Text-to-Image Generation on NegGenBench (test)
100Positive ScoreNAG
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| NAGVariation=Standard2025.08 | 100 | 14.7 | 81.2 | |
| CFGSampling Steps=282025.08 | 100 | 30 | 95.6 | |
| NAG2025.08 | 99.3 | 22 | 96.8 | |
| Qwen-Image NegationModel access=Open-weight, Runtime=110s2025.08 | 99 | 10 | 93.7 | |
| NoneDescription=Baseline without negative prompt handling2025.08 | 99 | 19.5 | 96.8 | |
| Nano BananaModel access=Closed-weight, Runtime=14s2025.08 | 98.5 | 49.8 | 98 | |
| Nano Banana2025.08 | 98.5 | 40.6 | 81.7 | |
| VSF QualityModel access=Open-weight, Runtime=3s2025.08 | 98 | 42 | 98.6 | |
| VSFVariation=Quality2025.08 | 98 | 33 | 81.4 | |
| VSFHyperparameter Tuning=Quality2025.08 | 98 | 42 | 98.6 | |
| GPT-4oModel access=Closed-weight, Runtime=47s2025.08 | 97.8 | 70.5 | 95.4 | |
| GPT-4o2025.08 | 97.8 | 61.9 | 81.2 | |
| NAGHyperparameter Tuning=Strong2025.08 | 97.5 | 32 | 90.1 | |
| Qwen-Image NPModel access=Open-weight, Runtime=110s2025.08 | 97.3 | 19 | 93.5 | |
| NASA2025.08 | 97 | 38 | 86.7 | |
| NASA2025.08 | 95 | 22.4 | 72.7 | |
| NAGVariation=Strong2025.08 | 95 | 16.8 | 79.5 | |
| Janus-4oModel access=Open-weight, Runtime=20s2025.08 | 92.5 | 22.5 | 94.4 | |
| Generate+EditModel access=Open-weight, Runtime=55s2025.08 | 87.5 | 48.8 | 95.8 | |
| VSF StrongModel access=Open-weight, Runtime=3s2025.08 | 87 | 54.5 | 95.2 | |
| VSFVariation=Strong2025.08 | 87 | 41.5 | 81.4 | |
| VSFHyperparameter Tuning=Strong2025.08 | 87 | 54.5 | 95.2 |