Conditional Image Generation on ImageNet 512x512 (val)
1.52gFIDMacTok+SiT-XL‡
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| MacTok+SiT-XL‡# Params (G)=675M, Tok. Model=KL, Token Type=1D, # Params (T)=391M†, # Tokens=64, Classifier-Free Guidance (CFG)=true2026.03 | 1.52 | 306 | — | — | 0.89 | — | — | |
| MacTok+SiT-XL‡# Params (G)=675M, Tok. Model=KL, Token Type=1D, # Params (T)=176M, # Tokens=128, Classifier-Free Guidance (CFG)=true2026.03 | 1.52 | 316 | — | — | 0.79 | — | — | |
| MacTok+SiT-XL# Params(G)=675M, Tok. Model=KL, Token Type=1D, # Params(T)=391M, # Tokens=64, Tok. rFID=0.89, CFG setting=w/ CFG2026.03 | 1.52 | 306 | 0.8 | 0.63 | — | — | — | |
| MacTok+SiT-XL# Params(G)=675M, Tok. Model=KL, Token Type=1D, # Params(T)=176M, # Tokens=128, Tok. rFID=0.79, CFG setting=w/ CFG2026.03 | 1.52 | 316 | 0.8 | 0.63 | — | — | — | |
| MAETok‡# Params (G)=675M, Tok. Model=AE, Token Type=1D, # Params (T)=176M, # Tokens=128, Classifier-Free Guidance (CFG)=true2026.03 | 1.69 | 304.2 | — | — | 0.62 | — | — | |
| MAETok# Params(G)=675M, Tok. Model=AE, Token Type=1D, # Params(T)=176M, # Tokens=128, Tok. rFID=0.62, CFG setting=w/ CFG2026.03 | 1.69 | 304.2 | 0.82 | 0.62 | — | — | — | |
| MAR-LType=Mask., Estimated inference time=via linear scaling2026.02 | 1.73 | 279.9 | — | — | — | 214.4 | — | |
| MAR-H# Params (G)=943M, Tok. Model=KL, Token Type=2D, # Params (T)=66M, # Tokens=1024, Classifier-Free Guidance (CFG)=true2026.03 | 1.73 | 279.9 | — | — | — | — | — | |
| MAR-H# Params(G)=943M, Tok. Model=KL, Token Type=2D, # Params(T)=66M, # Tokens=1024, CFG setting=w/ CFG2026.03 | 1.73 | 279.9 | 0.77 | 0.61 | — | — | — | |
| TexTok-128# Params (G)=675M, Tok. Model=KL, Token Type=1D, # Params (T)=176M, # Tokens=128, Classifier-Free Guidance (CFG)=true2026.03 | 1.8 | 305.4 | — | — | 0.97 | — | — | |
| TexTok-128# Params(G)=675M, Tok. Model=KL, Token Type=1D, # Params(T)=176M, # Tokens=128, Tok. rFID=0.97, CFG setting=w/ CFG2026.03 | 1.8 | 305.4 | 0.81 | 0.63 | — | — | — | |
| MAGVIT-v2# Params (G)=307M, Tok. Model=LFQ, Token Type=2D, # Params (T)=116M, # Tokens=1024, Classifier-Free Guidance (CFG)=true2026.03 | 1.91 | 324.3 | — | — | — | — | — | |
| MAGVIT-v2# Params(G)=307M, Tok. Model=LFQ, Token Type=2D, # Params(T)=116M, # Tokens=1024, CFG setting=w/ CFG2026.03 | 1.91 | 324.3 | — | — | — | — | — | |
| TiTok-BTok. Model=VQ, # Tokens=128, CFG=true2024.12 | 2.13 | 261.2 | — | — | 1.52 | — | — | |
| TiTok-B-128# Params (G)=177M, Tok. Model=VQ, Token Type=1D, # Params (T)=202M, # Tokens=128, Classifier-Free Guidance (CFG)=true2026.03 | 2.13 | 261.2 | — | — | 1.52 | — | — | |
| TiTok-B-128# Params(G)=177M, Tok. Model=VQ, Token Type=1D, # Params(T)=202M, # Tokens=128, Tok. rFID=1.52, CFG setting=w/ CFG2026.03 | 2.13 | 261.2 | — | — | — | — | — | |
| SiT-XLTok. Model=SoftVQ-BL, # Tokens=64, Gflops=29.23, Throughput (imgs/sec)=5.38, CFG=true2024.12 | 2.21 | 290.5 | 0.85 | 0.59 | 0.71 | — | — | |
| SoftVQ-VAE‡# Params (G)=675M, Tok. Model=SoftVQ, Token Type=1D, # Params (T)=391M, # Tokens=64, Classifier-Free Guidance (CFG)=true2026.03 | 2.21 | 290.5 | — | — | 0.71 | — | — | |
| SoftVQ-VAE# Params(G)=675M, Tok. Model=SoftVQ, Token Type=1D, # Params(T)=391M, # Tokens=64, Tok. rFID=0.71, CFG setting=w/ CFG2026.03 | 2.21 | 290.5 | 0.85 | 0.59 | — | — | — | |
| UViT-2BGflops=104.18, Throughput (imgs/sec)=9.44, Sampler=DPM-Solver, Sampling steps=50, CFG=true2024.12 | 2.25 | — | — | — | — | — | — | |
| UViT-2B# Params (G)=2B, Tok. Model=AE, Token Type=2D, # Params (T)=323M, # Tokens=256, Classifier-Free Guidance (CFG)=true2026.03 | 2.25 | — | — | — | 0.22 | — | — | |
| UViT-2B# Params(G)=2B, Tok. Model=AE, Token Type=2D, # Params(T)=323M, # Tokens=256, Tok. rFID=0.22, CFG setting=w/ CFG2026.03 | 2.25 | — | — | — | — | — | — | |
| VAR-d36 + SSGType=VAR2026.02 | 2.39 | 320.6 | — | — | — | 1 | — | |
| StyleGAN-XLCFG=true2024.12 | 2.41 | 267.7 | — | — | — | — | — | |
| StyleGAN-XL# Params (G)=168M, Classifier-Free Guidance (CFG)=true2026.03 | 2.41 | 267.7 | — | — | — | — | — | |
| StyleGAN-XL# Params(G)=168M, CFG setting=w/ CFG2026.03 | 2.41 | 267.7 | — | — | — | — | — | |
| UViT-HGflops=128.9, Throughput (imgs/sec)=6.14, Sampler=DPM-Solver, Sampling steps=50, CFG=true2024.12 | 2.53 | — | — | — | — | — | — | |
| UViT-H# Params (G)=501M, Tok. Model=KL, Token Type=2D, # Params (T)=84M, # Tokens=4096, Classifier-Free Guidance (CFG)=true2026.03 | 2.53 | — | — | — | 0.62 | — | — | |
| UViT-H# Params(G)=501M, Tok. Model=KL, Token Type=2D, # Params(T)=84M, # Tokens=4096, Tok. rFID=0.62, CFG setting=w/ CFG2026.03 | 2.53 | — | — | — | — | — | — | |
| SiT-XL/2Gflops=373.32, Throughput (imgs/sec)=0.1, CFG=true2024.12 | 2.62 | 252.2 | 0.84 | 0.57 | 0.1 | — | — | |
| SiT-XL/2# Params (G)=675M, Token Type=2D, Classifier-Free Guidance (CFG)=true2026.03 | 2.62 | 252.2 | — | — | — | — | — | |
| SiT-XL/2# Params(G)=675M, Token Type=2D, CFG setting=w/ CFG2026.03 | 2.62 | 252.2 | 0.84 | 0.57 | — | — | — | |
| UViT-HTok. Model=AE, # Tokens=64, Gflops=32.99, Throughput (imgs/sec)=15.23, Sampler=DPM-Solver, Sampling steps=50, CFG=true2024.12 | 2.66 | — | — | — | 0.22 | — | — | |
| UViT-H# Params (G)=501M, Token Type=2D, Classifier-Free Guidance (CFG)=true2026.03 | 2.66 | — | — | — | — | — | — | |
| UViT-H# Params(G)=501M, Token Type=2D, CFG setting=w/ CFG2026.03 | 2.66 | — | — | — | — | — | — | |
| DiffiTType=Diff.2026.02 | 2.67 | 252.1 | — | — | — | — | — | |
| VAR-d36Type=VAR, Status=Reproduced2026.02 | 2.7 | 290.6 | — | — | — | 1 | — | |
| MAR-H# Params (G)=943M, Tok. Model=KL, Token Type=2D, # Params (T)=66M, # Tokens=1024, Classifier-Free Guidance (CFG)=false2026.03 | 2.74 | 205.2 | — | — | — | — | — | |
| TiTok-L-64# Params (G)=177M, Tok. Model=VQ, Token Type=1D, # Params (T)=644M, # Tokens=64, Classifier-Free Guidance (CFG)=true2026.03 | 2.74 | 221.1 | — | — | 1.77 | — | — | |
| MAR-H# Params(G)=943M, Tok. Model=KL, Token Type=2D, # Params(T)=66M, # Tokens=1024, CFG setting=w/o CFG2026.03 | 2.74 | 205.2 | 0.69 | 0.59 | — | — | — | |
| TiTok-L-64# Params(G)=177M, Tok. Model=VQ, Token Type=1D, # Params(T)=644M, # Tokens=64, Tok. rFID=1.77, CFG setting=w/ CFG2026.03 | 2.74 | 221.1 | — | — | — | — | — | |
| MAETok‡# Params (G)=675M, Tok. Model=AE, Token Type=1D, # Params (T)=176M, # Tokens=128, Classifier-Free Guidance (CFG)=false2026.03 | 2.79 | 204.3 | — | — | 0.62 | — | — | |
| MAETok# Params(G)=675M, Tok. Model=AE, Token Type=1D, # Params(T)=176M, # Tokens=128, Tok. rFID=0.62, CFG setting=w/o CFG2026.03 | 2.79 | 204.3 | 0.81 | 0.62 | — | — | — | |
| DiT-XLTok. Model=AE, # Tokens=256, Gflops=80.75, Throughput (imgs/sec)=1.02, CFG=true2024.12 | 2.84 | — | — | — | 0.22 | — | — | |
| DiT-XL# Params (G)=675M, Token Type=2D, Classifier-Free Guidance (CFG)=true2026.03 | 2.84 | — | — | — | — | — | — | |
| DiT-XL# Params(G)=675M, Token Type=2D, CFG setting=w/ CFG2026.03 | 2.84 | — | — | — | — | — | — | |
| HMAR-d24Type=VAR2026.02 | 2.99 | 304.1 | — | — | — | — | — | |
| DiT-XL/2Gflops=373.34, Throughput (imgs/sec)=0.1, CFG=true2024.12 | 3.04 | 240.8 | 0.84 | 0.54 | 0.1 | — | — | |
| DiT-XL/2Type=Diff.2026.02 | 3.04 | 240.8 | — | — | — | 81 | — | |
| DiT-XL/2# Params (G)=675M, Token Type=2D, Classifier-Free Guidance (CFG)=true2026.03 | 3.04 | 240.8 | — | — | — | — | — | |
| DiT-XL/2# Params(G)=675M, Token Type=2D, CFG setting=w/ CFG2026.03 | 3.04 | 240.8 | 0.84 | 0.54 | — | — | — | |
| DIFFUSSM-XL-GType=Diff.2026.02 | 3.41 | 255.1 | — | — | — | — | — | |
| MAR-HTok. Model=SoftVQ-BL, # Tokens=64, Gflops=86.55, Throughput (imgs/sec)=1.5, CFG=true2024.12 | 3.42 | 261.8 | 0.77 | 0.61 | 0.71 | — | — | |
| ADMCFG=true2024.12 | 3.85 | 221.7 | 0.84 | 0.53 | — | — | — | |
| ADMClassifier-Free Guidance (CFG)=true2026.03 | 3.85 | 221.7 | — | — | — | — | — | |
| ADMCFG setting=w/ CFG2026.03 | 3.85 | 221.7 | 0.84 | 0.53 | — | — | — | |
| U-ViT-H/4Tok. Model=KL, # Tokens=4096, Gflops=128.92, Throughput (imgs/sec)=0.58, Sampler=DPM-Solver, Sampling steps=50, CFG=true2024.12 | 4.05 | 263.8 | 0.84 | 0.48 | 0.62 | — | — | |
| U-ViT-H/4# Params (G)=501M, Token Type=2D, Classifier-Free Guidance (CFG)=true2026.03 | 4.05 | 263.8 | — | — | — | — | — | |
| U-ViT-H/4# Params(G)=501M, Token Type=2D, CFG setting=w/o CFG2026.03 | 4.05 | 263.8 | 0.84 | 0.48 | — | — | — | |
| SiT-XLTok. Model=SoftVQ-L, # Tokens=32, Gflops=14.73, Throughput (imgs/sec)=10.12, CFG=true2024.12 | 4.23 | 218 | 0.83 | 0.52 | 0.64 | — | — | |
| MacTok+SiT-XL‡# Params (G)=675M, Tok. Model=KL, Token Type=1D, # Params (T)=391M†, # Tokens=64, Classifier-Free Guidance (CFG)=false2026.03 | 4.63 | 163.7 | — | — | 0.89 | — | — | |
| MacTok+SiT-XL# Params(G)=675M, Tok. Model=KL, Token Type=1D, # Params(T)=391M, # Tokens=64, Tok. rFID=0.89, CFG setting=w/o CFG2026.03 | 4.63 | 163.7 | 0.8 | 0.61 | — | — | — | |
| MacTok+SiT-XL‡# Params (G)=675M, Tok. Model=KL, Token Type=1D, # Params (T)=176M, # Tokens=128, Classifier-Free Guidance (CFG)=false2026.03 | 5.12 | 156.3 | — | — | 0.79 | — | — | |
| MacTok+SiT-XL# Params(G)=675M, Tok. Model=KL, Token Type=1D, # Params(T)=176M, # Tokens=128, Tok. rFID=0.79, CFG setting=w/o CFG2026.03 | 5.12 | 156.3 | 0.79 | 0.61 | — | — | — | |
| UViT-2B# Params (G)=2B, Tok. Model=AE, Token Type=2D, # Params (T)=323M, # Tokens=256, Classifier-Free Guidance (CFG)=false2026.03 | 6.5 | — | — | — | 0.22 | — | — | |
| UViT-2B# Params(G)=2B, Tok. Model=AE, Token Type=2D, # Params(T)=323M, # Tokens=256, Tok. rFID=0.22, CFG setting=w/o CFG2026.03 | 6.5 | — | — | — | — | — | — | |
| MaskGITTok. Model=VQ, # Tokens=1024, CFG=false2024.12 | 7.32 | 156 | — | — | 1.97 | — | — | |
| MaskGITType=Mask.2026.02 | 7.32 | 156 | — | — | — | 0.5 | — | |
| MaskGIT# Params (G)=227M, Tok. Model=VQ, Token Type=2D, # Params (T)=66M, # Tokens=1024, Classifier-Free Guidance (CFG)=false2026.03 | 7.32 | 156 | — | — | 1.97 | — | — | |
| MaskGIT# Params(G)=227M, Tok. Model=VQ, Token Type=2D, # Params(T)=66M, # Tokens=1024, Tok. rFID=1.97, CFG setting=w/o CFG2026.03 | 7.32 | 156 | — | — | — | — | — | |
| SiT-XLTok. Model=SoftVQ-BL, # Tokens=64, Gflops=29.23, Throughput (imgs/sec)=5.38, CFG=false2024.12 | 7.96 | 133.9 | 0.73 | 0.63 | 0.71 | — | — | |
| SoftVQ-VAE‡# Params (G)=675M, Tok. Model=SoftVQ, Token Type=1D, # Params (T)=391M, # Tokens=64, Classifier-Free Guidance (CFG)=false2026.03 | 7.96 | 133.9 | — | — | 0.71 | — | — | |
| SoftVQ-VAE# Params(G)=675M, Tok. Model=SoftVQ, Token Type=1D, # Params(T)=391M, # Tokens=64, Tok. rFID=0.71, CFG setting=w/o CFG2026.03 | 7.96 | 133.9 | 0.73 | 0.63 | — | — | — | |
| MAR-HTok. Model=SoftVQ-BL, # Tokens=64, Gflops=86.55, Throughput (imgs/sec)=1.5, CFG=false2024.12 | 8.21 | 152.9 | 0.69 | 0.59 | 0.71 | — | — | |
| BigGANCFG=true2024.12 | 8.43 | 177.9 | — | — | — | — | — | |
| BigGANType=GAN2026.02 | 8.43 | 177.9 | — | — | — | — | — | |
| BigGANClassifier-Free Guidance (CFG)=true2026.03 | 8.43 | 177.9 | — | — | — | — | — | |
| BigGANCFG setting=w/ CFG2026.03 | 8.43 | 177.9 | — | — | — | — | — | |
| DiT-XL# Params (G)=675M, Token Type=2D, Classifier-Free Guidance (CFG)=false2026.03 | 9.56 | — | — | — | — | — | — | |
| DiT-XL# Params(G)=675M, Token Type=2D, CFG setting=w/o CFG2026.03 | 9.56 | — | — | — | — | — | — | |
| DiT-XL/2Gflops=373.34, Throughput (imgs/sec)=0.1, CFG=false2024.12 | 9.62 | 121.5 | — | — | 0.1 | — | — | |
| DiT-XL/2# Params (G)=675M, Token Type=2D, Classifier-Free Guidance (CFG)=false2026.03 | 9.62 | 121.5 | — | — | — | — | — | |
| DiT-XL/2# Params(G)=675M, Token Type=2D, CFG setting=w/o CFG2026.03 | 9.62 | 121.5 | — | — | — | — | — | |
| UViT-H# Params (G)=501M, Tok. Model=KL, Token Type=2D, # Params (T)=84M, # Tokens=4096, Classifier-Free Guidance (CFG)=false2026.03 | 9.83 | — | — | — | 0.62 | — | — | |
| UViT-H# Params(G)=501M, Tok. Model=KL, Token Type=2D, # Params(T)=84M, # Tokens=4096, Tok. rFID=0.62, CFG setting=w/o CFG2026.03 | 9.83 | — | — | — | — | — | — | |
| SiT-XLTok. Model=SoftVQ-L, # Tokens=32, Gflops=14.73, Throughput (imgs/sec)=10.12, CFG=false2024.12 | 10.17 | 119.2 | 0.65 | 0.59 | 0.64 | — | — | |
| UViT-H# Params (G)=501M, Token Type=2D, Classifier-Free Guidance (CFG)=false2026.03 | 12.26 | — | — | — | — | — | — | |
| UViT-H# Params(G)=501M, Token Type=2D, CFG setting=w/o CFG2026.03 | 12.26 | — | — | — | — | — | — | |
| ADMCFG=false2024.12 | 23.24 | 58.06 | — | — | — | — | — | |
| ADMClassifier-Free Guidance (CFG)=false2026.03 | 23.24 | 58.1 | — | — | — | — | — | |
| ADMCFG setting=w/o CFG2026.03 | 23.24 | 58.1 | — | — | — | — | — | |
| VQGANType=AR2026.02 | 26.52 | 66.8 | — | — | — | 25 | — | |
| EDM2-LNFE=632026.03 | — | 250.07 | 0.753 | 0.61 | — | — | 67.7 | |
| EDM2-L + CC-DPONFE=632026.03 | — | 298.89 | 0.812 | 0.544 | — | — | 51.92 | |
| EDM2-L + CCANFE=632026.03 | — | 281.45 | 0.79 | 0.574 | — | — | 51.45 | |
| EDM2-L + CFGNFE=1262026.03 | — | 360.3 | 0.844 | 0.512 | — | — | 39.86 | |
| EDM2-L + DDONFE=632026.03 | — | 268.72 | 0.737 | 0.652 | — | — | 49.47 | |
| EDM2-L + MCLRNFE=632026.03 | — | 332.02 | 0.849 | 0.492 | — | — | 42.5 |