Safe Text-to-Image Synthesis on FLUX.1 12B (dev)
19Inference Time (s)FLUX.1-dev
Evaluation results
| Method | Links | ||
|---|---|---|---|
| FLUX.1-devResolution=1024x1024, Inference steps=28, Hardware=3×RTX 4090 GPUs, Backbone=FLUX.1-dev (12B)2026.06 | 19 | 35,328.23 | |
| DESResolution=1024x1024, Inference steps=28, Hardware=3×RTX 4090 GPUs, Method Type=Fine-tuning-based2026.06 | 19 | 35,328.23 | |
| ESDResolution=1024x1024, Inference steps=28, Hardware=3×RTX 4090 GPUs, Method Type=Fine-tuning-based2026.06 | 19 | 35,331.79 | |
| EraseAnythingResolution=1024x1024, Inference steps=28, Hardware=3×RTX 4090 GPUs, Method Type=Fine-tuning-based2026.06 | 19 | 35,331.79 | |
| UCEResolution=1024x1024, Inference steps=28, Hardware=3×RTX 4090 GPUs2026.06 | 19 | 35,328.23 | |
| UVRResolution=1024x1024, Inference steps=28, Hardware=3×RTX 4090 GPUs, Method Type=Inference-time, Training-free=true2026.06 | 25 | 35,331.35 | |
| SLDResolution=1024x1024, Inference steps=28, Hardware=3×RTX 4090 GPUs, Method Type=Inference-time2026.06 | 31 | 35,337.86 |