List your company
ResearchFormerly WizWand

Class-conditional generation on ImageNet 256 x 256 1k (val)

1.13FIDDiTDH-XL + RAE
0.9282.29153.6555.0185Oct 1, 2025Oct 22, 2025Nov 12, 2025Dec 3, 2025Dec 24, 2025Jan 14, 2026Feb 4, 2026
Updated 4mo ago

Evaluation results

MethodLinks
DiTDH-XL + RAE# Params=839M, NFE=50×2, CFG=true2025.121.13——262.6——
xAR-H# Params=1.1B, NFE=⋆, CFG=true2025.121.24——301.6——
DDT-XL/2# Params=675M, NFE=250×2, CFG=true2025.121.26——310.6——
DDT-XL/2NFE=250x2, #Params=675M, #Epochs=4002026.021.26—————
LightningDiT-XL/1# Params=675M, NFE=250×2, CFG=true2025.121.35——295.3——
Light.DiTNFE=250x2, #Params=675M, #Epochs=8002026.021.35—————
SiT-XL/2 + REPA# Params=675M, NFE=250×2, CFG=true2025.121.42——305.7——
REPA-XL/2NFE=250x2, #Params=675M, #Epochs=8002026.021.42—————
RAR-XXL# Params=1.5B, NFE=⋆, CFG=true2025.121.48——326——
RAR-XXLTokenizer=VQ, Model Type=Raster-scan AR, # Params=1.5B2025.111.48——3268063
RAR-XLTokenizer=VQ, Model Type=Raster-scan AR, # Params=955M2025.111.5——306.98062
MAR-H# Params=943M, NFE=⋆, CFG=true2025.121.55——303.7——
MAR-HTokenizer=VAE, Model Type=Bidirectional AR, # Params=943M2025.111.55——303.78162
STAR-XXLTokenizer=VQ, Model Type=Randomized AR, # Params=1.5B2025.111.55——338.88162
STAR-XLTokenizer=VQ, Model Type=Randomized AR, # Params=955M2025.111.65——333.28062
RAR-LTokenizer=VQ, Model Type=Raster-scan AR, # Params=461M2025.111.7——299.58160
iMF-XL/2# Params=610M, NFE=1, CFG=true2025.121.72——282——
VAR-r30-reNFE=10x2, #Params=2.0B, #Epochs=3502026.021.73—————
FACM-XL/2# Params=675M, NFE=1, CFG=true2025.121.76——290——
MAR-LTokenizer=VAE, Model Type=Bidirectional AR, # Params=479M2025.111.78——2968160
DUMO-XL/2 (longer training)NFE=2, #Params=679M, #Epochs=4322026.021.79—————
JiT-G/16# Params=2B, NFE=100×2, CFG=true2025.121.82——292.6——
UCGM⊕DDT-XL/2NFE=2, #Params=675M, #Epochs=4012026.021.9—————
VAR-d30# Params=2.0B, NFE=⋆, CFG=true2025.121.92——323.1——
VAR-d30Tokenizer=VQ, Model Type=Block-wise AR, # Params=2.0B2025.111.92——323.18259
UCGM-XL/2NFE=2, #Params=675M, #Epochs=4242026.021.94—————
REPA SiT-XL/2 + VeCoRw=1.85, sigma_low=0.0, sigma_high=0.652025.111.94—4.45———
UCGM⊕REPA-XL/2NFE=2, #Params=675M, #Epochs=8012026.021.95—————
RAR-BTokenizer=VQ, Model Type=Raster-scan AR, # Params=261M2025.111.95——290.58258
REPA-XL/2NFE=250x2, #Params=675M, #Epochs=2002026.021.96—————
REPA SiT-XL/2 + ΔFMw=1.85, sigma_low=0.0, sigma_high=0.652025.111.97—4.49———
STAR-LTokenizer=VQ, Model Type=Randomized AR, # Params=461M2025.111.98——3228258
SiT-XL/2# Params=675M, NFE=250×2, CFG=true2025.122.06——252.2——
SiT-XL/2NFE=250x2, #Params=675M, #Epochs=14002026.022.06—————
UCGM⊕Light.DiTNFE=2, #Params=675M, #Epochs=8012026.022.06—————
SiT-XL/2#Parameters=675M2025.102.06—————
REPA SiT-XL/2w=1.75, sigma_low=0.0, sigma_high=0.752025.112.09—5.55———
VAR-d24Tokenizer=VQ, Model Type=Block-wise AR, # Params=1.0B2025.112.09——312.98259
UCGM-XL/2NFE=1, #Params=675M, #Epochs=4242026.022.1—————
Light.DiTNFE=250x2, #Params=675M, #Epochs=642026.022.11—————
RCG, conditional# Params=512M, NFE=⋆, CFG=true2025.122.12——267.7——
RandAR-XXLTokenizer=VQ, Model Type=Randomized AR, # Params=1.4B2025.112.15——3227962
DMF-XL/2+# Params=675M, NFE=1, CFG=true2025.122.16—————
MeanFlow-XL/2 (longer training)NFE=2, #Params=676M, #Epochs=10002026.022.2—————
STAR-BTokenizer=VQ, Model Type=Randomized AR, # Params=261M2025.112.24——295.38257
RandAR-XLTokenizer=VQ, Model Type=Randomized AR, # Params=775M2025.112.25——314.28060
DiT-XL/2# Params=675M, NFE=250×2, CFG=true2025.122.27——278.2——
DiT-XL/2NFE=250x2, #Params=675M, #Epochs=14002026.022.27—————
DiT-XL/2#Parameters=675M2025.102.27—————
DiT-XL/2Tokenizer=VAE, Model Type=Diffusion, # Params=675M2025.112.27——278.28357
U-ViT-H/2NFE=50x2, #Params=501M, #Epochs=4002026.022.29—————
UViT-XL/2Tokenizer=VAE, Model Type=Diffusion, # Params=501M2025.112.29——263.98257
StyleGAN-XL# Params=166M, NFE=1, CFG=true2025.122.3——265.1——
MAR-BTokenizer=VAE, Model Type=Bidirectional AR, # Params=208M2025.112.31——281.78257
BiFlow-B/2# Params=133M, NFE=1, CFG=true2025.122.39——303——
STARFlow-XL/1# Params=1.4B, NFE=⋆, CFG=true2025.122.4—————
Open-MAGVIT2-L#Parameters=804M2025.102.51—————
RandAR-LTokenizer=VQ, Model Type=Randomized AR, # Params=343M2025.112.55——288.88158
VAR-d20Tokenizer=VQ, Model Type=Block-wise AR, # Params=600M2025.112.57——302.68356
α-Flow-XL/2+# Params=676M, NFE=1, CFG=true2025.122.58—————
DUMO-XL/2 (longer training)NFE=1, #Params=679M, #Epochs=4322026.022.6—————
DUMO-XL/2NFE=2, #Params=679M, #Epochs=2402026.022.82—————
π-Flow-XL/2# Params=675M, NFE=1, CFG=true2025.122.85—————
π-Flow⊕REPA-XL/2NFE=1, #Params=675M, #Epochs=4482026.022.85—————
MeanFlow-XL/2NFE=2, #Params=676M, #Epochs=2402026.022.93—————
ViT-VQGAN#Parameters=1.7B2025.103.04—————
LlamaGen-3BTokenizer=VQ, Model Type=Raster-scan AR, # Params=3.0B2025.113.05——222.38058
DUMO-XL/2NFE=1, #Params=679M, #Epochs=2402026.023.07—————
LlamaGen-XXLTokenizer=VQ, Model Type=Raster-scan AR, # Params=1.4B2025.113.09——253.68353
TiM-XL/2# Params=664M, NFE=1, CFG=true2025.123.26——210.3——
VAR-d16Tokenizer=VQ, Model Type=Block-wise AR, # Params=310M2025.113.3——274.48451
π-FlowNFE=1, #Params=675M, #Epochs=4482026.023.34—————
LlamaGen-XL#Parameters=775M2025.103.39—————
LlamaGen-XLTokenizer=VQ, Model Type=Raster-scan AR, # Params=804M2025.113.39——227.18154
UViT-L/2Tokenizer=VAE, Model Type=Diffusion, # Params=287M2025.113.4——219.98352
MeanFlow-XL/2# Params=676M, NFE=1, CFG=true2025.123.43——247.5——
MeanFlow-XL/2NFE=1, #Params=676M, #Epochs=2402026.023.43—————
GigaGAN# Params=569M, NFE=1, CFG=true2025.123.45——225.5——
GigaGANNFE=1, #Params=569M2026.023.45—————
LDM-4-G# Params=400M, NFE=250×2, CFG=true2025.123.6——247.7——
LDM-4#Parameters=400M2025.103.6—————
RQTransformer#Parameters=3.8B2025.103.8—————
LlamaGen-LTokenizer=VQ, Model Type=Raster-scan AR, # Params=343M2025.113.8——248.38351
Purrception#Parameters=750M, τ=0.9, cfg=1.3, Tokenizer=vq-ds8-c2i, ODE solver=Euler, Integration steps=2502025.103.88—————
IMM-XL/2 (ω = 1.5)NFE=2x2, #Params=675M, #Epochs=38402026.023.99—————
iTARFlow-XL/2# Params=690M, NFE=⋆, CFG=true2025.124.54——259.3——
ADM-G# Params=554M, NFE=250×2, CFG=true2025.124.59—————
ADM-GNFE=250x2, #Params=559M, #Epochs=3962026.024.59—————
iTARFlow-L/2# Params=448M, NFE=⋆, CFG=true2025.124.82——254.8——
CDM2025.104.88—————
DiT-L/2Tokenizer=VAE, Model Type=Diffusion, # Params=458M2025.115.02——167.27557
VQGAN#Parameters=1.4B2025.105.2—————
iTARFlow-M/2# Params=296M, NFE=⋆, CFG=true2025.125.22——255.5——
Implicit Timestep Model#Parameters=546M2025.105.3—————
IMM-XL/2NFE=2x2, #Params=675M, #Epochs=38402026.025.33—————
TARFlow-XL/8@pix# Params=1.3B, NFE=⋆, CFG=true2025.125.56—————
VQ-Diffusion2025.105.84—————
MaskGIT# Params=227M, NFE=⋆, CFG=true2025.126.18——182.1——
MaskGITNFE=8, #Params=227M, #Epochs=3002026.026.18—————
MaskGIT#Parameters=227M2025.106.18—————