Class-conditional image generation on ImageNet 256x256 (train)
362ISIAR-XXL
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| IAR-XXLType=AR, #Para.=1.4B, CFG=2.52025.01 | 362 | 3.74 | — | 86 | 54 | — | — | — | — | — | — | — | |
| VAR-d30-reType=VAR, Epoch=350, #Params=2.0B, rejection sampling=True, Classifier-free guidance=True2024.06 | 356.4 | 1.8 | — | 83 | 57 | — | — | — | — | — | — | — | |
| MaskGIT-reType=Mask., Epoch=300, #Params=227M, Gflops=300, rejection sampling=True, Classifier-free guidance=True2024.06 | 355.6 | 4.02 | — | — | — | — | — | — | — | — | — | — | |
| MaskGIT-reType=Mask, #Para.=227M2025.01 | 355.6 | 4.02 | — | — | — | — | — | — | — | — | — | — | |
| VAREpochs=350, Params=2.0B, Guidance=true2025.12 | 350.2 | — | — | 82 | 60 | — | — | — | 1.73 | — | — | — | |
| LEGO-XL-UArchitecture=Transformer, Classifier-free guidance=true2023.10 | 338.08 | 2.59 | 4.21 | 83 | 56 | 350 | — | — | — | — | — | — | |
| LEGO-XL-UArchitecture=Transformer, Classifier-free guidance=false2023.10 | 337.85 | 5.28 | 4.91 | 67 | 65 | 350 | — | — | — | — | — | — | |
| VAR-d30Type=VAR, Epoch=350, #Params=2.0B, Classifier-free guidance=True2024.06 | 334.7 | 1.97 | — | 81 | 61 | — | — | — | — | — | — | — | |
| MaskBitType=Mask., Epoch=1080, #Params=305M, Classifier-free guidance=True2024.06 | 328.6 | 1.52 | — | — | — | — | — | — | — | — | — | — | |
| MaskBit# Params(G)=305M, Tok. Model=LFQ, Token Type=2D, # Params(T)=54M, # Tokens=256, CFG=true2026.03 | 328.6 | — | — | — | — | — | — | — | 1.52 | 1.61 | — | — | |
| RQTran-reType=AR, Epoch=50, #Params=3.8B, rejection sampling=True, Classifier-free guidance=True2024.06 | 323.7 | 3.8 | — | — | — | — | — | — | — | — | — | — | |
| RQTran.-reType=AR, #Para.=3.8B2025.01 | 323.7 | 3.8 | — | — | — | — | — | — | — | — | — | — | |
| RQ-Trans.# Params(G)=3.8B, Tok. Model=RQ, Token Type=2D, # Params(T)=66M, # Tokens=256, CFG=true2026.03 | 323.7 | — | — | — | — | — | — | — | 3.8 | 3.2 | — | — | |
| VAREpochs=350, Params=2.0B, Guidance=false2025.12 | 323.1 | — | — | 82 | 59 | — | — | — | 1.92 | — | — | — | |
| VAR# Params(G)=2B, Tok. Model=MSRQ, Token Type=2D, # Params(T)=109M, # Tokens=680, CFG=true2026.03 | 323.1 | — | — | 75 | 63 | — | — | — | 1.92 | 0.9 | — | — | |
| REGEpochs=80, #Params=677M, Classifier-Free Guidance (CFG)=true2025.12 | 321.4 | 1.86 | 4.49 | 76 | 63 | — | — | — | — | — | — | — | |
| VAR-d24Type=VAR, Epoch=350, #Params=1.0B, Classifier-free guidance=True2024.06 | 320.1 | 2.33 | — | 82 | 57 | — | — | — | — | — | — | — | |
| MAGVIT-v2#params=307M, Guidance=w/ Guidance2023.12 | 319.4 | 1.78 | — | — | — | — | — | — | — | — | — | — | |
| FlexVAR-d24 w/ SAREpochs=+ 10, Param=1.0B, Step=10, Time=0.52025.12 | 315.5 | 2.14 | — | 81 | 59 | — | — | — | — | — | — | — | |
| REPA-EEpochs=800, #Params=675M, Classifier-Free Guidance (CFG)=true2025.12 | 314.9 | 1.26 | 4.11 | 79 | 66 | — | — | — | — | — | — | — | |
| REPA-EEpochs=800, Params=675M, Guidance=true2025.12 | 314.9 | — | — | 79 | 66 | — | — | — | 1.26 | — | — | — | |
| MDTv2-XL/2#params=676M, Guidance=w/ Guidance2023.12 | 314.7 | 1.58 | — | — | — | — | — | — | — | — | — | — | |
| MDTv2-GType=Mask. + Diff., Epoch=919, #Params=675M, Gflops=119, Classifier-free guidance=True2024.06 | 314.7 | 1.58 | — | 79 | 65 | — | — | — | — | — | — | — | |
| MDTv2Epochs=1080, #Params=675M, Classifier-Free Guidance (CFG)=true2025.12 | 314.7 | 1.58 | 4.52 | 79 | 65 | — | — | — | — | — | — | — | |
| MDTv2Epochs=1080, Params=675M, Guidance=true2025.12 | 314.7 | — | — | 79 | 65 | — | — | — | 1.58 | — | — | — | |
| MDTv2-XL/2# Params(G)=676M, Tok. Model=KL, Token Type=2D, # Params(T)=55M, # Tokens=4096, CFG=true2026.03 | 314.7 | — | — | 79 | 65 | — | — | — | 1.58 | 0.27 | — | — | |
| VAR-d24Epochs=350, Param=1.0B, Step=10, Time=0.52025.12 | 312.9 | 2.33 | — | 82 | 59 | — | — | — | — | — | — | — | |
| MAETok# Params(G)=675M, Tok. Model=AE, Token Type=1D, # Params(T)=176M, # Tokens=128, CFG=true2026.03 | 311.2 | — | — | 81 | 63 | — | — | — | 1.67 | 0.48 | — | — | |
| DDTEpochs=400, Params=675M, Guidance=true2025.12 | 310.6 | — | — | 79 | 65 | — | — | — | 1.26 | — | — | — | |
| MacTok+SiT-XL# Params(G)=675M, Tok. Model=KL, Token Type=1D, # Params(T)=176M, # Tokens=64, CFG=true2026.03 | 310.4 | — | — | 78 | 66 | — | — | — | 1.58 | 0.75 | — | — | |
| MacTok+LightningDiT# Params(G)=675M, Tok. Model=KL, Token Type=1D, # Params(T)=176M, # Tokens=64, CFG=true2026.03 | 307.3 | — | — | 77 | 66 | — | — | — | 1.68 | 0.75 | — | — | |
| VAR-d20Type=VAR, Epoch=250, #Params=600M, Classifier-free guidance=True2024.06 | 306.1 | 2.95 | — | 84 | 53 | — | — | — | — | — | — | — | |
| SD-VAE + SiT-XL/2+REPATokenizer Type=2D, Tokenizer Param.=84M, rFID=0.62, Generator Type=Diff, Generator Param.=675M, # Tokens=1024, Resolution=256x2562026.01 | 305.7 | 1.42 | — | 80 | 65 | — | — | — | — | — | — | — | |
| REPAEpochs=800, #Params=675M, Classifier-Free Guidance (CFG)=true2025.12 | 305.7 | 1.42 | 4.7 | 80 | 65 | — | — | — | — | — | — | — | |
| REPAEpochs=800, Params=675M, Guidance=true2025.12 | 305.7 | — | — | 80 | 64 | — | — | — | 1.42 | — | — | — | |
| SiT-XL/2 + REPA# Params(G)=675M, Tok. Model=KL, Token Type=2D, # Params(T)=84M, # Tokens=1024, CFG=true2026.03 | 305.7 | — | — | 82 | 59 | — | — | — | 1.42 | 0.62 | — | — | |
| REPAArchitecture=Representation-Guided DiT, Generator Params=675M, Epochs=80, Guidance=with Guidance2026.05 | 305.7 | — | — | 80 | 65 | — | — | — | 1.42 | 0.61 | — | — | |
| SpectralAR + SpectralAR-d20Tokenizer Type=Flex, rFID=4.03, Generator Type=AR, Generator Param.=600M, # Tokens=64, Resolution=256x2562026.01 | 305.4 | 2.49 | — | — | — | — | — | — | — | — | — | — | |
| l-DeTokEpochs=800, Params=479M, Guidance=true2025.12 | 304.1 | — | — | 81 | 62 | — | — | — | 1.35 | — | — | — | |
| l-DeTok# Params(G)=479M, Tok. Model=KL, Token Type=2D, # Params(T)=172M, # Tokens=256, CFG=true2026.03 | 304.1 | — | — | 81 | 62 | — | — | — | 1.35 | 0.62 | — | — | |
| MAREpochs=800, Params=943M, Guidance=true2025.12 | 303.7 | — | — | 81 | 62 | — | — | — | 1.55 | — | — | — | |
| MAR-H# Params(G)=943M, Tok. Model=KL, Token Type=2D, # Params(T)=66M, # Tokens=256, CFG=true2026.03 | 303.7 | — | — | 81 | 62 | — | — | — | 1.55 | 1.22 | — | — | |
| TexTok-256# Params(G)=675M, Tok. Model=KL, Token Type=1D, # Params(T)=176M, # Tokens=256, CFG=true2026.03 | 303.1 | — | — | 79 | 64 | — | — | — | 1.46 | 0.73 | — | — | |
| VAR-d20Category=Autoregressive Model, # Avg. Activated Params.=600M2025.03 | 302.6 | 2.57 | — | 83 | 56 | — | — | — | — | — | — | — | |
| VAR + VAR-d20Tokenizer Type=2D, Tokenizer Param.=109M, rFID=0.9, Generator Type=VAR, Generator Param.=600M, # Tokens=680, Resolution=256x2562026.01 | 302.6 | 2.57 | — | 83 | 56 | — | — | — | — | — | — | — | |
| VAR-d20Epochs=250, Param=600M, Step=10, Time=0.32025.12 | 302.6 | 2.95 | — | 83 | 56 | — | — | — | — | — | — | — | |
| MacTok+SiT-XL# Params(G)=675M, Tok. Model=KL, Token Type=1D, # Params(T)=176M, # Tokens=128, CFG=true2026.03 | 302.5 | — | — | 79 | 66 | — | — | — | 1.44 | 0.43 | — | — | |
| RCG, conditional (MAGE-L)#params=512M, Guidance=w/ Guidance2023.12 | 300.7 | 2.25 | — | — | — | — | — | — | — | — | — | — | |
| MacTok+LightningDiT# Params(G)=675M, Tok. Model=KL, Token Type=1D, # Params(T)=176M, # Tokens=128, CFG=true2026.03 | 299.8 | — | — | 78 | 67 | — | — | — | 1.5 | 0.43 | — | — | |
| FlexVAR-d20Epochs=250, Param=600M, Step=10, Time=0.32025.12 | 299.3 | 2.41 | — | 85 | 58 | — | — | — | — | — | — | — | |
| FlexVAR-d24Epochs=350, Param=1.0B, Step=10, Time=0.52025.12 | 299.1 | 2.21 | — | 83 | 59 | — | — | — | — | — | — | — | |
| PixNerdEpochs=160, #Params=700M, Classifier-Free Guidance (CFG)=true2025.12 | 297 | 2.15 | 4.55 | 79 | 59 | — | — | — | — | — | — | — | |
| PixNerdEpochs=160, Params=700M, Guidance=true2025.12 | 297 | — | — | 79 | 59 | — | — | — | 2.15 | — | — | — | |
| REGEpochs=480, #Params=677M, Classifier-Free Guidance (CFG)=true2025.12 | 296.9 | 1.4 | 4.24 | 77 | 66 | — | — | — | — | — | — | — | |
| PAE (DINOv2)Architecture=Representation-Native DiT, Generator Params=675M, Epochs=800, Guidance=with Guidance2026.05 | 296.9 | — | — | 79 | 67 | — | — | — | 1.03 | 0.26 | — | — | |
| VA-VAE + LightningDiTTokenizer Type=2D, Tokenizer Param.=70M, rFID=0.28, Generator Type=Diff, Generator Param.=675M, # Tokens=256, Resolution=256x2562026.01 | 295.3 | 1.35 | — | 79 | 65 | — | — | — | — | — | — | — | |
| LightningDiTEpochs=800, #Params=675M, Classifier-Free Guidance (CFG)=true2025.12 | 295.3 | 1.35 | 4.15 | 79 | 65 | — | — | — | — | — | — | — | |
| VA-VAEEpochs=800, Params=675M, Guidance=true2025.12 | 295.3 | — | — | 79 | 65 | — | — | — | 1.35 | — | — | — | |
| LightningDiT# Params(G)=675M, Tok. Model=KL, Token Type=2D, # Params(T)=70M, # Tokens=256, CFG=true2026.03 | 295.3 | — | — | — | — | — | — | — | 1.35 | 0.28 | — | — | |
| ReDiArchitecture=Representation-Guided DiT, Generator Params=675M, Epochs=800, Guidance=with Guidance2026.05 | 295.1 | — | — | 78 | 64 | — | — | — | 1.61 | 0.58 | — | — | |
| ImageFolder + ImageFolder-VARTokenizer Type=2D, Tokenizer Param.=176M, rFID=0.8, Generator Type=VAR, Generator Param.=362M, # Tokens=286, Resolution=256x2562026.01 | 295 | 2.6 | — | 75 | 63 | — | — | — | — | — | — | — | |
| ImageFolder# Params(G)=362M, Tok. Model=MSRQ, Token Type=1D, # Params(T)=176M, # Tokens=286, CFG=true2026.03 | 295 | — | — | 75 | 63 | — | — | — | 2.6 | 0.8 | — | — | |
| AFMEpochs=800, Params=675M, Guidance=true2025.12 | 293.6 | — | — | 79 | 65 | — | — | — | 1.37 | — | — | — | |
| FlexVAR-d20 w/ SAREpochs=+ 10, Param=600M, Step=10, Time=0.32025.12 | 293.3 | 2.35 | — | 82 | 59 | — | — | — | — | — | — | — | |
| TREADEpochs=740, #Params=675M, Classifier-Free Guidance (CFG)=true2025.12 | 292.7 | 1.69 | 4.73 | 81 | 63 | — | — | — | — | — | — | — | |
| JiTEpochs=600, Params=2.0B, Guidance=true2025.12 | 292.6 | — | — | — | — | — | — | — | 1.82 | — | — | — | |
| DiMR-G/2RType=Diff., Epoch=800, #Params=1.1B, Gflops=331, Classifier-free guidance=True2024.06 | 292.5 | 1.63 | — | 79 | 63 | — | — | — | — | — | — | — | |
| FlexVAR-d16Epochs=180, Param=310M, Step=10, Time=0.22025.12 | 291.3 | 3.05 | — | 83 | 52 | — | — | — | — | — | — | — | |
| RecTokEpochs=600, Params=839M, Guidance=true2025.12 | 289.2 | — | — | 79 | 67 | — | — | — | 1.13 | — | — | — | |
| LEGO-XL-PGArchitecture=Transformer, Classifier-free guidance=true2023.10 | 289.12 | 2.05 | 4.77 | 84 | 55 | 512 | — | — | — | — | — | — | |
| DiMR-XL/2RType=Diff., Epoch=800, #Params=505M, Gflops=160, Classifier-free guidance=True2024.06 | 289 | 1.7 | — | 79 | 63 | — | — | — | — | — | — | — | |
| PAE (SigLIP2)Architecture=Representation-Native DiT, Generator Params=675M, Epochs=800, Guidance=with Guidance2026.05 | 287.4 | — | — | 77 | 68 | — | — | — | 1.07 | 0.27 | — | — | |
| DiMR-XL/2RType=Diff., Epoch=400, #Params=505M, Gflops=160, Classifier-free guidance=True2024.06 | 285.7 | 1.77 | — | 79 | 62 | — | — | — | — | — | — | — | |
| DiffMoE-L-E8-DDPM-GCategory=Diffusion Model, Guidance=with, cfg=1.5, Sampling=DDPM, # Avg. Activated Params.=458M2025.03 | 284.78 | 2.3 | — | 82 | 59 | — | — | — | — | — | — | — | |
| DiffMoE-L-E8-DDPM-GAvg. Activated Params=458M, Diffusion Paradigm=DDPM, Guidance=with guidance, cfg=1.5, Training Steps=7000K2025.03 | 284.78 | 2.3 | — | 82 | 59 | — | — | — | — | — | — | — | |
| LEGO-XL-PRArchitecture=Transformer, Classifier-free guidance=true2023.10 | 284.73 | 2.35 | 5.21 | 83 | 60 | 512 | — | — | — | — | — | — | |
| MDT-XL/2Architecture=Transformer, Classifier-free guidance=true2023.10 | 283.01 | 1.79 | 4.57 | 81 | 60 | 1,664 | — | — | — | — | — | — | |
| MDT-GType=Mask. + Diff., Epoch=1299, #Params=676M, Gflops=119, Classifier-free guidance=True2024.06 | 283 | 1.79 | — | 81 | 61 | — | — | — | — | — | — | — | |
| SpectralAR + SpectralAR-d16Tokenizer Type=Flex, rFID=4.03, Generator Type=AR, Generator Param.=310M, # Tokens=64, Resolution=256x2562026.01 | 282.2 | 3.02 | — | 81 | 55 | — | — | — | — | — | — | — | |
| PixelFlowEpochs=320, #Params=677M, Classifier-Free Guidance (CFG)=true2025.12 | 282.1 | 1.98 | 5.83 | 81 | 60 | — | — | — | — | — | — | — | |
| PixelFlowEpochs=320, Params=677M, Guidance=true2025.12 | 282.1 | — | — | 81 | 60 | — | — | — | 1.98 | — | — | — | |
| TiTok-S-128Type=Mask., Epoch=800, #Params=287M, Classifier-free guidance=True2024.06 | 281.8 | 1.97 | — | — | — | — | — | — | — | — | — | — | |
| TiTok-S + MaskGIT-UViT-LTokenizer Type=1D, Tokenizer Param.=72M, rFID=1.71, Generator Type=Mask, Generator Param.=287M, # Tokens=128, Resolution=256x2562026.01 | 281.8 | 1.87 | — | — | — | — | — | — | — | — | — | — | |
| TiTok-S-128# Params(G)=287M, Tok. Model=VQ, Token Type=1D, # Params(T)=72M, # Tokens=128, CFG=true2026.03 | 281.8 | — | — | — | — | — | — | — | 1.97 | 1.61 | — | — | |
| VAE + MAR-BTokenizer Type=2D, Tokenizer Param.=66M, rFID=0.53, Generator Type=AR+Diff, Generator Param.=208M, # Tokens=256, Resolution=256x2562026.01 | 281.7 | 2.31 | — | 82 | 57 | — | — | — | — | — | — | — | |
| VAR-d16Epochs=180, Param=310M, Step=10, Time=0.22025.12 | 280.4 | 3.55 | — | 84 | 51 | — | — | — | — | — | — | — | |
| VQGAN-reType=AR, Epoch=100, #Params=1.4B, rejection sampling=True, Classifier-free guidance=True2024.06 | 280.3 | 5.2 | — | — | — | — | — | — | — | — | — | — | |
| VQGAN-reType=AR, #Para.=1.4B2025.01 | 280.3 | 5.2 | — | — | — | — | — | — | — | — | — | — | |
| Dense-DiT-XL-DDPM-G (reproduced)Category=Diffusion Model, Guidance=with, cfg=1.5, Sampling=DDPM, Reproduction=true, # Avg. Activated Params.=675M2025.03 | 279.18 | 2.32 | — | 83 | 57 | — | — | — | — | — | — | — | |
| Dense-DiT-XL-DDPM-GAvg. Activated Params=675M, Diffusion Paradigm=DDPM, Guidance=with guidance, cfg=1.5, Training Steps=7000K, Source=Our reproduction2025.03 | 279.18 | 2.32 | — | 83 | 57 | — | — | — | — | — | — | — | |
| SoftVQ-VAE# Params(G)=675M, Tok. Model=SoftVQ, Token Type=1D, # Params(T)=176M, # Tokens=64, CFG=true2026.03 | 279 | — | — | 80 | 63 | — | — | — | 1.78 | 0.88 | — | — | |
| IAR-XXLType=AR, #Para.=1.4B, CFG=1.72025.01 | 278.9 | 2.19 | — | 81 | 58 | — | — | — | — | — | — | — | |
| DiT-XL/2Architecture=Transformer, Classifier-free guidance=true2023.10 | 278.24 | 2.27 | 4.6 | 83 | 57 | 1,792 | — | — | — | — | — | — | |
| DiT-XL/2Depth=28, #Param=675 M, Iters=7,000 K2024.12 | 278.24 | 2.27 | 4.6 | 83 | 57 | — | — | 6.91 | — | — | — | — | |
| DiT-XL/2#params=675M, Guidance=w/ Guidance2023.12 | 278.2 | 2.27 | — | — | — | — | — | — | — | — | — | — | |
| DiT-XL/2Type=Diff., Epoch=1399, #Params=675M, Gflops=119, Classifier-free guidance=True2024.06 | 278.2 | 2.27 | — | 83 | 57 | — | — | — | — | — | — | — | |
| DiT-XL/2Type=Diffusion, #Para.=675M2025.01 | 278.2 | 2.27 | — | 83 | 57 | — | — | — | — | — | — | — | |
| Dense-DiT-XL-DDPM-GCategory=Diffusion Model, Guidance=with, cfg=1.5, Sampling=DDPM, Source=Peebles & Xie, 2023a, # Avg. Activated Params.=675M2025.03 | 278.2 | 2.27 | — | 83 | 57 | — | — | — | — | — | — | — | |
| Dense-DiT-XL-DDPM-GAvg. Activated Params=675M, Diffusion Paradigm=DDPM, Guidance=with guidance, cfg=1.5, Training Steps=7000K, Source=Original (Peebles & Xie, 2023a)2025.03 | 278.2 | 2.27 | — | 83 | 57 | — | — | — | — | — | — | — | |
| SD-VAE + DiT-XL/2Tokenizer Type=2D, Tokenizer Param.=84M, rFID=0.62, Generator Type=Diff, Generator Param.=675M, # Tokens=1024, Resolution=256x2562026.01 | 278.2 | 2.27 | — | 83 | 57 | — | — | — | — | — | — | — |