Class-conditional Image Generation on ImageNet 256x256 2012 (val)
1.28FIDAliTok-XL‡
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| AliTok-XL‡Token Type=Discrete Tokens, Autoregressive Strategy=Standard AR, Tokenizer #Param(Enc+Dec).=390M, Tokenizer Codebook=4K, Generator #Param.=662M2026.05 | 1.28 | 306.3 | 79 | 65 | 1.88 | |
| LightningDiT-XLTokenizer Type=KL, Tokenizer #Param.=70M, Tokenizer #Tokens=256, Generator Type=Diff., Generator #Param.=675M, Generator #Steps=2502026.01 | 1.35 | 295.3 | 0.79 | 0.65 | — | |
| LightningDiT-XL†Token Type=Continuous Tokens, Tokenizer #Param(Enc+Dec).=70M, Generator #Param.=675M2026.05 | 1.35 | 295.3 | 79 | 65 | 2.17 | |
| REPA-SiT-XL†Token Type=Continuous Tokens, Tokenizer #Param(Enc+Dec).=84M, Generator #Param.=675M2026.05 | 1.42 | 305.7 | 80 | 65 | 5.9 | |
| AliTok-B‡Token Type=Discrete Tokens, Autoregressive Strategy=Standard AR, Tokenizer #Param(Enc+Dec).=390M, Tokenizer Codebook=4K, Generator #Param.=177M2026.05 | 1.44 | 319.5 | 77 | 65 | 2.4 | |
| Prologue L–XLToken Type=Discrete Tokens, Autoregressive Strategy=Standard AR, Tokenizer #Param(Enc+Dec).=86+305M, Tokenizer Codebook=1K+4K, Generator #Param.=685M2026.05 | 1.46 | 257.7 | 78 | 66 | 2.26 | |
| RAR-XXLToken Type=Discrete Tokens, Autoregressive Strategy=Non-standard AR / Non-AR, Tokenizer #Param(Enc+Dec).=36+36M, Tokenizer Codebook=1K, Generator #Param.=1.5B2026.05 | 1.48 | 326 | 80 | 63 | — | |
| Prologue L–LToken Type=Discrete Tokens, Autoregressive Strategy=Standard AR, Tokenizer #Param(Enc+Dec).=86+305M, Tokenizer Codebook=1K+4K, Generator #Param.=305M2026.05 | 1.52 | 251.6 | 77 | 66 | 2.81 | |
| MAR-HToken Type=Continuous Tokens, Tokenizer #Param(Enc+Dec).=70M, Generator #Param.=943M2026.05 | 1.55 | 303.7 | 81 | 62 | 2.35 | |
| FlexTok d18-d28†§⋄Token Type=Discrete Tokens, Autoregressive Strategy=Standard AR, Tokenizer #Param(Enc+Dec).=1.4B, Tokenizer Codebook=64K, Generator #Param.=1.4B2026.05 | 1.71 | 284.9 | 82 | 61 | — | |
| MAGVIT-v22026.01 | 1.78 | 319.4 | — | — | — | |
| MAR-L2026.01 | 1.78 | 296 | 81 | 60 | — | |
| MAGVIT-v2Token Type=Discrete Tokens, Autoregressive Strategy=Non-standard AR / Non-AR, Tokenizer Codebook=262K, Generator #Param.=307M2026.05 | 1.78 | 319.4 | — | — | 3.65 | |
| VAR-𝑑30Token Type=Discrete Tokens, Autoregressive Strategy=Non-standard AR / Non-AR, Tokenizer Codebook=4K, Generator #Param.=2.0B2026.05 | 1.92 | 323.1 | 82 | 59 | — | |
| TiTok-S-128★Token Type=Discrete Tokens, Autoregressive Strategy=Non-standard AR / Non-AR, Tokenizer #Param(Enc+Dec).=22+22M, Tokenizer Codebook=4K, Generator #Param.=287M2026.05 | 1.97 | 281.8 | — | — | 4.44 | |
| Mediator2026.01 | 2.01 | 271.04 | 82 | 60 | — | |
| FlexTok d18-d18Tokenizer Type=FSQ, Tokenizer #Param.=950M, Tokenizer #Tokens=32, Generator Type=AR+Flow, Generator #Param.=1.33B, Generator #Steps=26-2812026.01 | 2.02 | — | — | — | — | |
| GigaTok-B-L†Token Type=Discrete Tokens, Autoregressive Strategy=Standard AR, Tokenizer #Param(Enc+Dec).=622M, Tokenizer Codebook=16K, Generator #Param.=1.4B2026.05 | 2.03 | 238.5 | 80 | 63 | — | |
| SiT-XL2026.01 | 2.06 | 277.5 | 83 | 59 | — | |
| DiG-XL/2-G2026.01 | 2.07 | 278.95 | 82 | 60 | — | |
| Prologue L–BToken Type=Discrete Tokens, Autoregressive Strategy=Standard AR, Tokenizer #Param(Enc+Dec).=86+305M, Tokenizer Codebook=1K+4K, Generator #Param.=115M2026.05 | 2.15 | 219.9 | 79 | 60 | 5.02 | |
| VFMTok-SigLIP2-XXL†Token Type=Discrete Tokens, Autoregressive Strategy=Standard AR, Tokenizer Codebook=16K, Generator #Param.=1.4B2026.05 | 2.16 | 272 | 83 | 60 | 1.98 | |
| LINA-Hcfg=2.42026.01 | 2.18 | 275.73 | 81 | 58 | — | |
| DiM-H2026.01 | 2.21 | — | — | — | — | |
| DiT-XL/22026.01 | 2.27 | 278.24 | 83 | 57 | — | |
| DiT-XL/2Tokenizer Type=KL, Tokenizer #Param.=84M, Tokenizer #Tokens=1024, Generator Type=Diff., Generator #Param.=675M, Generator #Steps=2502026.01 | 2.27 | 278.2 | 0.83 | 0.57 | — | |
| DiT-XL/2Token Type=Continuous Tokens, Tokenizer #Param(Enc+Dec).=84M, Generator #Param.=675M2026.05 | 2.27 | 278.2 | 83 | 57 | 9.62 | |
| DiffuSSM-XL-G2026.01 | 2.28 | 259.13 | 86 | 56 | — | |
| U-ViT-H/2-G2026.01 | 2.29 | 263.9 | 82 | 57 | — | |
| MAR-B2026.01 | 2.31 | 281.7 | 82 | 57 | — | |
| MAR-BTokenizer Type=KL, Tokenizer #Param.=66M, Tokenizer #Tokens=256, Generator Type=Mask.+Diff., Generator #Param.=208M, Generator #Steps=642026.01 | 2.31 | 281.7 | 0.82 | 0.57 | — | |
| LiT-XL/22026.01 | 2.32 | 265.2 | 82.4 | 57.4 | — | |
| ResTokTokenizer Type=VQ, Tokenizer #Param.=662M, Tokenizer #Tokens=128, Generator Type=HAR, Generator #Param.=326M, Generator #Steps=92026.01 | 2.34 | 257.8 | 0.79 | 0.6 | — | |
| ResTok†§Token Type=Discrete Tokens, Autoregressive Strategy=Non-standard AR / Non-AR, Tokenizer #Param(Enc+Dec).=662M, Tokenizer Codebook=8K, Generator #Param.=326M2026.05 | 2.34 | 257.8 | 79 | 60 | 4.56 | |
| LlamaGen-XXLToken Type=Discrete Tokens, Autoregressive Strategy=Standard AR, Tokenizer #Param(Enc+Dec).=36+36M, Tokenizer Codebook=16K, Generator #Param.=1.4B2026.05 | 2.34 | 253.9 | 80 | 59 | 14.64 | |
| Prologue B–XLToken Type=Discrete Tokens, Autoregressive Strategy=Standard AR, Tokenizer #Param(Enc+Dec).=86+86M, Tokenizer Codebook=1K+16K, Generator #Param.=685M2026.05 | 2.43 | 252.6 | 80 | 59 | 5.22 | |
| TiTok-B-64★Token Type=Discrete Tokens, Autoregressive Strategy=Non-standard AR / Non-AR, Tokenizer #Param(Enc+Dec).=86+86M, Tokenizer Codebook=4K, Generator #Param.=177M2026.05 | 2.48 | 214.7 | — | — | 3.08 | |
| ImageFolderTokenizer Type=MSRQ, Tokenizer #Param.=176M, Tokenizer #Tokens=286, Generator Type=VAR, Generator #Param.=362M, Generator #Steps=102026.01 | 2.6 | 295 | 0.75 | 0.63 | — | |
| ImageFolder★†§Token Type=Discrete Tokens, Autoregressive Strategy=Non-standard AR / Non-AR, Tokenizer #Param(Enc+Dec).=86+86M, Tokenizer Codebook=4K×2, Generator #Param.=362M2026.05 | 2.6 | 295 | 75 | 63 | — | |
| DiM-L2026.01 | 2.64 | — | — | — | — | |
| Prologue B–LToken Type=Discrete Tokens, Autoregressive Strategy=Standard AR, Tokenizer #Param(Enc+Dec).=86+86M, Tokenizer Codebook=1K+16K, Generator #Param.=305M2026.05 | 2.67 | 251.2 | 82 | 56 | 6.56 | |
| VFMTokTokenizer Type=VQ, Tokenizer #Tokens=256, Generator Type=AR, Generator #Param.=343M, Generator #Steps=2562026.01 | 2.75 | 278.8 | 0.84 | 0.57 | — | |
| TiTok-L-32Tokenizer Type=VQ, Tokenizer #Param.=641M, Tokenizer #Tokens=32, Generator Type=Mask., Generator #Param.=177M, Generator #Steps=82026.01 | 2.77 | 199.8 | — | — | — | |
| IBQ-BTokenizer Type=IBQ, Tokenizer #Param.=128M, Tokenizer #Tokens=256, Generator Type=AR, Generator #Param.=342M, Generator #Steps=2562026.01 | 2.88 | 254.7 | 0.84 | 0.51 | — | |
| FlowAR-BTokenizer Type=KL, Tokenizer #Param.=66M, Tokenizer #Tokens=256, Generator Type=VAR+Flow, Generator #Param.=300M, Generator #Steps=52026.01 | 2.9 | 272.5 | 0.84 | 0.54 | — | |
| DetailFlow-16Tokenizer Type=VQ, Tokenizer #Param.=271M, Tokenizer #Tokens=128, Generator Type=PAR, Generator #Param.=326M, Generator #Steps=232026.01 | 2.96 | 221.4 | 0.82 | 0.57 | — | |
| SpectralAR-d16Tokenizer Type=VQ, Tokenizer #Tokens=64, Generator Type=AR, Generator #Param.=310M, Generator #Steps=642026.01 | 3.02 | 282.2 | 0.81 | 0.55 | — | |
| LlamaGen-LTokenizer Type=VQ, Tokenizer #Param.=72M, Tokenizer #Tokens=576, Generator Type=AR, Generator #Param.=343M, Generator #Steps=5762026.01 | 3.07 | 256.1 | 0.83 | 0.52 | — | |
| GigaTok-B-LTokenizer Type=VQ, Tokenizer #Param.=622M, Tokenizer #Tokens=256, Generator Type=AR, Generator #Param.=111M, Generator #Steps=2562026.01 | 3.26 | 221 | 0.81 | 0.56 | — | |
| VAR-d16Tokenizer Type=MSRQ, Tokenizer #Param.=109M, Tokenizer #Tokens=680, Generator Type=VAR, Generator #Param.=310M, Generator #Steps=102026.01 | 3.3 | 274.4 | 0.84 | 0.51 | — | |
| LDM-42026.01 | 3.6 | 247.67 | 87 | 48 | — | |
| LDM-4-GTokenizer Type=KL, Tokenizer #Param.=55M, Tokenizer #Tokens=4096, Generator Type=Diff., Generator #Param.=400M, Generator #Steps=2502026.01 | 3.6 | 247.7 | — | — | — | |
| LDM-4-GToken Type=Continuous Tokens, Tokenizer #Param(Enc+Dec).=55M, Generator #Param.=400M2026.05 | 3.6 | 247.7 | 87 | 48 | — | |
| PAR-L-4xTokenizer Type=VQ, Tokenizer #Param.=72M, Tokenizer #Tokens=576, Generator Type=PAR, Generator #Param.=343M, Generator #Steps=1472026.01 | 3.76 | 218.9 | 0.84 | 0.5 | — | |
| Prologue B–BToken Type=Discrete Tokens, Autoregressive Strategy=Standard AR, Tokenizer #Param(Enc+Dec).=86+86M, Tokenizer Codebook=1K+16K, Generator #Param.=115M2026.05 | 4.11 | 210.3 | 83 | 48 | 10.75 | |
| LINA-Hcfg=12026.01 | 4.49 | 162.64 | 74 | 62 | — | |
| ADM2026.01 | 4.59 | 186.7 | 82 | 52 | — | |
| CDM2026.01 | 4.88 | 158.71 | — | — | — | |
| LlamaGen-BToken Type=Discrete Tokens, Autoregressive Strategy=Standard AR, Tokenizer #Param(Enc+Dec).=36+36M, Tokenizer Codebook=16K, Generator #Param.=111M2026.05 | 5.46 | 193.6 | 83 | 45 | 26.26 | |
| Mask-GIT2026.01 | 6.18 | 182.1 | — | — | — | |
| MaskGITTokenizer Type=VQ, Tokenizer #Param.=66M, Tokenizer #Tokens=256, Generator Type=Mask., Generator #Param.=227M, Generator #Steps=82026.01 | 6.18 | 182.1 | 0.8 | 0.51 | — | |
| RQTran.Tokenizer Type=RQ, Tokenizer #Param.=66M, Tokenizer #Tokens=256, Generator Type=AR, Generator #Param.=3.8B, Generator #Steps=682026.01 | 7.55 | 134 | — | — | — | |
| VQGANTokenizer Type=VQ, Tokenizer #Param.=23M, Tokenizer #Tokens=256, Generator Type=AR, Generator #Param.=1.4B, Generator #Steps=2562026.01 | 15.78 | 74.3 | — | — | — |