Image Generation on ImageNet 512x512 (val)
1.366FID-50KSiD2A
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| SiD2ACFG=N, NFE=1, Model Scale=XXL, Parameters=1.5B2024.10 | 1.366 | — | — | — | — | — | — | |
| SiD2ACFG=N, NFE=1, Model Scale=XL, Parameters=1.1B2024.10 | 1.379 | — | — | — | — | — | — | |
| EDM2-XXL (interval)NFE=632024.10 | 1.4 | — | — | — | — | — | — | |
| EDM2-XXLNFE=32×1.2, #Params=1.5B, Guidance Interval=true2025.03 | 1.4 | — | — | — | — | — | — | |
| SiD2ACFG=N, NFE=1, Model Scale=L, Parameters=777M2024.10 | 1.413 | — | — | — | — | — | — | |
| SiDACFG=N, NFE=1, Model Scale=XL, Parameters=1.1B2024.10 | 1.428 | — | — | — | — | — | — | |
| SiD2, Flop HeavyNFE=5122024.10 | 1.48 | — | — | — | — | — | — | |
| SiD2ACFG=N, NFE=1, Model Scale=M, Parameters=498M2024.10 | 1.488 | — | — | — | — | — | — | |
| SiD2, Flop Heavy distilledNFE=162024.10 | 1.5 | — | — | — | — | — | — | |
| SiDACFG=N, NFE=1, Model Scale=L, Parameters=777M2024.10 | 1.501 | — | — | — | — | — | — | |
| SiDACFG=N, NFE=1, Model Scale=XXL, Parameters=1.5B2024.10 | 1.503 | — | — | — | — | — | — | |
| SiDACFG=N, NFE=1, Model Scale=M, Parameters=498M2024.10 | 1.546 | — | — | — | — | — | — | |
| SiD2ACFG=N, NFE=1, Model Scale=S, Parameters=280M2024.10 | 1.669 | — | — | — | — | — | — | |
| EDM2-S (interval)NFE=632024.10 | 1.68 | — | — | — | — | — | — | |
| xAR-L#params=608M2025.02 | 1.7 | 281.5 | — | — | — | — | — | |
| MARNFE=256×2, #Params=481M2025.03 | 1.73 | — | — | — | — | — | — | |
| SiDACFG=N, NFE=1, Model Scale=S, Parameters=280M2024.10 | 1.756 | — | — | — | — | — | — | |
| eMIGM-LNFE=64×1.25, #Params=478M2025.03 | 1.77 | — | — | — | — | — | — | |
| PaGoDANFE=12024.10 | 1.8 | — | — | — | — | — | — | |
| EDM2CFG=Y, NFE=126, Model Scale=XXL, Parameters=1.5B2024.10 | 1.81 | — | — | — | — | — | — | |
| EDM2-XXLNFE=32×2, #Params=1.5B2025.03 | 1.81 | — | — | — | — | — | — | |
| EDM2CFG=Y, NFE=126, Model Scale=XL, Parameters=1.1B2024.10 | 1.85 | — | — | — | — | — | — | |
| EDM2CFG=Y, NFE=126, Model Scale=L, Parameters=777M2024.10 | 1.88 | — | — | — | — | — | — | |
| sCDCFG=Y, NFE=2, Model Scale=XXL, Parameters=1.5B2024.10 | 1.88 | — | — | — | — | — | — | |
| sCD-XXLNFE=2, #Params=1.5B2025.03 | 1.88 | — | — | — | — | — | — | |
| SiDCFG=N, NFE=1, Model Scale=XL, Parameters=1.1B2024.10 | 1.888 | — | — | — | — | — | — | |
| SiDCFG=N, NFE=1, Model Scale=L, Parameters=777M2024.10 | 1.907 | — | — | — | — | — | — | |
| EDM2CFG=N, NFE=63, Model Scale=XXL, Parameters=1.5B2024.10 | 1.91 | — | — | — | — | — | — | |
| MAGVIT-v2NFE=642024.10 | 1.91 | — | — | — | — | — | — | |
| sCDCFG=Y, NFE=2, Model Scale=XL, Parameters=1.1B2024.10 | 1.93 | — | — | — | — | — | — | |
| EDM2CFG=N, NFE=63, Model Scale=XL, Parameters=1.1B2024.10 | 1.96 | — | — | — | — | — | — | |
| SiDCFG=N, NFE=1, Model Scale=XXL, Parameters=1.5B2024.10 | 1.969 | — | — | — | — | — | — | |
| EDM2CFG=Y, NFE=126, Model Scale=M, Parameters=498M2024.10 | 2.01 | — | — | — | — | — | — | |
| sCDCFG=Y, NFE=2, Model Scale=L, Parameters=777M2024.10 | 2.04 | — | — | — | — | — | — | |
| EDM2CFG=N, NFE=63, Model Scale=L, Parameters=777M2024.10 | 2.06 | — | — | — | — | — | — | |
| SiDCFG=N, NFE=1, Model Scale=M, Parameters=498M2024.10 | 2.06 | — | — | — | — | — | — | |
| REPA#params=675M, Backbone=DINOv22025.02 | 2.08 | 274.6 | — | — | — | — | — | |
| SiD2ACFG=N, NFE=1, Model Scale=XS, Parameters=125M2024.10 | 2.156 | — | — | — | — | — | — | |
| SiD2, smallNFE=5122024.10 | 2.19 | — | — | — | — | — | — | |
| eMIGM-LNFE=16×1.2, #Params=478M2025.03 | 2.19 | — | — | — | — | — | — | |
| SiDACFG=N, NFE=1, Model Scale=XS, Parameters=125M2024.10 | 2.228 | — | — | — | — | — | — | |
| EDM2CFG=Y, NFE=126, Model Scale=S, Parameters=280M2024.10 | 2.23 | — | — | — | — | — | — | |
| SiT-XL/2 + MGBackbone=SiT-XL/2, Guidance=Model-guidance (MG)2025.02 | 2.24 | 276.9 | 4.03 | 86 | — | 60 | — | |
| EDM2CFG=N, NFE=63, Model Scale=M, Parameters=498M2024.10 | 2.25 | — | — | — | — | — | — | |
| sCDCFG=Y, NFE=2, Model Scale=M, Parameters=498M2024.10 | 2.26 | — | — | — | — | — | — | |
| sCDCFG=Y, NFE=1, Model Scale=XXL, Parameters=1.5B2024.10 | 2.28 | — | — | — | — | — | — | |
| eMIGM-BNFE=64×1.25, #Params=244M2025.03 | 2.31 | — | — | — | — | — | — | |
| DiP-XL/32Model Type=Pixel Generative Models, Params=631M, CFG=With CFG2025.11 | 2.31 | 291.68 | 4.48 | 84 | — | 58 | — | |
| sCDCFG=Y, NFE=1, Model Scale=XL, Parameters=1.1B2024.10 | 2.4 | — | — | — | — | — | — | |
| StyleGAN-XLNFE=12024.10 | 2.4 | — | — | — | — | — | — | |
| StyleGAN-XLModel=GAN, Label fraction (#labels/class)=100%2023.02 | 2.41 | 267.75 | — | — | — | — | — | |
| StyleGAN-XLNFE=1×22025.03 | 2.41 | — | — | — | — | — | — | |
| FlowDCN-XLModel Type=Latent Generative Models, Params=618M+86M, CFG=With CFG2025.11 | 2.44 | 252.8 | 4.53 | 84 | — | 54 | — | |
| sCDCFG=Y, NFE=2, Model Scale=S, Parameters=280M2024.10 | 2.5 | — | — | — | — | — | — | |
| MaskDiT-GModel Type=Latent Generative Models, Params=675M+86M, CFG=With CFG2025.11 | 2.5 | 256.3 | 5.1 | 83 | — | 56 | — | |
| Large-DiTNFE=250×2, #Params=3B2025.03 | 2.52 | — | — | — | — | — | — | |
| sCDCFG=Y, NFE=1, Model Scale=L, Parameters=777M2024.10 | 2.55 | — | — | — | — | — | — | |
| EDM2CFG=N, NFE=63, Model Scale=S, Parameters=280M2024.10 | 2.56 | — | — | — | — | — | — | |
| SiT-XL/2Backbone=SiT-XL/22025.02 | 2.62 | 252.2 | 4.18 | 84 | — | 57 | — | |
| SiT-XLNFE=250×2, #Params=675M2025.03 | 2.62 | — | — | — | — | — | — | |
| SiT-XLModel Type=Latent Generative Models, Params=675M+86M, CFG=With CFG2025.11 | 2.62 | 252.2 | 4.18 | 84 | — | 57 | — | |
| VAR-d36#params=2.3B2025.02 | 2.63 | 303.2 | — | — | — | — | — | |
| VAR-d36-sNFE=10×2, #Params=2.3B2025.03 | 2.63 | — | — | — | — | — | — | |
| FreqWarm (USiT-H + DC-AE-f64c128)Diffusion Model=USiT-H, Autoencoder=DC-AE-f64c128, Warm-up=w/ FreqWarm, Params (B)=0.50, CFG=true, CFG Scale=1.52025.11 | 2.64 | 276.28 | — | — | — | — | — | |
| VDM++NFE=2502024.10 | 2.65 | — | — | — | — | — | — | |
| VDM++NFE=512×2, #Params=2B2025.03 | 2.65 | — | — | — | — | — | — | |
| VDM++Model Type=Pixel Generative Models, Params=2.46B, CFG=With CFG2025.11 | 2.65 | 278.1 | — | — | — | — | — | |
| DiffiTNFE=2502024.10 | 2.67 | — | — | — | — | — | — | |
| SiDCFG=N, NFE=1, Model Scale=S, Parameters=280M2024.10 | 2.707 | — | — | — | — | — | — | |
| OUSAC w/o cacheSteps=50, CFG-Steps=9, MACs (T)=30.94, Latency (s)=12.912025.12 | 2.71 | 228.3 | 4.38 | 83.43 | — | 57 | — | |
| OUSACSteps=50, CFG-Steps=9, MACs (T)=24.93, Latency (s)=11.282025.12 | 2.72 | 228.6 | 4.13 | 83.62 | — | 55.3 | — | |
| sCDCFG=Y, NFE=1, Model Scale=M, Parameters=498M2024.10 | 2.75 | — | — | — | — | — | — | |
| FreqWarm (UViT-H + DC-AE-f64c128)Diffusion Model=UViT-H, Autoencoder=DC-AE-f64c128, Warm-up=w/ FreqWarm, Params (B)=0.50, CFG=true, CFG Scale=1.52025.11 | 2.76 | 246.89 | — | — | — | — | — | |
| DiT-XL/2 + MGBackbone=DiT-XL/2, Guidance=Model-guidance (MG)2025.02 | 2.78 | 257.2 | 4.86 | 83 | — | 58 | — | |
| eMIGM-BNFE=16×1.2, #Params=244M2025.03 | 2.78 | — | — | — | — | — | — | |
| FreqWarm (DiT-XL + DC-AE-f32c128)Diffusion Model=DiT-XL, Autoencoder=DC-AE-f32c128, Warm-up=w/ FreqWarm, Params (B)=0.67, CFG=true, CFG Scale=1.52025.11 | 2.87 | 240.05 | — | — | — | — | — | |
| DiMR-XL/3R#params=525M2025.02 | 2.89 | 289.8 | — | — | — | — | — | |
| EDM2CFG=Y, NFE=126, Model Scale=XS, Parameters=125M2024.10 | 2.91 | — | — | — | — | — | — | |
| DiT-XL + Asym-VAE-f8Diffusion Model=DiT-XL, Autoencoder=Asym-VAE-f8, Warm-up=None, Params (B)=0.67, CFG=true, CFG Scale=1.52025.11 | 2.97 | — | — | — | — | — | — | |
| DDIMSteps=1000, CFG-Steps=1000, MACs (T)=1049.1, Latency (s)=416.792025.12 | 2.99 | 210.6 | 4.38 | 83.17 | — | 55.6 | — | |
| SD, U-ViT-LNFE=5122024.10 | 3.02 | — | — | — | — | — | — | |
| SimDiffNFE=512×2, #Params=2B2025.03 | 3.02 | — | — | — | — | — | — | |
| SiDModel Type=Pixel Generative Models, Params=2.00B, CFG=With CFG2025.11 | 3.02 | 248.7 | — | — | — | — | — | |
| DiT-XL/2-GModel=Diff., Label fraction (#labels/class)=100%2023.02 | 3.04 | 240.82 | — | — | — | — | — | |
| DiT-XL/2Backbone=DiT-XL/22025.02 | 3.04 | 240.8 | 5.02 | 84 | — | 54 | — | |
| DiT-XL/2#params=675M2025.02 | 3.04 | 240.8 | — | — | — | — | — | |
| DiT-XL/2NFE=2502024.10 | 3.04 | — | — | — | — | — | — | |
| DiT-XL + SD-VAE-f8c4Diffusion Model=DiT-XL, Autoencoder=SD-VAE-f8c4, Warm-up=None, Params (B)=0.67, CFG=true, CFG Scale=1.52025.11 | 3.04 | — | — | — | — | — | — | |
| DiT-XL/2NFE=250×2, #Params=675M2025.03 | 3.04 | — | — | — | — | — | — | |
| DiT-XLModel Type=Latent Generative Models, Params=675M+86M, CFG=With CFG2025.11 | 3.04 | 240.8 | 5.02 | 84 | — | 54 | — | |
| sCDCFG=Y, NFE=1, Model Scale=S, Parameters=280M2024.10 | 3.07 | — | — | — | — | — | — | |
| DiT-XL + DC-AE-f32c128Diffusion Model=DiT-XL, Autoencoder=DC-AE-f32c128, Warm-up=None, Params (B)=0.67, CFG=true, CFG Scale=1.52025.11 | 3.11 | 227.69 | — | — | — | — | — | |
| USiT-H + DC-AE-f64c128Diffusion Model=USiT-H, Autoencoder=DC-AE-f64c128, Warm-up=None, Params (B)=0.50, CFG=true, CFG Scale=1.52025.11 | 3.13 | 207.66 | — | — | — | — | — | |
| FreqWarm (USiT-2B + DC-AE-f64c128)Diffusion Model=USiT-2B, Autoencoder=DC-AE-f64c128, Warm-up=w/ FreqWarm, Params (B)=1.58, CFG=true, CFG Scale=1.52025.11 | 3.18 | 311.79 | — | — | — | — | — | |
| FreqWarm (USiT-H + DC-AE-f32c128)Diffusion Model=USiT-H, Autoencoder=DC-AE-f32c128, Warm-up=w/ FreqWarm, Params (B)=0.50, CFG=true, CFG Scale=1.52025.11 | 3.2 | 244.21 | — | — | — | — | — | |
| DDIMSteps=50, CFG-Steps=50, MACs (T)=52.45, Latency (s)=20.862025.12 | 3.2 | 203.8 | 4.53 | 83.27 | — | 56.4 | — | |
| UViT-H + DC-AE-f64c128Diffusion Model=UViT-H, Autoencoder=DC-AE-f64c128, Warm-up=None, Params (B)=0.50, CFG=true, CFG Scale=1.52025.11 | 3.23 | 219.3 | — | — | — | — | — | |
| eMIGM-SNFE=64×1.25, #Params=132M2025.03 | 3.29 | — | — | — | — | — | — | |
| OUSAC w/o cacheSteps=30, CFG-Steps=8, MACs (T)=19.93, Latency (s)=8.22025.12 | 3.33 | 214.8 | 4.85 | 82.99 | — | 55.3 | — | |
| SiDCFG=N, NFE=1, Model Scale=XS, Parameters=125M2024.10 | 3.353 | — | — | — | — | — | — |