List your company

Image Generation on ImageNet 256x256 (val)

0.58FIDVQGAN-f-4
0.52080.92041.321.7196Dec 20, 2021Sep 19, 2022Jun 20, 2023Mar 19, 2024Dec 18, 2024Sep 17, 2025Jun 18, 2026

Evaluation results

MethodLinks
VQGAN-f-4mode=first stage, Generator Compute (V100 days)=29, Overall Compute (V100 days)=29, Nparams=55M2021.120.58————————————————
JiT-H + FD-lossNFE=1, space=pixel, #params=953M2026.040.75313—7666———————2.65————
iMF-XL + FD-lossNFE=1, space=latent, #params=610M2026.040.76301.3—7767———————2.45————
JiT-L + FD-lossNFE=1, space=pixel, #params=459M2026.040.77317.3—7766———————3.24————
pMF-H + FD-lossNFE=1, space=pixel, #params=935M2026.040.77310.1—7768———————1.89————
pMF-L + FD-lossNFE=1, space=pixel, #params=410M2026.040.78309.2—7667———————2.09————
BAR-LNFE=256×2×4, space=discrete, #params=1.1B2026.041.01281.9—7768———————3.57————
REPA-EEpochs=800, Backbone=SiT-XL, Sampling protocol=Class-balanced, Fine-tuning autoencoders=true, Guidance Scale (w)=1.82026.021.12302.9—79664.09———————————
VQGAN-f-8mode=first stage, Generator Compute (V100 days)=66, Overall Compute (V100 days)=66, Nparams=68M2021.121.14————————————————
RAE-XLNFE=50×2†, space=latent, #params=839M2026.041.16261—7767———————3.26————
REPA-ENFE=250×2†, space=latent, #params=676M2026.041.17298.3—7966———————3.04————
xAR-Htype=xAR, Parameters=1.1B, Sampling Steps=502025.021.24—————————1.3——————
xAR-Htype=xAR, #params=1.1B2025.021.24301.6—8364————————————
DDT-XLNFE=250×2, space=latent, #params=675M2026.041.26309.3—7966———————5.7————
xAR-Ltype=xAR, Parameters=608M, Sampling Steps=502025.021.28—————————3.2——————
xAR-Ltype=xAR, #params=608M2025.021.28292.5—8262————————————
Ours (StableVM/VA-REPA)Epochs=400, Backbone=SiT-XL, Sampling protocol=Class-balanced, Guidance Scale (w)=1.82026.021.34305—80644.53———————————
SiD2, Flop HeavyNFE=5122024.101.38————————————————
SiD2NFE=512×2, Space=pixel2026.061.38————————————————
MAR-L, DeTokNFE=256×2×100, space=latent, #params=478M2026.041.39306.2—8162———————5.49————
REGEpochs=480, Backbone=SiT-XL, Guidance Scale (w)=1.82026.021.4296.9—77664.24———————————
REPAtype=Flow., Parameters=675M, Sampling Steps=2502025.021.42—————————0.6——————
REPAtype=Flow-Matching, #params=675M, guidance interval sampling=true2025.021.42305.7—8065————————————
REPAEpochs=800, Backbone=SiT-XL, Guidance Scale (w)=1.82026.021.42305.7—80654.7———————————
REPA-XL/2Params=675M + 86M, Epochs=800, NFE=250x2, Latency(s)=3.44, Type=Latent Diffusion Model (VAE)2025.111.42305.7—80644.7———————————
SiT-XL/2-REPANFE=250×2†, space=latent, #params=675M2026.041.42306.1—8065———————5.45————
LightningDiTNFE=250×2, space=latent, #params=675M2026.041.42294.3—8064———————4.57————
REPA-XL/2Params=675M + 86M, Epochs=800, GFLOPs=2382026.051.42305.7———————————————
SiT-XL/2 + REPANFE=250×2, Space=latent, Params=675M2026.061.42————————————————
Drift-L (pixel)NFE=1, space=pixel, #params=465M2026.041.43305.8—8160———————10.51————
Ours (StableVM/VA-REPA)Epochs=400, Backbone=SiT-XL, Sampling protocol=Standard REPA, Guidance Scale (w)=1.82026.021.47300.3—80634.51———————————
RAR-XXLtype=AR, #params=1.5B2025.021.48326—8063————————————
MaskBittype=Mask., #params=305M2025.021.52328.6———————————————
Drift-L (latent)NFE=1, space=latent, #params=463M2026.041.53257.2—7963———————10.92————
REGNFE=250×2†, space=latent, #params=685M2026.041.54302.9—7862———————4.64————
MAR-Htype=MAR, Parameters=943M, Sampling Steps=2562025.021.55—————————0.3——————
MAR-Htype=MAR, #params=943M2025.021.55303.7—8162————————————
PaGoDANFE=12024.101.56————————————————
MAR-HNFE=256×2×100, space=latent, #params=942M2026.041.56299.5—8062———————5.61————
CausalFusion-H†type=Diffusion, #params=1B, guidance interval sampling=true2025.021.57————————————————
MDTv2-XL/2type=Diffusion, #params=676M2025.021.58314.7—7965————————————
SiT-XL/2 + routing + SGFID#Epoch=4002026.011.58249.7—80634.45———————————
iMF-XLNFE=2, space=latent, #params=610M2026.041.61289.1—7963———————7.48————
PixelDiT-XLParams=797M, Epochs=320, GFLOPs=3112026.051.61292.7———————————————
PixelDiT-XL/16NFE=100×2, Space=pixel, Params=797M2026.061.61————————————————
DeCo-XL/16Params=682M, Epochs=800, NFE=250x2, Latency(s)=2.632025.111.62301—80624.41———————————
CrossFlow-XLNFE=1, Space=latent→pixel, Params=461M2026.061.62————————————————
DiMR-G/2Rtype=Diffusion, #params=1.1B2025.021.63292.5—7963————————————
FlowAR-Htype=AR, #params=1.9B2025.021.65296.5—8360————————————
REPA-EEpochs=80, Backbone=SiT-XL, Sampling protocol=Class-balanced, Fine-tuning autoencoders=true, Guidance Scale (w)=1.82026.021.67266.3—80634.12———————————
reference (real images)NFE=50k validation images, space=N/A, #params=N/A2026.041.68232.2—7566———————1————
FlowAR-HNFE=50×2†, space=latent, #params=1.9B2026.041.68274.1—8062———————6.13————
SiT-XL/2+REPA-EParam=675M, Epoch=800, Guidance=None, NFE=250, latent_space=representational2025.111.69————————————————
BitDance-B-4xType=AR, Order=raster, #Params=260M, Steps=64, Throughput=24.18 img/s, Parallel tokens per step=42026.021.69291.2—7963————————————
DeCo-XL/16 (Heun)Params=682M, Epochs=600, NFE=100x2, Latency(s)=1.05, Sampler=Heun2025.111.69304—79634.59———————————
CrossFlow-LNFE=1, Space=latent→pixel, Params=315M2026.061.69————————————————
Ours (StableVM/VA-REPA)Epochs=80, Backbone=SiT-XL, Sampling protocol=Class-balanced, Guidance Scale (w)=1.82026.021.71274.2—81614.54———————————
DeCo-XL/16Params=682M, Epochs=800, NFE=100x2, Latency(s)=1.052025.111.71304—80614.54———————————
xAR-Btype=xAR, Parameters=172M, Sampling Steps=502025.021.72—————————9.8——————
xAR-Btype=xAR, #params=172M2025.021.72280.4—8259————————————
SiD2, smallNFE=5122024.101.72————————————————
iMF-XL/2NFE=1, Space=latent, Params=659M2026.061.72————————————————
MAR-Ltype=AR, Parameters=479M, Sampling Steps=2562025.021.78—————————0.5——————
MAGVIT-v2type=Mask., #params=307M2025.021.78319.4———————————————
MAR-Ltype=MAR, #params=479M2025.021.78296—8160————————————
MAR-LType=Mask., Order=random, #Params=479M, Steps=256, Throughput=1.39 img/s2026.021.78296—8160————————————
MDT-XL/2#Epoch=13002026.011.79283.01—81614.57———————————
REPAtype=Flow-Matching, #params=675M, guidance interval sampling=false2025.021.8284—8161————————————
Ours (StableVM/VA-REPA)Epochs=80, Backbone=SiT-XL, Sampling protocol=Standard REPA, Guidance Scale (w)=1.82026.021.8272.4—81604.52———————————
SiT-XL/2 + REPA#Epoch=8002026.011.8284—81614.5———————————
MAR-LNFE=256×2×100, space=latent, #params=478M2026.041.8293.4—8060———————6.68————
iMF-XLNFE=1, space=latent, #params=610M2026.041.82278.9—7863———————8.39————
JiT-G/16Params=2B, Epochs=600, GFLOPs=7662026.051.82292.6———————————————
DiT-XL/2-G++Diffusion Space=Latent2022.111.83281.53—78645.16—70—————————
REGEpochs=80, Backbone=SiT-XL, Guidance Scale (w)=1.82026.021.86321.4—76634.49———————————
JiT-H/16params=953M2025.121.86303.4———————————————
JiT-H/16 (Heun)Params=953M, Epochs=600, NFE=100x2, Sampler=Heun2025.111.86303———————————————
RAE-XLParam=676M, Epoch=800, Guidance=None, NFE=250, latent_space=representational2025.111.87————————————————
RDMNFE=3562024.101.89————————————————
DeCo-XL/16Params=682M, Epochs=320, NFE=100x2, Latency(s)=1.052025.111.9303—80614.47———————————
DeCo-XL/16Params=682M, Epochs=320, GFLOPs=2372026.051.9303———————————————
BitDance-B-16xType=AR, Order=raster, #Params=260M, Steps=16, Throughput=90.26 img/s, Parallel tokens per step=162026.021.91283.8—7862————————————
FREPix-XLParams=674M, Epochs=320, GFLOPs=2302026.051.91295.6———————————————
iREPAEpochs=80, Backbone=SiT-XL, Guidance Scale (w)=1.82026.021.93268.8—80604.59———————————
PixNerd-XL/16Params=700M, Epochs=320, NFE=100x2, Latency(s)=1.182025.111.95300—80604.54———————————
TiTok-S-128type=Mask., #params=287M2025.021.97281.8———————————————
VAR-d30type=VAR, #params=2.0B2025.021.97323.1—8259————————————
VAR-d30NFE=10×2, space=discrete, #params=2B2026.041.97304.6—8259———————6.7————
JiT-HNFE=50×2×2†, space=pixel, #params=953M2026.041.97296—7863———————7.66————
REPAEpochs=80, Backbone=SiT-XL, Guidance Scale (w)=1.82026.021.98263—80614.6———————————
PixelFlow, XL/4params=677M2025.121.98282.1———————————————
PixelFlow-XL/4Params=677M, Epochs=320, NFE=120x2, Latency(s)=9.782025.111.98282.1—81605.83———————————
PixelFlow-XL/4Params=677M, Epochs=320, GFLOPs=58182026.051.98282.1———————————————
RDMParams=553M + 553M, Epochs=400, NFE=250, Latency(s)=38.4, Type=Latent Diffusion Model (VAE)2025.111.99260.4—81583.99———————————
SiT-XL/2type=Flow., Parameters=675M, Sampling Steps=2502025.022.02—————————0.5——————
Faster-DiTEpochs=400, Guidance Scale (w)=1.82026.022.03264—81604.63———————————
FasterDiT-XL/2#Epoch=4002026.012.03264—81604.63———————————
OUSACSteps=50, CFG-Steps=8, MACs (T)=5.07, Latency (s)=2.812025.122.04266.5—82.0958.74.29———————————
DiCo-XLType=Diff., Order=-, #Params=701M, Steps=250, Throughput=2.62 img/s2026.022.05282.2—8359————————————
SiT-XL/2type=Flow-Matching, #params=675M2025.022.06277.5—8359————————————