List your company
ResearchFormerly WizWand

Text-to-Image Generation on GenEval (GenEval Metric)

0.96GenEval ScoreUDM-GRPO
0.46080.59040.720.8496Mar 9, 2025May 16, 2025Jul 24, 2025Oct 1, 2025Dec 8, 2025Feb 15, 2026Apr 25, 2026
Updated 6mo ago

Evaluation results

MethodLinks
UDM-GRPO2026.040.96
SD 3.5 M + FlowGRPO#Steps=>5K, NFE_pi_theta_old=40, NFE_pi_theta=402026.040.95
SD 3.5 M + DiffusionNFT#Steps=1.7K, NFE_pi_theta_old=40 + 40, NFE_pi_theta=402026.040.94
SD 3.5 M + V-GRPO#Steps=580, NFE_pi_theta_old=40 + 6.9, NFE_pi_theta=6.92026.040.91
UniGRPOThinking=true2026.030.9
FlowGRPOThinking=false2026.030.88
TextGRPOThinking=true2026.030.88
FPOThinking=false2026.030.87
ReFL (w/ Thinking) + TextGRPOThinking=true2026.030.87
Nucleus-ImageConfiguration=1024 × 10242026.040.87
FlowGRPOThinking=true2026.030.86
Tar-7B W/Self-ReflectParams=7B, Type=AR, #Training Images=46M, Res.=2562025.030.85
Ming-UniVisionParams=16B-A3B, Type=AR-Continuous, Res.=5122025.030.85
ReFLThinking=false2026.030.85
FPOThinking=true2026.030.85
Tar-7BParams=7B, Type=AR, #Training Images=46M, Res.=2562025.030.84
ReFLThinking=true2026.030.84
SFTThinking=false2026.030.83
ReFL (w/ Thinking)Thinking=true2026.030.83
BAGEL2026.030.82
SFTThinking=true2026.030.82
BLIP3o-4B2026.030.81
xLARD2026.030.81
Janus-Pro-7B2026.030.8
OmniGen22026.030.8
SANA-1.5AutoEncoders=DC-AE(f32c32p1), Token-nums=32×32, Params (B)=4.8, Throughput (img/s)=0.262026.030.8
UniWorld-V1Type=AR+Diffusion, Params=20B2026.030.8
Janus-proType=AR, Params=7B2026.030.8
BagelThinking=true2026.030.79
BAGEL*Type=AR+RAG, Params=14B, Note=Re-generated using official pre-trained models2026.030.79
Tar-1.5B W/Self-ReflectParams=3B, Type=AR, #Training Images=46M, Res.=2562025.030.78
BagelThinking=false2026.030.78
FIRM-SD3.5Base Model=SD3.5-Medium, Training=RL with FIRM-Gen-8B2026.030.77
OmniGen2*Type=Diffusion+AR, Params=7B, Note=Re-generated using official pre-trained models2026.030.77
Tar-1.5BParams=3B, Type=AR, #Training Images=46M, Res.=2562025.030.76
Show-o22026.030.76
SD3.5-MediumTraining=RL with Qwen3VL-8B2026.030.76
ILLUME+Params=3B, Type=AR, #Training Images=46M, Res.=5122025.030.72
SemHiTokParams=7B, Type=AR, #Training Images=15M, Res.=2562025.030.71
SD3.5-L2026.040.71
SD 3.5 LResolution=1024×10242026.040.71
LACON-AModel Scale=Qwen3-1.7B2026.030.703
SD3.5-MediumTraining=RL with Qwen3VL-32B2026.030.7
LACON-SModel Scale=Qwen3-1.7B2026.030.697
URSAClassifier-Free Guidance (CFG)=true2026.040.69
OmniGenType=AR+Diffusion, Params=7B2026.030.68
Show-OType=AR+Diffusion, Params=1.3B2026.030.68
FLUX-devType=Diffusion, Params=12B2026.030.68
DALL-E 3Type=Diff, Res.=10242025.030.67
DALLE-32026.030.67
FLUX-devAutoEncoders=FLUX-VAE(f8c16p2), Token-nums=64×64, Params (B)=12, Throughput (img/s)=0.042026.030.67
Baseline-BModel Scale=Qwen3-1.7B2026.030.664
Emu3Params=8B, Type=AR, Res.=2562025.030.66
FLUX.1-Dev2026.030.66
FLUX.1-Dev2026.040.66
SD 3.5 M + FlowGRPO#Steps=2K, NFE_pi_theta_old=40, NFE_pi_theta=402026.040.66
UniFlowModel Size=0.6B, Type=Diffusion, Res.=2562025.100.65
SANAModel Size=0.6B, Type=Diffusion, Res.=5122025.100.64
DA-VAE (SD3.5-M)AutoEncoders=DA-VAE (f16c32p2), Token-nums=32×32, Params (B)=2.5, Throughput (img/s)=1.032026.030.64
TransfusionParams=7.3B, Type=AR, #Training Images=3.5B, Res.=2562025.030.63
TokenFlowParams=7B, Type=AR, #Training Images=60M, Res.=2562025.030.63
MMaDAParams=8B, Type=Diff, Res.=5122025.030.63
JanusFlow2026.030.63
Hunyuan-DiTAutoEncoders=—, Token-nums=64×64, Params (B)=1.5, Throughput (img/s)=0.052026.030.63
SD3.5-mediumAutoEncoders=SD3-VAE (f8c16p2), Token-nums=64×64, Params (B)=2.5, Throughput (img/s)=0.252026.030.63
SD3.5-medium†AutoEncoders=SD3-VAE (f8c16p2), Token-nums=32×32, Params (B)=2.5, Throughput (img/s)=1.032026.030.63
SD 3.5 M + CFGCFG=true2026.040.63
Ming-Lite-UniParams=8B, Type=AR-Scale, #Training Images=5M, Res.=5122025.030.62
SD3-mediumAutoEncoders=SD3-VAE (f8c16p2), Token-nums=64×64, Params (B)=2.0, Throughput (img/s)=0.362026.030.62
LACON-SModel Scale=Qwen3-0.6B2026.030.613
LACON-AModel Scale=Qwen3-0.6B2026.030.613
JanusParams=1.3B, Type=AR, #Training Images=65M, Res.=3842025.030.61
SynerGen-VLParams=2.4B, Type=AR, #Training Images=25M, Res.=2562025.030.61
ILLUMEParams=7B, Type=AR, #Training Images=15M, Res.=5122025.030.61
UniTokParams=7B, Type=AR, #Training Images=30M, Res.=2562025.030.59
Baseline-BModel Scale=Qwen3-0.6B2026.030.589
MUSE-VLParams=7B, Type=AR, #Training Images=10M, Res.=2562025.030.57
SDXLParams=2.6B, Type=Diff, #Training Images=2000M, Res.=10242025.030.55
LiquidParams=7B, Type=AR, #Training Images=30M, Res.=5122025.030.55
SD XLModel Size=2.6B, Type=Diffusion, Res.=10242025.100.55
TokenFlowModel Size=7B, Type=AR, Res.=2562025.100.55
SD-XL2026.030.55
SDXL2026.040.55
SD XLResolution=1024×10242026.040.55
EMU3-GenModel Size=8B, Type=AR, Res.=5122025.100.54
EMU3-Gen2026.030.54
PixArt-ΣAutoEncoders=—, Token-nums=64×64, Params (B)=0.6, Throughput (img/s)=0.402026.030.54
Emu3Type=AR, Params=8B2026.030.54
SD 3.5 M + FlowGRPO#Steps=4K, NFE_pi_theta_old=40, NFE_pi_theta=402026.040.54
Show-oParams=1.5B, Type=Diff, #Training Images=36M, Res.=2562025.030.53
Show-oModel Size=1.5B, Type=AR+Diffusion, Res.=2562025.100.53
Show-o2026.030.53
DALL-E 2Type=Diff, #Training Images=650M, Res.=10242025.030.52
SD3.5-Medium2026.030.52
LongAlign2026.040.5075
SD v2.1Type=Diff, #Training Images=2000M, Res.=10242025.030.5
SD v2.1Model Size=866M, Type=Diffusion, Res.=5122025.100.5
ELLA2026.040.4971
PRISM2026.040.496
PixArt-αParams=0.6B, Type=Diff, Res.=10242025.030.48