List your company
ResearchFormerly WizWand

Text-to-Image Generation on GenEval (Overall Score)

0.98Overall Score (GenEval)PRECISE
0.664880.746690.82850.91031May 8, 2026May 10, 2026May 12, 2026May 15, 2026May 17, 2026May 19, 2026May 22, 2026
Updated 5mo ago

Evaluation results

MethodLinks
PRECISENFE=10, Iterations=30002026.050.98
CPSNFE=10, Iterations=30002026.050.972
Flow-GRPONFE=10, Iterations=30002026.050.969
CGPOGPU=8, RL Method=CGPO2026.050.96
GRPO-GuardGPU=–, RL Method=GRPO2026.050.95
Flow-GRPOGPU=24, RL Method=GRPO2026.050.95
Flow-GRPOGPU=8, RL Method=GRPO2026.050.94
ReCA-HarmonParams=1.5B, LLM rewriter=false2026.050.9
UniLIPParams=2B, LLM rewriter=false2026.050.9
SGT-BAGELParams=7B+7B‡, LLM rewriter=true2026.050.9
AlphaGRPOResolution=1024, Self-Reflective Refinement=true2026.050.895
AlphaGRPOResolution=512, Self-Reflective Refinement=true2026.050.882
BAGELParams=7B+7B‡, LLM rewriter=true2026.050.88
AlphaGRPO (RT2I)Resolution=1024, Reasoning text-to-image generation task=RT2I2026.050.874
Qwen-ImageType=DM2026.050.87
Nucleus-ImageType=DM2026.050.87
Qwen-ImageModel Category=Latent Text-to-Image Model2026.050.87
BAGELResolution=10242026.050.866
AlphaGRPOResolution=10242026.050.861
DecoModel Category=Pixel Text-to-Image Model2026.050.86
AlphaGRPO (RT2I)Resolution=512, Reasoning text-to-image generation task=RT2I2026.050.85
AlphaGRPOResolution=5122026.050.842
Seedream 3.0Type=DM2026.050.84
Seedream 3.0Model Category=Latent Text-to-Image Model2026.050.84
BAGELResolution=5122026.050.84
Visual-CoGGPU=–, RL Method=PPO2026.050.84
UniWorld-v1Params=7B+12B‡, LLM rewriter=true2026.050.84
HiDream-I1-FullType=DM2026.050.83
RAEv2Training Stage=Finetuning2026.050.827
Full CacheBudget=100%, Model=Infinity-8B2026.050.826
HeatKVBudget=20%, Model=Infinity-8B2026.050.824
ScaleKVBudget=20%, Model=Infinity-8B2026.050.821
HeatKVBudget=4%, Model=Infinity-8B2026.050.821
NTMType=NF, Training Setting=from scratch, Resolution=256 × 2562026.050.82
BAGEL2026.050.82
BAGEL2026.050.82
Z-Image-turboModel Category=Latent Text-to-Image Model2026.050.82
FLUX.1 devCategory=Generation Only Models2026.050.82
HACKBudget=20%, Model=Infinity-8B2026.050.82
ScaleKVBudget=4%, Model=Infinity-8B2026.050.82
ScaleKVBudget=10%, Model=Infinity-8B2026.050.819
StreamingLLMBudget=4%, Model=Infinity-8B2026.050.819
StreamingLLMBudget=10%, Model=Infinity-8B2026.050.816
HeatKVBudget=10%, Model=Infinity-8B2026.050.816
SnapKVBudget=20%, Model=Infinity-8B2026.050.815
RAETraining Stage=Finetuning2026.050.815
StreamingLLMBudget=20%, Model=Infinity-8B2026.050.814
HACKBudget=4%, Model=Infinity-8B2026.050.814
SnapKVBudget=4%, Model=Infinity-8B2026.050.814
HACKBudget=10%, Model=Infinity-8B2026.050.813
SnapKVBudget=10%, Model=Infinity-8B2026.050.813
BLIP3o-4B2026.050.81
Janus-Pro-7BType=DM2026.050.8
OmniGen22026.050.8
w/ RPO-Gemini 3.1 Pro-ARRBase Generator=FLUX.1-Dev2026.050.8
Janus-Pro-7B2026.050.8
OmniGen22026.050.8
ARR-RPOVariant=w/ RPO-Gemini 3.1 Pro-ARR2026.050.8
JanusProCategory=Unified Multimodal Models2026.050.8
Janus-ProParams=7B, LLM rewriter=false2026.050.8
PixelGenModel Category=Pixel Text-to-Image Model2026.050.79
SGT-Gen2Params=3B+4B‡, LLM rewriter=false2026.050.789
Flux-VAETraining Stage=Finetuning2026.050.783
w/ RPO-GPT-5-ARRBase Generator=FLUX.1-Dev2026.050.78
ARR-RPOVariant=w/ RPO-GPT-5-ARR2026.050.78
ARR-RPOVariant=w/ RPO-Gemini 3.1 Pro2026.050.77
OmniGen2Params=3B+4B‡, LLM rewriter=false2026.050.766
NTMType=NF, Training Setting=finetune, Resolution=512 × 5122026.050.76
Show-o22026.050.76
Show-o22026.050.76
ARR-RPOVariant=w/ RPO-GPT-52026.050.76
L2PModel Category=Pixel Text-to-Image Model2026.050.76
w/ RPO-Qwen3vl-8B-ARRBase Generator=FLUX.1-Dev2026.050.74
ARR-RPOVariant=w/ RPO-Qwen3vl-8B-ARR2026.050.74
PixelDiTModel Category=Pixel Text-to-Image Model2026.050.74
SD3 MediumCategory=Generation Only Models2026.050.74
PixNerdModel Category=Pixel Text-to-Image Model2026.050.73
HarmonParams=1.5B, LLM rewriter=false2026.050.73
ARR-RPOVariant=w/ RPO-Qwen3vl-8B2026.050.72
FKS + StitchVMModel=SD3.5-Large, Reward Model=CLIP Reward, N=42026.050.71
FKS + StitchVMModel=SD3.5-Large, Reward Model=HPSv2 Reward, N=42026.050.7
Show-oCategory=Unified Multimodal Models2026.050.69
DanceGRPOGPU=32, RL Method=GRPO2026.050.69
FKS + StitchVMModel=SD3.5-Medium, Reward Model=HPSv2 Reward, N=42026.050.69
HeatKVBudget=20%, Model=Infinity-2B2026.050.684
HeatKVBudget=10%, Model=Infinity-2B2026.050.684
HeatKVBudget=4%, Model=Infinity-2B2026.050.683
SnapKVBudget=4%, Model=Infinity-2B2026.050.682
Full CacheBudget=100%, Model=Infinity-2B2026.050.68
Show-o512Params=1.3B, LLM rewriter=false2026.050.68
BoNModel=SD3.5-Large, Reward Model=HPSv2 Reward, N=42026.050.68
FKSModel=SD3.5-Large, Reward Model=HPSv2 Reward, N=42026.050.68
FKS + StitchVMModel=FLUX, Reward Model=HPSv2 Reward, N=42026.050.68
FKS + StitchVMModel=SD3.5-Large, Reward Model=Aesthetic Reward, N=42026.050.68
FKS + StitchVMModel=SD3.5-Medium, Reward Model=CLIP Reward, N=42026.050.68
FKSModel=SD3.5-Large, Reward Model=CLIP Reward, N=42026.050.68
StreamingLLMBudget=20%, Model=Infinity-2B2026.050.679
ScaleKVBudget=20%, Model=Infinity-2B2026.050.678
HACKBudget=20%, Model=Infinity-2B2026.050.677
SnapKVBudget=20%, Model=Infinity-2B2026.050.677