List your company

Image Reconstruction on ImageNet (val)

0.011rFIDFlux VAE
-0.035360.277570.59050.90343Dec 19, 2024Mar 18, 2025Jun 16, 2025Sep 13, 2025Dec 12, 2025Mar 11, 2026Jun 9, 2026

Evaluation results

MethodLinks
Flux VAEScale Factor=4.02026.030.0110.99141.710.0034——0.68———
Flux VAEScale Factor=2.02026.030.0120.9941.510.0061——0.71———
Flux VAEScale Factor=3.02026.030.0120.9941.60.0061——0.7———
Flux VAEScale Factor=1.52026.030.0330.97737.180.0081——1.96———
FLUX.1[dev]-VAETokenizer Category=Specialized, Resolution=512, Note=re-evaluated using official checkpoints2026.030.060.9333.65———————
FLUX.1[dev]-VAERes.=5122026.040.060.9333.65———————
Flux VAEScale Factor=0.52026.030.0690.92730.820.0126——8.31———
Flux VAEScale Factor=1.02026.030.1120.91428.280.0184——15.05———
Distilled VAEScale Factor=4.02026.030.1150.98336.20.0041——2.4———
Tuna-RRes.=5122026.040.120.9332.22———————
Tuna-2Res.=5122026.040.150.9332.8———————
FLUX-VAETokenizer Category=Generation-oriented Tokenizer2026.010.1760.91732.860.044——————
PS-VAERes.=2562026.040.20.8228.79———————
SD3-VAETokenizer Category=Generation-oriented Tokenizer2026.010.2010.88631.290.059——————
SD3-VAESemantics Distillation=false2025.070.21—————————
OpenVision 3Tokenizer Category=Unified Tokenizer2026.010.2160.88530.330.061——————
DC-AETokenizer Category=Specialized, Resolution=5122026.030.220.7126.15———————
DC-AERes.=5122026.040.220.7126.15———————
VA-VAETokenizer Category=Specialized, Resolution=2562026.030.260.828.59———————
VA-VAERes.=2562026.040.260.828.59———————
AlignTokf=14, #Tok.=256, Dim.=322026.050.26—25.8———————
Distilled VAEScale Factor=3.02026.030.2680.98535.780.0046——2.64———
VA-VAESemantics Distillation=true2025.070.28—————————
VA-VAEf=16, #Tok.=256, Dim.=322026.050.280.7928———————
VMFAE(DINOv2)f=16, #Tok.=256, Dim.=322026.050.290.8225.5————225.6——
VFMAE(SigLIP2)f=16, #Tok.=256, Dim.=322026.050.290.7924.6————225.4——
UniTokTokenizer Category=Unified Tokenizer2026.010.3620.74225.340.132——————
UniTokTokenizer Category=Unified, Resolution=2562026.030.38—————————
UniComTokenizer Category=Unified, Resolution=1024, d (feature dimension)=11522026.030.380.6122.6———————
UniTokRes.=2562026.040.38—————————
UniComTokenizer Category=Unified, Resolution=1024, d (feature dimension)=642026.030.420.6122.28———————
GQbpp (K x N)=0.28 (2^18 x 1024), ImageNet Trained Epochs=40, Params(M)=872025.120.4240.80427.860.054——————
Emu3.5-IBQ#Tokens=1,024, Codebook=131,072, BPP=0.0664, Resolution=512 x 5122026.050.46—23.82———————
MAETokSemantics Distillation=true2025.070.48—————————
GQbpp (K x N)=0.22 (2^14 x 1024), ImageNet Trained Epochs=40, Params(M)=872025.120.4910.76225.310.064——————
GigaTokTokenizer Category=Specialized, Resolution=2562026.030.510.6921.32———————
GigaTokRes.=2562026.040.510.6921.32———————
VFMVAEf=14, #Tok.=256, Dim.=322026.050.520.5923————214.1——
MAR-VAESemantics Distillation=false2025.070.53—————————
MingTokTokenizer Category=Unified, Resolution=512, Note=re-evaluated using official checkpoints2026.030.530.6123.49———————
MingTokRes.=5122026.040.530.6123.49———————
FlowMo-Hibpp (K x N)=0.22 (2^14 x 1024), ImageNet Trained Epochs=300, Params(M)=9452025.120.560.78524.930.073——————
Distilled VAEScale Factor=2.02026.030.5630.95433.930.0334——4.08———
RAEf=14, #Tok.=256, Dim.=7682026.050.570.6118.8————226.8——
VAR#Tokens=2,240, Codebook=4,096, BPP=0.1025, Resolution=512 x 5122026.050.58—23.43———————
SD-VAEf=16, #Tok.=256, Dim.=162026.050.580.7225.7———————
Llama-Gen-32bpp (K x N)=0.22 (2^14 x 1024), ImageNet Trained Epochs=40, Params(M)=702025.120.590.76824.440.064——————
SD-VAETokenizer Category=Generation-oriented Tokenizer2026.010.6060.74526.260.133——————
SD-VAESemantics Distillation=false2025.070.61—————————
RAERes.=2562026.040.610.4419.2———————
Idealf=16, Size=16384, Dim.=64, #Res.=3842026.060.61———100——230.4——
MAE-TokTokenizer Category=Specialized, Resolution=5122026.030.62—————————
MAE-TokRes.=5122026.040.62—————————
TokenFlowTokenizer Category=Unified, Resolution=3842026.030.630.7322.77———————
TokenFlowRes.=3842026.040.630.7322.77———————
SD-VAEType=cont., #tok=10242026.030.64————223.8————
IBQ-16k#Tokens=1,024, Codebook=16,384, BPP=0.0547, Resolution=512 x 5122026.050.64—23.39———————
O-MAGVIT2-262k#Tokens=1,024, Codebook=262,144, BPP=0.0703, Resolution=512 x 5122026.050.65—23.98———————
FAEf=14, #Tok.=256, Dim.=642026.050.66—————————
l-DeTokSemantics Distillation=false2025.070.68—————————
LlamaGen-T2I#Tokens=1,024, Codebook=16,384, BPP=0.0547, Resolution=512 x 5122026.050.69—22.8———————
InsightTok#Tokens=1,024, Codebook=16,384, BPP=0.0547, Resolution=512 x 5122026.050.69—23.64———————
ImageFolderf=-, Size=32768, Dim.=32, #Tok.=2862026.050.69——————201.5100—
ImageFolderf=-, Size=32768, Dim.=32, #Res.=2562026.060.69———100——201.5——
TexTokTokenizer Category=Specialized, Resolution=5122026.030.730.6624.45———————
TexTokRes.=5122026.040.730.6624.45———————
NVG#Tokens=511*, Ratio=16, Codebook=4,0962025.080.74——0.1875100—————
Distilled VAEScale Factor=1.52026.030.8220.92931.330.0395——7.43———
l-DeTok + DistillationSemantics Distillation=true2025.070.85—————————
VFMTokf=-, Size=16384, Dim.=12, #Tok.=2562026.050.89——————215.4100—
VFMTokf=-, Size=16384, Dim.=12, #Res.=3362026.060.89———100——215.4——
OptVQLatent Size=16 x 16 x 8, From Scratch=true, #Tokens=16,3842024.120.910.72927.570.066——————
VARf=16, Size=4096, Dim.=32, #Tok.=6802026.050.92——————196—100
VARf=16, Size=4096, Dim.=32, #Res.=2562026.060.92———100——196——
Chameleon#Tokens=1,024, Codebook=8,192, BPP=0.0508, Resolution=512 x 5122026.050.93—22.53———————
Wan2.1-VAETokenizer Category=Generation-oriented Tokenizer2026.010.9450.88631.340.058——————
VQGANf=8, #Tok.=576, Ref.=[72]2026.050.95——————197.3—99.7
LlamaGenf=16, Size=16384, Dim.=8, #Res.=3842026.060.95———99.7——197.3——
CosmosTokenizer Category=Generation-oriented Tokenizer2026.010.9590.725.070.167——————
BSQbpp (K x N)=0.28 (2^18 x 1024), ImageNet Trained Epochs=200, Params(M)=1752025.120.990.81727.780.063——————
OptVQLatent Size=16 x 16 x 4, From Scratch=true, #Tokens=16,3842024.1210.71726.590.076——————
IBQ#Tokens=16 × 16, Ratio=16, Codebook=262,1442025.081——0.20384—————
LooC-VAEK x d*=1024 x 42026.011.010.71629.15—100—————
LooC-VAECodebook size (K x d*)=1024 × 42026.011.010.71629.150.716100—————
Distilled VAEScale Factor=0.52026.031.0310.85926.990.0478——20.04———
GQbpp (K x N)=0.07 (2^14 x 256), ImageNet Trained Epochs=40, Params(M)=872025.121.040.64222.30.116——————
TiTokf=-, Size=8192, Dim.=64, #Tok.=2562026.051.05——————191.5100—
TiTokf=-, Size=8192, Dim.=64, #Res.=2562026.061.05———100——191.5——
RAE (CLIP)Tokenizer Category=Unified Tokenizer2026.011.060.40317.440.324——————
VAR#Tokens=680, Ratio=16, Codebook=4,0962025.081.06——0.1863100—————
SD-VAETokenizer Category=Specialized, Resolution=2562026.031.060.8628.62———————
SD-VAERes.=2562026.041.060.8628.62———————
TokenBridgebpp (K x N)=0.375 (2^6 x 4096), Params(M)=832025.121.11—————————
Mo-VQGANLatent Size=16 x 16 x 4, From Scratch=true, #Tokens=1,0242024.121.120.67322.420.113——————
MoVQK x d*=1024 x 642026.011.120.673122.42—63—————
MoVQCodebook size (K x d*)=1024 × 642026.011.120.673122.420.113263—————
VQGAN-SDbpp (K x N)=0.22 (2^14 x 1024), ImageNet Trained Epochs=OpenImages, Params(M)=832025.121.14—————————
DINO-Tokf=16, Size=16384×2, Dim.=832, #Res.=2562026.061.15—————————
OpenMagViT-V2bpp (K x N)=0.07 (2^14 x 256), ImageNet Trained Epochs=300, Params(M)=1702025.121.170.6421.630.111——————
Open-MAGVIT2#Tokens=16 × 16, Ratio=16, Codebook=262,1442025.081.17——0.2038100—————