Conditional Image Generation on ImageNet-1K 256x256 (val)
1.28gFIDAliTok-XL
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| AliTok-XLTraining epochs=400, #Parameters=662M, cfg=true2025.06 | 1.28 | 306.3 | 0.79 | 0.65 | — | |
| LightningDiTTraining epochs=800, #Parameters=675M, cfg=true2025.06 | 1.35 | 295.3 | 0.79 | 0.65 | — | |
| AliTok-LTraining epochs=800, #Parameters=318M, cfg=true2025.06 | 1.38 | 326.2 | 0.78 | 0.65 | — | |
| REPATraining epochs=800, #Parameters=675M, cfg=true2025.06 | 1.42 | 305.7 | 0.8 | 0.65 | — | |
| AliTok-BTraining epochs=800, #Parameters=177M, cfg=true2025.06 | 1.44 | 319.5 | 0.77 | 0.65 | — | |
| RAR-XXLTraining epochs=400, #Parameters=1.5B, cfg=true2025.06 | 1.48 | 326 | 0.8 | 0.63 | — | |
| RAR-XLTraining epochs=400, #Parameters=955M, cfg=true2025.06 | 1.5 | 306.9 | 0.8 | 0.62 | — | |
| MAR-HTraining epochs=800, #Parameters=943M, cfg=true2025.06 | 1.55 | 303.7 | 0.81 | 0.62 | — | |
| MAR-H, Diff Loss#params=943M, CFG=True, Token Representation=continuous-valued tokens2024.06 | 1.55 | 303.7 | 81 | 62 | — | |
| MDTv2-XL/2#params=676M, CFG=True, Token Representation=continuous-valued tokens2024.06 | 1.58 | 314.7 | 79 | 65 | — | |
| RAR-LTraining epochs=400, #Parameters=461M, cfg=true2025.06 | 1.7 | 299.5 | 0.81 | 0.6 | — | |
| DiffiTCFG=True, Token Representation=continuous-valued tokens2024.06 | 1.73 | 276.5 | 80 | 62 | — | |
| MAGVIT-v2Training epochs=1080, #Parameters=307M, cfg=true2025.06 | 1.78 | 319.4 | — | — | — | |
| MAGVIT-v2#params=307M, CFG=True, Token Representation=vector-quantized tokens2024.06 | 1.78 | 319.4 | — | — | — | |
| MAR-L, Diff Loss#params=479M, CFG=True, Token Representation=continuous-valued tokens2024.06 | 1.78 | 296 | 81 | 60 | — | |
| AliTok-XLTraining epochs=400, #Parameters=662M, cfg=false2025.06 | 1.88 | 228.4 | — | — | — | |
| VAR-d30Training epochs=350, #Parameters=2.0B, cfg=true2025.06 | 1.92 | 323.1 | 0.82 | 0.59 | — | |
| RAR-BTraining epochs=400, #Parameters=261M, cfg=true2025.06 | 1.95 | 290.5 | 0.82 | 0.58 | — | |
| TiTok-S-128Training epochs=800, #Parameters=287M, cfg=true2025.06 | 1.97 | 281.8 | — | — | — | |
| MAR-LTraining epochs=800, #Parameters=479M, cfg=true2025.06 | 1.98 | 290.3 | 0.81 | 0.6 | — | |
| AliTok-LTraining epochs=800, #Parameters=318M, cfg=false2025.06 | 1.98 | 200.8 | — | — | — | |
| SiT-XLTraining epochs=800, #Parameters=675M, cfg=true2025.06 | 2.06 | 270.3 | 0.82 | 0.59 | — | |
| VAR-d24Training epochs=350, #Parameters=1.0B, cfg=true2025.06 | 2.09 | 312.9 | 0.82 | 0.59 | — | |
| VDM++#params=2B, CFG=True, Token Representation=pixel-based2024.06 | 2.12 | 267.7 | — | — | — | |
| LightningDiTTraining epochs=800, #Parameters=675M, cfg=false2025.06 | 2.17 | 205.6 | — | — | — | |
| LlamaGen-3BTraining epochs=300, #Parameters=3B, cfg=true2025.06 | 2.18 | 263.3 | 0.81 | 0.59 | — | |
| AsCANFLOPS=52G, Throughput (A100, B=64, samples/sec)=556, Resolution=256x256, sampled cfg=true2024.11 | 2.23 | — | — | — | — | |
| DiT-XL/2#params=675M, CFG=True, Token Representation=continuous-valued tokens2024.06 | 2.27 | 278.2 | 83 | 57 | — | |
| DiT-XL/2-GFLOPS=118G, Throughput (A100, B=64, samples/sec)=293, Resolution=256x2562024.11 | 2.27 | — | — | — | — | |
| U-ViT-H/2-G#params=501M, CFG=True, Token Representation=continuous-valued tokens2024.06 | 2.29 | 263.9 | 82 | 57 | — | |
| U-ViT-H/2FLOPS=133G, Throughput (A100, B=64, samples/sec)=271, Resolution=256x2562024.11 | 2.29 | — | — | — | — | |
| MAR-BTraining epochs=800, #Parameters=208M, cfg=true2025.06 | 2.31 | 281.7 | 0.82 | 0.57 | — | |
| MAR-B, Diff Loss#params=208M, CFG=True, Token Representation=continuous-valued tokens2024.06 | 2.31 | 281.7 | 82 | 57 | — | |
| LlamaGen-XXLTraining epochs=300, #Parameters=1.4B, cfg=true2025.06 | 2.34 | 253.9 | 0.8 | 0.59 | — | |
| MAR-HTraining epochs=800, #Parameters=943M, cfg=false2025.06 | 2.35 | 227.8 | — | — | — | |
| MAR-H, Diff Loss#params=943M, CFG=False, Token Representation=continuous-valued tokens2024.06 | 2.35 | 227.8 | 79 | 62 | — | |
| AliTok-BTraining epochs=800, #Parameters=177M, cfg=false2025.06 | 2.4 | 177.1 | — | — | — | |
| VDM++#params=2B, CFG=False, Token Representation=pixel-based2024.06 | 2.4 | 225.3 | — | — | — | |
| AsCANFLOPS=52G, Throughput (A100, B=64, samples/sec)=556, Resolution=256x2562024.11 | 2.41 | — | — | — | — | |
| MAR-LTraining epochs=800, #Parameters=479M, cfg=false2025.06 | 2.6 | 221.4 | — | — | — | |
| MAR-L, Diff Loss#params=479M, CFG=False, Token Representation=continuous-valued tokens2024.06 | 2.6 | 221.4 | 79 | 60 | — | |
| LlamaGen-XLTraining epochs=300, #Parameters=775M, cfg=true2025.06 | 2.62 | 244.1 | 0.8 | 0.57 | — | |
| RAR-XXLTraining epochs=400, #Parameters=1.5B, cfg=false2025.06 | 3.26 | 193.3 | — | — | — | |
| GIVT#params=304M, CFG=True, Token Representation=continuous-valued tokens2024.06 | 3.35 | — | 84 | 53 | — | |
| U-ViT-L/2FLOPS=77G, Throughput (A100, B=64, samples/sec)=498, Resolution=256x2562024.11 | 3.4 | — | — | — | — | |
| MAR-BTraining epochs=800, #Parameters=208M, cfg=false2025.06 | 3.48 | 192.4 | — | — | — | |
| MAR-B, Diff Loss#params=208M, CFG=False, Token Representation=continuous-valued tokens2024.06 | 3.48 | 192.4 | 78 | 58 | — | |
| LDM-4#params=400M, CFG=True, Token Representation=continuous-valued tokens2024.06 | 3.6 | 247.7 | 87 | 48 | — | |
| LDMFLOPS=104G, Throughput (A100, B=64, samples/sec)=362, Resolution=256x2562024.11 | 3.6 | — | — | — | — | |
| MAGVIT-v2Training epochs=1080, #Parameters=307M, cfg=false2025.06 | 3.65 | 200.5 | — | — | — | |
| MAGVIT-v2#params=307M, CFG=False, Token Representation=vector-quantized tokens2024.06 | 3.65 | 200.5 | — | — | — | |
| RAR-XLTraining epochs=400, #Parameters=955M, cfg=false2025.06 | 3.72 | 179.9 | — | — | — | |
| VIT-VQGANParams=1.7B, Steps=10242022.12 | 4.17 | 175.1 | — | — | — | |
| TiTok-S-128Training epochs=800, #Parameters=287M, cfg=false2025.06 | 4.44 | 168.2 | — | — | — | |
| SeQ-GAN + NAR-LParams=364M, Steps=122022.12 | 4.55 | 200.4 | — | — | — | |
| ADM#params=554M, CFG=True, Token Representation=pixel-based2024.06 | 4.59 | 186.7 | 82 | 52 | — | |
| SeQ-GAN + NARParams=229M, Steps=122022.12 | 4.99 | 189.1 | — | — | — | |
| MDTv2-XL/2#params=676M, CFG=False, Token Representation=continuous-valued tokens2024.06 | 5.06 | 155.6 | 72 | 66 | — | |
| RAR-LTraining epochs=400, #Parameters=461M, cfg=false2025.06 | 5.39 | 149.1 | — | — | — | |
| GIVT#params=304M, CFG=False, Token Representation=continuous-valued tokens2024.06 | 5.67 | — | 75 | 59 | — | |
| REPATraining epochs=800, #Parameters=675M, cfg=false2025.06 | 5.9 | 157.8 | — | — | — | |
| MaskGITTraining epochs=300, #Parameters=227M, cfg=false2025.06 | 6.18 | 182.1 | — | — | — | |
| MaskGITParams=227M, Steps=82022.12 | 6.18 | 182.1 | — | — | — | |
| MaskGIT#params=227M, CFG=False, Token Representation=vector-quantized tokens2024.06 | 6.18 | 182.1 | 80 | 51 | — | |
| SeQ-GAN + AR-LParams=364M, Steps=2562022.12 | 6.25 | 140.9 | — | — | — | |
| MAGE#params=230M, CFG=False, Token Representation=vector-quantized tokens2024.06 | 6.93 | 195.8 | — | — | — | |
| BigGAN-DeepParams=160M, Steps=12022.12 | 6.95 | 198.2 | — | — | — | |
| RAR-BTraining epochs=400, #Parameters=261M, cfg=false2025.06 | 7.12 | 124.8 | — | — | — | |
| MoVQ + ARParams=389M, Steps=10242022.12 | 7.13 | 138.3 | — | — | — | |
| MoVQ + NARParams=389M, Steps=122022.12 | 7.22 | 130.1 | — | — | — | |
| SeQ-GAN + ARParams=229M, Steps=2562022.12 | 7.55 | 121.3 | — | — | — | |
| SiT-XLTraining epochs=800, #Parameters=675M, cfg=false2025.06 | 8.61 | 131.7 | — | — | — | |
| RQ-VAEParams=1.4B, Steps=10242022.12 | 8.71 | 119 | — | — | — | |
| LlamaGen-3BTraining epochs=300, #Parameters=3B, cfg=false2025.06 | 9.38 | 112.9 | — | — | — | |
| DiT-XL/2#params=675M, CFG=False, Token Representation=continuous-valued tokens2024.06 | 9.62 | 121.5 | 67 | 67 | — | |
| LDM-4#params=400M, CFG=False, Token Representation=continuous-valued tokens2024.06 | 10.56 | 103.5 | 71 | 62 | — | |
| ADMFLOPS=110G, Resolution=256x2562024.11 | 10.6 | — | — | — | — | |
| ADMParams=554M, Steps=2502022.12 | 10.94 | 101 | — | — | — | |
| ADM#params=554M, CFG=False, Token Representation=pixel-based2024.06 | 10.94 | 101 | 69 | 63 | — | |
| VIT-VQGANParams=714M, Steps=2562022.12 | 11.2 | 97.2 | — | — | — | |
| VQ-DiffusionParams=370M, Steps=1002022.12 | 11.89 | — | — | — | — | |
| Improved DDPMParams=280M, Steps=2502022.12 | 12.26 | — | — | — | — | |
| VQGANParams=1.4B, Steps=2562022.12 | 15.78 | 78.3 | — | — | — | |
| Autoreg. w/ VQGAN#params=1.4B, CFG=False, Token Representation=vector-quantized tokens2024.06 | 15.78 | 78.3 | — | — | — | |
| VQ-VAE-2Params=13.5B, Steps=51202022.12 | 31.11 | 45 | — | — | — | |
| DCTParams=738M, Steps=>10242022.12 | 36.51 | — | — | — | — | |
| Baseline (AR)type=autoregressive2026.02 | — | 282.6 | 80 | 59 | 2.02 | |
| Baseline (CDM)type=conditional diffusion modeling2026.02 | — | 259.6 | 81 | 58 | 3.26 | |
| Baseline (L=1)GFlops=26.82024.12 | — | — | — | — | 31.1 | |
| De-MAR2026.02 | — | 305.8 | 83 | 62 | 1.47 | |
| DiffiT2026.02 | — | 276.5 | 80 | 62 | 1.73 | |
| DiT-L/2GFlops=80.02024.12 | — | — | — | — | 23.3 | |
| DiT-L/2 + REPAGFlops=80.02024.12 | — | — | — | — | 15.6 | |
| DiT-XL/2GFlops=118.62024.12 | — | — | — | — | 19.5 | |
| DiT-XL/2Model Scale=XL/22026.02 | — | 278.2 | 83 | 57 | 2.27 | |
| DiT-XL/2 + REPAGFlops=118.62024.12 | — | — | — | — | 12.3 | |
| GIVT2026.02 | — | — | 84 | 53 | 3.35 | |
| LDM-4downsampling factor=42026.02 | — | 247.7 | 87 | 48 | 3.6 | |
| MAR2026.02 | — | 303.7 | 81 | 62 | 1.55 | |
| MDTv2-XL/2Model Scale=XL/22026.02 | — | 314.7 | 79 | 65 | 1.58 |