=0.63s, Inference 1 iter=0.013s2025.11 | 41.93 | 36.52 | — | — | 13.76 | 0.59 | |
| PixDDTParams=434M, Train Mem (GB)=23.6, Train Speed (s/it)=0.22, Inference Mem (GB)=2.4, Inference 1 image=0.49s, Inference 1 iter=0.010s2025.11 | 46.37 | 36.24 | — | — | 17.14 | 0.63 | |
| PixelFlowParams=459M, Train Mem (GB)=73.2, Train Speed (s/it)=1.61, Inference Mem (GB)=3.9, Inference 1 image=6.61s†, Inference 1 iter=0.066s2025.11 | 54.33 | 24.67 | — | — | 9.71 | 0.58 | |
| BaselineParams=459M, Train Mem (GB)=24.8, Train Speed (s/it)=0.22, Inference Mem (GB)=2.5, Inference 1 image=0.46s, Inference 1 iter=0.009s2025.11 | 61.1 | 16.81 | — | — | 15.86 | 0.6 | |
| DeCo w/o REPAParams=426M, Train Mem (GB)=26.4, Train Speed (s/it)=0.22, Inference Mem (GB)=2.4, Inference 1 image=0.46s, Inference 1 iter=0.009s2025.11 | 67.55 | 19.1 | — | — | 10.58 | 0.56 |