Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

FLUX

Benchmarks

Task NameDataset NameSOTA ResultTrend
Text-to-Image GenerationFLUX.1 DrawBench dev 1.0 (test)
ImageReward1.0804
35
Single-task capability preservationFLUX.1 (dev)
DINO Score0.7443
32
Text-to-image generationFLUX.1-schnell 1.0 (dev)
Latency (s)1.15
23
Aesthetic Image GenerationFLUX
Aesthetic Score8.5394
22
Text-aligned Image GenerationFLUX
Pick-Score0.2439
22
Image Quality AssessmentFLUX-1 Preserved Concepts (non-target concepts)
LPIPS0.0362
22
Deepfake AttributionFLUX unseen
Accuracy57.92
20
KL Divergence EvaluationFLUX2-9B
KL Divergence170
15
Image GenerationFLUX (test)
ImgReward1.085
14
Text-to-Image GenerationFlux (dev)
HPSv2.1 Score31.47
14
Text-to-Image GenerationFLUX.1 dev (test)
ImageReward1.095
13
LoRA MergingFLUX.1 (dev)
Wall-clock Time (s)6.96
12
Text-to-image generationFLUX.1 80 held-out prompts (dev)
RMSE12.18
12
Fake Image DetectionFlux (dev)
Accuracy99.89
12
Identity CustomizationFLUX.1 (dev)
Similarity Score78.3
11
Text-to-Image GenerationFLUX
Training Memory (GiB)10.42
11
AI-generated image detectionFLUX.1 generator (test)
Accuracy83.57
11
Text-to-Image GenerationFLUX.1
Win Rate54
10
Text-to-Image GenerationFLUX.1 (dev)
PSNR30.37
8
Image GenerationFlux
SO0.98
8
Time Series ForecastingFlux (test)
RMSE (H=6)0.105
7
AIGI DetectionFLUX 2 (test)
Accuracy94.84
7
Nudity Jailbreaking TransferFLUX.1 (Universal nudity jailbreaking prompts)
Transfer Success Rate100
7
Image GenerationFLUX 3-mixed-resolution (512 + 1024 + 2048)
ImgR.98.3
6
Deepfake DetectionFLUX.2 (test)
Accuracy97.1
6
Showing 25 of 51 rows