Image Generation on ImageNet 256x256 (gFID w/o/w CFG)
1.35gFID (w/ CFG)VA-VAE + LightningDiT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| VA-VAE + LightningDiT#Tokens=256, Generator=LightningDiT [10]2026.06 | 1.35 | 2.17 | |
| VAEGenerative Model=MAR, Param=943M, Latent Tokens=2562026.04 | 1.55 | 2.35 | |
| MAR-H + MAR-H#Tokens=256, Generator=MAR-H [7]2026.06 | 1.55 | 2.35 | |
| SelfBootTok + SiT#Tokens=64, Generator=SiT [12]2026.06 | 1.56 | 2.22 | |
| MAETokGenerative Model=LightningDiT, Param=675M, Latent Tokens=2562026.04 | 1.73 | 2.21 | |
| MAETok + LightningDiT#Tokens=128, Generator=LightningDiT [10]2026.06 | 1.73 | — | |
| SoftVQ + SiT#Tokens=64, Generator=SiT [12]2026.06 | 1.78 | 5.98 | |
| HieraTok + DiT#Tokens=256, Generator=DiT [1]2026.06 | 1.82 | 3.53 | |
| FlexTok d18-d28 + LlamaGen#Tokens=1 to 256, Generator=LlamaGen [63]2026.06 | 1.86 | — | |
| VAR-Tok + VAR-d30#Tokens=680, Generator=VAR-d30 [5]2026.06 | 1.92 | — | |
| TiTok-S + MaskGIT-UViT-L#Tokens=128, Generator=MaskGIT-UViT-L [25, 66]2026.06 | 1.97 | — | |
| GigaTok XL-XXL + LlamaGen XXL#Tokens=256, Generator=LlamaGen XXL [63]2026.06 | 1.98 | — | |
| SD-VAE + SiT-XL/2#Tokens=256, Generator=SiT-XL/2 [12]2026.06 | 2.06 | 8.61 | |
| VQGANGenerative Model=LlamaGen, Param=3B, Latent Tokens=2562026.04 | 2.18 | 9.38 | |
| VQGAN + LlamaGen#Tokens=256, Generator=LlamaGen [63]2026.06 | 2.18 | 9.38 | |
| LlamaGen-3B + LlamaGen#Tokens=576, Generator=LlamaGen [63]2026.06 | 2.18 | — | |
| SD-VAEGenerative Model=DiT, Param=600M, Latent Tokens=2562026.04 | 2.27 | 9.62 | |
| SD-VAE + DiT-XL/2#Tokens=256, Generator=DiT-XL/2 [1]2026.06 | 2.27 | 9.62 | |
| ViTok S-BGenerative Model=DiT, Param=675M, Latent Tokens=2562026.04 | 2.45 | — | |
| ViTok S-B/16 + DiT-XL#Tokens=256, Generator=DiT-XL [1]2026.06 | 2.45 | — | |
| TiTok-B + MaskGIT-ViT#Tokens=64, Generator=MaskGIT-ViT [25]2026.06 | 2.48 | — | |
| TC-AEGenerative Model=DiT, Param=675M, Latent Tokens=64, CFG scale=1.42026.04 | 2.57 | 7.16 | |
| VAR-Tok.Generative Model=VAR-d20, Param=600M, Latent Tokens=6802026.04 | 2.95 | — | |
| LlamaGen-Tok.Generative Model=LlamaGen, Param=1.4B, Latent Tokens=2562026.04 | 3.09 | — | |
| GigaTok-B-LGenerative Model=LlamaGen, Param=111M, Latent Tokens=2562026.04 | 3.33 | — | |
| VAEGenerative Model=LDM-4, Param=400M, Latent Tokens=40962026.04 | 3.6 | — | |
| VAE + LDM-4#Tokens=4096, Generator=LDM-4 [24]2026.06 | 3.6 | 10.56 | |
| RQ-VAE + RQ-Transformer#Tokens=256, Generator=RQ-Transformer [64]2026.06 | 3.8 | — | |
| DC-AEGenerative Model=DiT, Param=675M, Latent Tokens=642026.04 | — | 26.44 | |
| DC-AE-1.5Generative Model=DiT, Param=675M, Latent Tokens=642026.04 | — | 17.31 | |
| MAGE + MAGE#Tokens=256, Generator=MAGE [65]2026.06 | — | 6.93 | |
| MaskGITGenerative Model=MaskGIT, Param=227M, Latent Tokens=2562026.04 | — | 6.18 | |
| MaskGIT + MaskGIT#Tokens=256, Generator=MaskGIT [25]2026.06 | — | 6.18 |