List your company

Image Generation on ImageNet 256x256 (test)

1.36FIDSiT-XL/2 + REG
0.5426.063511.58517.1065Oct 9, 2023Mar 22, 2024Sep 3, 2024Feb 15, 2025Jul 30, 2025Jan 11, 2026Jun 25, 2026

Evaluation results

MethodLinks
SiT-XL/2 + REGEpochs=800, w/o Ext.=false, CFG=true2026.051.36———299.477664.25
SiT-XL/2 + AHPAEpochs=800, w/o Ext.=true, CFG=true2026.051.4———317.581644.12
SiT-XL/2 + REPAEpochs=800, w/o Ext.=false, CFG=true2026.051.42———311.480634.7
MaskBitLatent Representation=discrete, #tokens=256, codebook size=16384, #Params=305M, #Steps=2562024.091.52———328.6———
SiT-XL/2 + SRA 2Epochs=800, w/o Ext.=true, CFG=true2026.051.52———316.282624.63
MAR-HLatent Representation=continuous, #tokens=256x16, #Params=943M, #Steps=256, Concurrent=true2024.091.55———303.7———
SiT-XL/2 + AHPAEpochs=400, w/o Ext.=true, CFG=true2026.051.56———308.281634.15
MDTv2Latent Representation=continuous, #tokens=1024x4, #Params=676M, #Steps=250, Tokenizer training data=OpenImages, Concurrent=true2024.091.58———314.7———
SiT-XL/2 + SRAEpochs=800, w/o Ext.=true, CFG=true2026.051.58———305.780634.65
MaskBitLatent Representation=discrete, #tokens=256, codebook size=16384, #Params=305M, #Steps=642024.091.62———338.7———
MaskBitLatent Representation=discrete, #tokens=256, codebook size=4096, #Params=305M, #Steps=642024.091.65———341.8———
RAEv2K=7, Training-free Encoder=true, Recon-Gen Tradeoff Control=true, Epochs=802026.051.65———228———
DiMRLatent Representation=continuous, #tokens=1024x4, #Params=505M, #Steps=250, Tokenizer training data=OpenImages, Concurrent=true2024.091.7———289———
DiffiTClass=Diffusion2023.121.73———276.498062—
MAGVIT-v2Latent Representation=discrete, #tokens=256, codebook size=262144, #Params=307M, #Steps=642024.091.78———319.4———
SiT-XL/2 + AHPAEpochs=200, w/o Ext.=true, CFG=true2026.051.78———291.882614.26
MDTLatent Representation=continuous, #tokens=1024x4, #Params=676M, #Steps=250, Tokenizer training data=OpenImages, LAION-Aesthetics/-Humans2024.091.79———283———
VARLatent Representation=discrete, #tokens=680, codebook size=4096, #Params=2.0B, #Steps=10, Tokenizer training data=OpenImages, Concurrent=true2024.091.92———350.2———
TiTok-S-128Latent Representation=discrete, #tokens=128, codebook size=4096, #Params=287M, #Steps=64, Concurrent=true2024.091.97———281.8———
DiT-XL/2 + iSSDEpochs=1400+0.8, w/o Ext.=false, CFG=true2026.052.02———25081604.22
SiT-XL/2Epochs=1400, w/o Ext.=true, CFG=true2026.052.06———270.382594.5
MaskGITLatent Representation=discrete, Tokenizer=VQGAN+ (ours), #tokens=256, codebook size=4096, #Params=310M, #Steps=642024.092.12———300.8———
SiT-XL/2 + AHPAEpochs=100, w/o Ext.=true, CFG=true2026.052.15———265.481584.38
LlamaGen-3BLatent Representation=discrete, #tokens=576, codebook size=16384, #Params=3.1B, #Steps=576, Concurrent=true2024.092.18———263.3———
RAETraining-free Encoder=true, Recon-Gen Tradeoff Control=false, Epochs=802026.052.23———214.8———
RPiAETraining-free Encoder=false, Recon-Gen Tradeoff Control=false, Epochs=802026.052.25———208.7———
DiT-XL/2-GClass=Diffusion2023.122.27———278.248357—
DiT-XL/2-GLatent Representation=continuous, #tokens=1024x4, #Params=675M, #Steps=250, Tokenizer training data=OpenImages, LAION-Aesthetics/-Humans2024.092.27———278.2———
DiT-XL/2Epochs=1400, w/o Ext.=true, CFG=true2026.052.27———278.283574.6
MaskDiTEpochs=1600, w/o Ext.=false, CFG=true2026.052.28———276.680615.67
U-ViTLatent Representation=continuous, #tokens=1024x4, #Params=501M, #Steps=50, Tokenizer training data=OpenImages, LAION-Aesthetics/-Humans2024.092.29———263.9———
StyleGAN-XLParam=166M, Epoch=-, Guidance=CG + cGAN, NFE=12025.112.3———————
AF-XL/2Param=673M, Epoch=125, Guidance=CG + DA, NFE=12025.112.38———————
GIVT-L-ALatent Representation=continuous, #tokens=256x32, #Params=1.67B, #Steps=2562024.092.59———————
AF-L/2Param=457M, Epoch=120, Guidance=CG + DA, NFE=12025.112.63———————
AlphaFlow-XL/2Param=676M, Epoch=240, Guidance=CFG, NFE=12025.112.81———————
AF-M/2Param=306M, Epoch=120, Guidance=CG + DA, NFE=12025.112.82———————
GAT-XL/2+REPAParam=602M, Epoch=40, Guidance=DA + cGAN, NFE=12025.112.96———————
RAEv2K=23, Training-free Encoder=true, Recon-Gen Tradeoff Control=true, Epochs=802026.053.02———206———
ViT-VQGANLatent Representation=discrete, #tokens=1024, codebook size=8192, #Params=1.7B, #Steps=10242024.093.04———227.4———
AF-B/2Param=130M, Epoch=200, Guidance=CG + DA, NFE=12025.113.05———————
DINO-SAETraining-free Encoder=false, Recon-Gen Tradeoff Control=false, Epochs=802026.053.07———209.7———
DiffMoE-L + SharpMoE# Activated Params.=470M, # Total Params.=1.99B, cfg=1.5, Sampling Steps=2502026.063.1———228.88———
SD-DiTEpochs=480, w/o Ext.=false, CFG=true2026.053.23———————
TiM-XL/2Param=664M, Epoch=300, Guidance=CFG, NFE=12025.113.26———————
EC-DiT-L + SharpMoE# Activated Params.=484M, # Total Params.=1.18B, cfg=1.5, Sampling Steps=2502026.063.27———221.36———
ReDi-XL/2#Params=675M, Iter.=4M, Classifier-Free Guidance=false2026.043.3———————
CoReDi-XL/2#Params=675M, Iter.=2M, Classifier-Free Guidance=false2026.043.3———————
VFM-VAETraining-free Encoder=false, Recon-Gen Tradeoff Control=false, Epochs=802026.053.41———160.4———
MeanFlow-XL/2Param=676M, Epoch=240, Guidance=CFG, NFE=12025.113.43———————
GigaGANParam=569M, Epoch=480, Guidance=Match-loss, NFE=12025.113.45———————
LDMLatent Shape=64x64x32023.103.6———————
LDM-4-GClass=Diffusion2023.123.6———247.678748—
LDM-4-GLatent Representation=continuous, #tokens=4096x3, #Params=400M, #Steps=250, Tokenizer training data=OpenImages2024.093.6———247.7———
AlignTokTraining-free Encoder=false, Recon-Gen Tradeoff Control=false, Epochs=802026.053.71———148.9———
TC-DiT-L + SharpMoE# Activated Params.=484M, # Total Params.=1.18B, cfg=1.5, Sampling Steps=2502026.063.72———206.93———
RQ-TransformerLatent Representation=discrete, #tokens=256, codebook size=16384, #Params=3.8B, #Steps=642024.093.8———323.7———
MeanFlow-L/2Param=459M, Epoch=240, Guidance=CFG, NFE=12025.113.84———————
DiffMoE-L# Activated Params.=458M, # Total Params.=1.98B, cfg=1.5, Sampling Steps=2502026.063.86———203———
MaskGITLatent Representation=discrete, #tokens=256, codebook size=1024, #Params=177M, #Steps=82024.094.02———355.6———
EC-DiT-L# Activated Params.=458M, # Total Params.=1.16B, cfg=1.5, Sampling Steps=2502026.064.09———195.12———
Dense-DiT-L# Activated Params.=458M, # Total Params.=458M, cfg=1.5, Sampling Steps=2502026.064.74———183.54———
MeanFlow-M/2Param=308M, Epoch=240, Guidance=CFG, NFE=12025.115.01———————
TC-DiT-L# Activated Params.=458M, # Total Params.=1.16B, cfg=1.5, Sampling Steps=2502026.065.07———174.98———
AlphaFlow-B/2Param=131M, Epoch=240, Guidance=CFG, NFE=12025.115.4———————
Taming-VQGANLatent Representation=discrete, #tokens=256, codebook size=1024, #Params=1.4B, #Steps=2562024.095.88———304.8———
REPA-XL/2#Params=675M, Iter.=4M, Classifier-Free Guidance=false2026.045.9———————
DINO-TokTraining-free Encoder=false, Recon-Gen Tradeoff Control=false, Epochs=802026.055.94———152.6———
MeanFlow-B/2Param=131M, Epoch=240, Guidance=CFG, NFE=12025.116.17———————
DiffMoE-B + SharpMoE# Activated Params.=133M, # Total Params.=562M, cfg=1.5, Sampling Steps=2502026.066.66———155.64———
BigGANParam=112M, Epoch=-, Guidance=cGAN, NFE=12025.116.95———————
EC-DiT-B + SharpMoE# Activated Params.=137M, # Total Params.=336M, cfg=1.5, Sampling Steps=2502026.067.67———144.5———
DiffMoE-B# Activated Params.=130M, # Total Params.=559M, cfg=1.5, Sampling Steps=2502026.068.03———138.49———
SiT-XL/2#Params=675M, Iter.=7M, Classifier-Free Guidance=false2026.048.3———————
FunctaLatent Shape=32x32x642023.108.8———————
TC-DiT-B + SharpMoE# Activated Params.=137M, # Total Params.=336M, cfg=1.5, Sampling Steps=2502026.069.28———130.52———
EC-DiT-B# Activated Params.=130M, # Total Params.=329M, cfg=1.5, Sampling Steps=2502026.069.28———126.6———
OursLatent Shape=256x2562023.109.3———————
Dense-DiT-B# Activated Params.=130M, # Total Params.=130M, cfg=1.5, Sampling Steps=2502026.0610.07———122.15———
Shortcut-XL/2Param=675M, Epoch=250, Guidance=CFG, NFE=12025.1110.6———————
DiT + DiffusionBlocksBackbone=DiT-L/2, Layers=24, B (blocks)=3, Resolution=256x2562025.0610.63———————
SpatialLatent Shape=16x16x2562023.1011.7———————
DiTBackbone=DiT-L/2, Layers=24, Resolution=256x2562025.0612.09———————
TC-DiT-B# Activated Params.=130M, # Total Params.=329M, cfg=1.5, Sampling Steps=2502026.0612.82———104.37———
DiffMoE-L + SharpMoE# Activated Params.=470M, # Total Params.=1.99B, cfg=1.0, Sampling Steps=2502026.0613.93———93.66———
EC-DiT-L + SharpMoE# Activated Params.=484M, # Total Params.=1.18B, cfg=1.0, Sampling Steps=2502026.0614.82———88.02———
DiffMoE-L# Activated Params.=458M, # Total Params.=1.98B, cfg=1.0, Sampling Steps=2502026.0615.13———83.62———
CoReDi-B/2#Params=130M, Iter.=400K, Classifier-Free Guidance=false2026.0416.4———————
VeCoRBackbone=SiT-XL/2, NFEs=50, Sampling Method=Heun2, Resolution=256x2562025.1116.55———76.1765627.21
EC-DiT-L# Activated Params.=458M, # Total Params.=1.16B, cfg=1.0, Sampling Steps=2502026.0616.61———78.2———
TC-DiT-L + SharpMoE# Activated Params.=484M, # Total Params.=1.18B, cfg=1.0, Sampling Steps=2502026.0616.63———81.29———
SiT-XL/2Backbone=SiT-XL/2, NFEs=50, Sampling Method=Heun2, Resolution=256x2562025.1118.59———72.0563645.3
Dense-DiT-L# Activated Params.=458M, # Total Params.=458M, cfg=1.0, Sampling Steps=2502026.0619.02———71.61———
DiffMoE-S + SharpMoE# Activated Params.=34M, # Total Params.=141M, cfg=1.5, Sampling Steps=2502026.0619.04———80.07———
VeCoRBackbone=SiT-L/2, NFEs=50, Sampling Method=Heun2, Resolution=256x2562025.1119.81———66.1663627.41
TC-DiT-L# Activated Params.=458M, # Total Params.=1.16B, cfg=1.0, Sampling Steps=2502026.0620.53———69.01———
TC-DiT-S + SharpMoE# Activated Params.=35M, # Total Params.=85M, cfg=1.5, Sampling Steps=2502026.0620.55———73.94———
DiffMoE-S# Activated Params.=33M, # Total Params.=140M, cfg=1.5, Sampling Steps=2502026.0620.59———73.46———
ReDi-B/2#Params=130M, Iter.=400K, Classifier-Free Guidance=false2026.0421.4———————
EC-DiT-S + SharpMoE# Activated Params.=35M, # Total Params.=85M, cfg=1.5, Sampling Steps=2502026.0621.81———70.56———