Class-conditional image generation on ImageNet 15 (val)
1.36gFIDRAR-XXL(DINOv2)
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| RAR-XXL(DINOv2)Type=AR., #Para.=1.5B, #Tok.=256, Generation Mode=w/ CFG2026.05 | 1.36 | 5.85 | 301.3 | 78 | 66 | |
| RAR-XL(DINOv2)Type=AR., #Epoch=400, #Para.=955M, #Tok.=256, Generation Mode=w/ CFG2026.05 | 1.38 | 5.86 | 310.2 | 78 | 65 | |
| RAR-XXL(SigLIP)Type=AR., #Para.=1.5B, #Tok.=256, Generation Mode=w/ CFG2026.05 | 1.38 | 6.18 | 298.4 | 78 | 65 | |
| RAR-XL(SigLIP)Type=AR., #Para.=955M, #Tok.=256, Generation Mode=w/ CFG2026.05 | 1.4 | 6.39 | 311.7 | 78 | 65 | |
| RAR-XXL(SigLIP2)Type=AR., #Para.=955M, #Tok.=256, Generation Mode=w/ CFG2026.05 | 1.43 | 6.07 | 291 | 79 | 65 | |
| RAR-L(DINOv2)Type=AR., #Para.=461M, #Tok.=256, Generation Mode=w/ CFG2026.05 | 1.44 | 6.03 | 312.8 | 78 | 66 | |
| RAR-L(SigLIP)Type=AR., #Epoch=400, #Para.=461M, #Tok.=256, Generation Mode=w/ CFG2026.05 | 1.46 | 6.18 | 312.9 | 78 | 64 | |
| RAR-XL(SigLIP2)Type=AR., #Para.=955M, #Tok.=256, Generation Mode=w/ CFG2026.05 | 1.46 | 6.02 | 288.4 | 78 | 65 | |
| RAR-XXLType=AR., #Para.=1.5B, #Tok.=256, Generation Mode=w/ CFG2026.05 | 1.48 | — | 326 | 80 | 63 | |
| RAR-XLType=AR., #Para.=955M, #Tok.=256, Generation Mode=w/ CFG2026.05 | 1.5 | — | 306.9 | 80 | 62 | |
| RAR-L(SigLIP2)Type=AR., #Epoch=400, #Para.=461M, #Tok.=256, Generation Mode=w/ CFG2026.05 | 1.5 | 6.37 | 292.7 | 77 | 66 | |
| RAR-XXL(DINOv2)Type=AR., #Para.=1.5B, #Tok.=256, Generation Mode=w/o CFG2026.05 | 1.65 | 5.55 | 253.7 | 80 | 63 | |
| RAR-XXL(SigLIP2)Type=AR., #Para.=955M, #Tok.=256, Generation Mode=w/o CFG2026.05 | 1.65 | 5.44 | 266.5 | 81 | 63 | |
| RAR-XXL(SigLIP)Type=AR., #Para.=1.5B, #Tok.=256, Generation Mode=w/o CFG2026.05 | 1.68 | 5.64 | 239.5 | 79 | 63 | |
| RAR-LType=AR., #Epoch=400, #Para.=461M, #Tok.=256, Generation Mode=w/ CFG2026.05 | 1.7 | — | 299.5 | 82 | 58 | |
| RAR-XL(SigLIP2)Type=AR., #Para.=955M, #Tok.=256, Generation Mode=w/o CFG2026.05 | 1.72 | 5.41 | 244.8 | 80 | 63 | |
| RAR-XL(DINOv2)Type=AR., #Epoch=400, #Para.=955M, #Tok.=256, Generation Mode=w/o CFG2026.05 | 1.74 | 5.33 | 233 | 80 | 63 | |
| SiT-XL/2+REPAType=Diff., #Epoch=1600, #Para.=675M, #Tok.=256, Generation Mode=w/ CFG2026.05 | 1.8 | 4.5 | 284 | 81 | 61 | |
| RAR-XL(SigLIP)Type=AR., #Para.=955M, #Tok.=256, Generation Mode=w/o CFG2026.05 | 1.87 | 5.77 | 226.7 | 79 | 63 | |
| VFMTok-2B(SigLIP)Type=AR., #Epoch=200, #Para.=2.2B, #Tok.=256, Generation Mode=w/o CFG2026.05 | 1.92 | 5.78 | 260.5 | 82 | 61 | |
| VFMTok-XXL(DINOv2)Type=AR., #Epoch=200, #Para.=1.4B, #Tok.=256, Generation Mode=w/o CFG2026.05 | 1.95 | 5.65 | 259.3 | 82 | 62 | |
| VFMTok-XXL(SigLIP2)Type=AR., #Epoch=200, #Para.=1.4B, #Tok.=256, Generation Mode=w/o CFG2026.05 | 1.98 | 5.53 | 265.3 | 82 | 62 | |
| VFMTok-2B(SigLIP2)Type=AR., #Epoch=200, #Para.=2.2B, #Tok.=256, Generation Mode=w/o CFG2026.05 | 1.98 | 5.41 | 269.7 | 82 | 62 | |
| RAR-L(DINOv2)Type=AR., #Para.=461M, #Tok.=256, Generation Mode=w/o CFG2026.05 | 2.02 | 5.51 | 210.4 | 79 | 63 | |
| FasterDITType=Diff., #Epoch=400, Generation Mode=w/ CFG2026.05 | 2.03 | 4.63 | 264 | 81 | 60 | |
| VFMTok-3B(DINOv2)Type=AR., #Epoch=200, #Para.=3.1B, #Tok.=256, Generation Mode=w/o CFG2026.05 | 2.04 | 5.43 | 267.6 | 82 | 61 | |
| VFMTok-2B(SigLIP)Type=AR., #Epoch=200, #Para.=2.2B, #Tok.=256, Generation Mode=w/ CFG2026.05 | 2.05 | 5.77 | 271.4 | 82 | 61 | |
| RAR-L(SigLIP2)Type=AR., #Epoch=400, #Para.=461M, #Tok.=256, Generation Mode=w/o CFG2026.05 | 2.05 | 5.5 | 217.7 | 79 | 62 | |
| SiT-XL/2Type=Diff., #Epoch=1600, #Para.=675M, #Tok.=256, Generation Mode=w/ CFG2026.05 | 2.06 | 4.5 | 270.3 | 82 | 59 | |
| VFMTok-3B(DINOv2)Type=AR., #Epoch=200, #Para.=3.1B, #Tok.=256, Generation Mode=w/ CFG2026.05 | 2.07 | 6.23 | 280.4 | 81 | 62 | |
| VFMTok-2B(DINOv2)Type=AR., #Epoch=200, #Para.=2.2B, #Tok.=256, Generation Mode=w/o CFG2026.05 | 2.08 | 5.49 | 263.9 | 82 | 61 | |
| VFMTok-2B(DINOv2)Type=AR., #Epoch=200, #Para.=2.2B, #Tok.=256, Generation Mode=w/ CFG2026.05 | 2.09 | 6.25 | 274.5 | 81 | 61 | |
| VFMTok-XXL(SigLIP)Type=AR., #Epoch=200, #Para.=1.4B, #Tok.=256, Generation Mode=w/ CFG2026.05 | 2.09 | 5.75 | 272.6 | 82 | 60 | |
| VFMTok-XXL(SigLIP)Type=AR., #Epoch=200, #Para.=1.4B, #Tok.=256, Generation Mode=w/o CFG2026.05 | 2.09 | 5.75 | 272.6 | 82 | 60 | |
| VFMTok-XL(DINOv2)Type=AR., #Para.=775M, #Tok.=256, Generation Mode=w/o CFG2026.05 | 2.1 | 5.54 | 258.1 | 82 | 60 | |
| VFMTok-L(DINOv2)Type=AR., #Para.=343M, #Tok.=256, Generation Mode=w/o CFG2026.05 | 2.11 | 5.46 | 230.1 | 82 | 60 | |
| VFMTok-L(SigLIP2)Type=AR., #Epoch=300, #Para.=343M, #Tok.=256, Generation Mode=w/o CFG2026.05 | 2.11 | 5.39 | 225.6 | 81 | 60 | |
| VFMTok-XXL(SigLIP2)Type=AR., #Epoch=200, #Para.=1.4B, #Tok.=256, Generation Mode=w/ CFG2026.05 | 2.16 | 5.45 | 272 | 83 | 60 | |
| VFMTok-2B(SigLIP2)Type=AR., #Epoch=200, #Para.=2.2B, #Tok.=256, Generation Mode=w/ CFG2026.05 | 2.17 | 5.43 | 281.4 | 83 | 60 | |
| LlamaGen-3BType=AR., #Para.=3.1B, #Tok.=576, Generation Mode=w/ CFG2026.05 | 2.19 | 5.97 | 263.3 | 82 | 58 | |
| VFMTok-XXL(DINOv2)Type=AR., #Epoch=200, #Para.=1.4B, #Tok.=256, Generation Mode=w/ CFG2026.05 | 2.19 | 5.53 | 278 | 83 | 60 | |
| RAR-L(SigLIP)Type=AR., #Epoch=400, #Para.=461M, #Tok.=256, Generation Mode=w/o CFG2026.05 | 2.26 | 5.78 | 204.3 | 79 | 62 | |
| DiTType=Diff., #Epoch=1600, #Para.=675M, #Tok.=256, Generation Mode=w/ CFG2026.05 | 2.27 | 4.6 | 278.2 | 83 | 57 | |
| MaskDiTType=Diff., #Epoch=1600, #Para.=675M, #Tok.=256, Generation Mode=w/ CFG2026.05 | 2.28 | 5.67 | 276.6 | 80 | 61 | |
| LlamaGen-XXLType=AR., #Para.=1.4B, #Tok.=576, Generation Mode=w/ CFG2026.05 | 2.34 | 6 | 253.9 | 81 | 60 | |
| VFMTok-XL(DINOv2)Type=AR., #Para.=775M, #Tok.=256, Generation Mode=w/ CFG2026.05 | 2.41 | 5.53 | 276.8 | 83 | 59 | |
| VFMTok-L(SigLIP)Type=AR., #Epoch=300, #Para.=343M, #Tok.=256, Generation Mode=w/ CFG2026.05 | 2.61 | 5.54 | 272.1 | 84 | 56 | |
| VFMTok-L(SigLIP)Type=AR., #Epoch=300, #Para.=343M, #Tok.=256, Generation Mode=w/o CFG2026.05 | 2.61 | 5.54 | 272.1 | 84 | 56 | |
| LlamaGen-XLType=AR., #Para.=775M, #Tok.=576, Generation Mode=w/ CFG2026.05 | 2.63 | 5.59 | 244.1 | 81 | 58 | |
| VFMTok-L(SigLIP2)Type=AR., #Epoch=300, #Para.=343M, #Tok.=256, Generation Mode=w/ CFG2026.05 | 2.69 | 5.31 | 273.4 | 84 | 56 | |
| VFMTok-L(DINOv2)Type=AR., #Para.=343M, #Tok.=256, Generation Mode=w/ CFG2026.05 | 2.75 | 5.58 | 278.8 | 84 | 57 | |
| LlamaGen-LType=AR., #Para.=343M, #Tok.=576, Generation Mode=w/ CFG2026.05 | 3.07 | 6.09 | 256.1 | 83 | 52 | |
| VFMTok-B(DINOv2)Type=AR., #Epoch=300, #Para.=111M, #Tok.=256, Generation Mode=w/o CFG2026.05 | 3.09 | 5.67 | 173.6 | 80 | 58 | |
| VARType=AR., #Epoch=350, #Para.=310M, #Tok.=680, Generation Mode=w/ CFG2026.05 | 3.3 | — | 274.4 | 84 | 51 | |
| VFMTok-B(DINOv2)Type=AR., #Epoch=300, #Para.=111M, #Tok.=256, Generation Mode=w/ CFG2026.05 | 3.43 | 5.88 | 252.2 | 85 | 53 | |
| RAR-XXLType=AR., #Para.=1.5B, #Tok.=256, Generation Mode=w/o CFG2026.05 | 3.85 | 5.18 | 176.2 | 78 | 61 | |
| MaskGiT-reType=Mask., Generation Mode=w/ CFG, rejection sampling=true2026.05 | 4.02 | — | 355.6 | — | — | |
| TiTok-LType=AR., #Para.=343M, #Tok.=256, re-implemented=true, Generation Mode=w/ CFG2026.05 | 4.03 | 6.93 | 219.5 | 84 | 52 | |
| RAR-XLType=AR., #Para.=955M, #Tok.=256, Generation Mode=w/o CFG2026.05 | 4.62 | 5.27 | 158.3 | 77 | 62 | |
| MaskDiTType=Diff., #Epoch=1600, #Para.=675M, #Tok.=256, Generation Mode=w/o CFG2026.05 | 5.69 | 10.34 | 177.9 | 74 | 60 | |
| SiT-XL/2+REPAType=Diff., #Epoch=1600, #Para.=675M, #Tok.=256, Generation Mode=w/o CFG2026.05 | 5.9 | — | — | — | — | |
| LlamaGen-BType=AR., #Epoch=300, #Para.=111M, #Tok.=576, Generation Mode=w/ CFG2026.05 | 6.09 | 7.24 | 182.5 | 85 | 42 | |
| MaskGiTType=Mask., #Epoch=555, #Para.=227M, #Tok.=256, Generation Mode=w/o CFG2026.05 | 6.18 | — | 182.1 | 80 | 51 | |
| RAR-LType=AR., #Epoch=400, #Para.=461M, #Tok.=256, Generation Mode=w/o CFG2026.05 | 6.72 | 5.56 | 129.2 | 74 | 61 | |
| TiTok-BType=AR., #Epoch=300, #Para.=111M, #Tok.=256, re-implemented=true, Generation Mode=w/ CFG2026.05 | 6.76 | 7.82 | 175.3 | 85 | 43 | |
| FasterDITType=Diff., #Epoch=400, Generation Mode=w/o CFG2026.05 | 7.91 | 5.45 | 131.3 | 67 | 69 | |
| SiT-XL/2Type=Diff., #Epoch=1600, #Para.=675M, #Tok.=256, Generation Mode=w/o CFG2026.05 | 8.61 | 6.32 | 131.7 | 68 | 67 | |
| LlamaGen-3BType=AR., #Para.=3.1B, #Tok.=576, Generation Mode=w/o CFG2026.05 | 9.38 | 8.24 | 112.9 | 69 | 67 | |
| DiTType=Diff., #Epoch=1600, #Para.=675M, #Tok.=256, Generation Mode=w/o CFG2026.05 | 9.62 | 6.85 | 121.5 | 67 | 67 | |
| TiTok-LType=AR., #Para.=343M, #Tok.=256, re-implemented=true, Generation Mode=w/o CFG2026.05 | 11.4 | 7.14 | 88.8 | 68 | 64 | |
| LlamaGen-XXLType=AR., #Para.=1.4B, #Tok.=576, Generation Mode=w/o CFG2026.05 | 14.6 | 8.69 | 86.3 | 63 | 68 | |
| LlamaGen-XLType=AR., #Para.=775M, #Tok.=576, Generation Mode=w/o CFG2026.05 | 15.5 | 7.04 | 79.2 | 62 | 69 | |
| LlamaGen-LType=AR., #Para.=343M, #Tok.=576, Generation Mode=w/o CFG2026.05 | 19.1 | 8.67 | 64.3 | 61 | 67 | |
| TiTok-BType=AR., #Epoch=300, #Para.=111M, #Tok.=256, re-implemented=true, Generation Mode=w/o CFG2026.05 | 19.6 | 7.54 | 57.9 | 64 | 60 | |
| LlamaGen-BType=AR., #Epoch=300, #Para.=111M, #Tok.=576, Generation Mode=w/o CFG2026.05 | 32.2 | 11.84 | 39.9 | 57 | 61 |