SDXL
Benchmarks
| Task Name | Dataset Name | SOTA Result | Trend | |
|---|---|---|---|---|
| Image Copy Detection | SDXL (test) | Avg Similarity81.8 | 28 | |
| Text-to-Image Generation | SDXL 200 prompts (val) | ImageReward6.85 | 24 | |
| Style Erasure | SDXL Erasing Van Gogh | CS (Van Gogh)33.81 | 13 | |
| AI-generated image detection | SDXL generator v1.0 (test) | Accuracy66.79 | 11 | |
| Text-to-image generation | SDXL U-Net (test) | FID29.08 | 10 | |
| Text-to-Image Generation | SDXL | FID25.54 | 10 | |
| Prompt Refinement | SDXL unseen v1.0 (test) | ImageReward0.983 | 10 | |
| Super-Resolution | SDXL-Turbo | PSNR33.25 | 9 | |
| Fake image detection | SDXL v1.0 | AA95.17 | 9 | |
| Synthetic Image Detection | SDXL (open-source) | AUCROC0.814 | 9 | |
| Text-to-Image Generation | SDXL Base | CLIP Score0.2912 | 8 | |
| Diffusion Model Fine-tuning | SDXL | GPU Hours4 | 8 | |
| Watermark Extraction | SDXL | Clean Extraction Rate100 | 8 | |
| Nudity Jailbreaking Transfer | SDXL Universal nudity jailbreaking prompts | Transfer Success Rate100 | 7 | |
| Style-Driven Image Generation | SDXL Easy + Complex Averaged 1.0 (test) | Content Score0.329 | 7 | |
| Text-to-Image Generation | SDXL | BRISQUE19.55 | 6 | |
| Style Erasure | SDXL Erasing Van Gogh and Monet | CS (Van Gogh)33.71 | 6 | |
| LoRA Combination | SDXL v1.0 | CLIP Score (Style)61.59 | 6 | |
| Jailbreak Attack | SDXL | TASR (%)94.25 | 6 | |
| Tamper Localization | SDXL | F1 Score97 | 6 | |
| Image Generation | SDXL | FID9.12 | 6 | |
| Synthetic Image Detection | SDXL | Acc@EER73.7 | 6 | |
| Synthetic Image Detection | SDXL | AP79.9 | 6 | |
| Synthetic Image Detection | SDXL-DPO | AP87.4 | 6 | |
| Text-to-Image Backdoor Attack | SDXL TOA | ASR (%)100 | 5 |
Showing 25 of 34 rows