Broad-concept Removal on I2P
47.9Self-harm Removal RateSD
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| SD2026.05 | 47.9 | 44.2 | 37.5 | 40 | 60.2 | 59.5 | 46.3 | 48.9 | |
| Ablating2026.05 | 47.4 | 40.8 | 32.9 | 37.9 | 60.3 | 57.8 | 43.3 | 45.9 | |
| FMN2026.05 | 46.8 | 37.7 | 25 | 37 | 59.1 | 58.1 | 47.8 | 47.8 | |
| EraseDiff2026.05 | 40.6 | — | — | — | 49.8 | 49.4 | — | 44.9 | |
| SafeDiffusion-R1Anchor Config=Unsafe Anchor2026.05 | 36.83 | 30.74 | 39.56 | 40.44 | 27.18 | 26.17 | 32.01 | 33.43 | |
| ESD-x2026.05 | 36.8 | 34.1 | 30.2 | 28.9 | 40.2 | 45.2 | 40.5 | 36.6 | |
| SD v1.42026.03 | 35.5 | 21.2 | 19.7 | 19.4 | 54.5 | 42.1 | 40.1 | 35.6 | |
| ESD-u2026.05 | 33.7 | 26.8 | 24 | 26.7 | 35 | 40.1 | 35.1 | 32.8 | |
| Safe-DPO2026.05 | 33.33 | — | 22.59 | 30.3 | 20.7 | — | 32.43 | 19.82 | |
| UCE2026.05 | 30.8 | 36.4 | 29.5 | 29 | 25.5 | 41.1 | 34.1 | 31.3 | |
| SLD2026.05 | 30 | 22.5 | 22.1 | 22.1 | 52.4 | 40.5 | 31.8 | 33.7 | |
| CASTEER2026.05 | 26.22 | 29 | 25.61 | 17.61 | 20.73 | 34 | 27.78 | 25.58 | |
| Receler2026.05 | 24.8 | 28.6 | 21.7 | 21.3 | 29.4 | 34.8 | 27.1 | 27 | |
| SLDsource=TRCE [4]2026.03 | 19.2 | 41.1 | 20.1 | 19.4 | 22.9 | 16 | 19.7 | 35.6 | |
| SPM2026.05 | 15.88 | — | — | — | 52.5 | 69.1 | — | 54.6 | |
| SafeDiffusion-R12026.05 | 15.86 | 16.02 | 25.12 | 26 | 11.6 | 14.6 | 17.33 | 18.07 | |
| ESDsource=TRCE [4]2026.03 | 11.1 | 3.5 | 6.4 | 16.7 | 16.4 | 16.1 | 6.3 | 12.2 | |
| AdaVD2026.03 | 11.1 | 10.4 | 8.1 | 8.7 | 2.7 | 17.5 | 12.7 | 10 | |
| RECEsource=TRCE [4]2026.03 | 8.5 | 4.3 | 6.1 | 6.1 | 8.6 | 9.7 | 14.2 | 8.5 | |
| Safree2026.03 | 7.2 | 4.8 | 9.8 | 10.9 | 5.3 | 13.7 | 9.6 | 8.8 | |
| TRCE2026.03 | 5 | 7.8 | 6.2 | 3.4 | 1.7 | 8.2 | 6.2 | 5.7 | |
| Prototype-Guided Concept Erasure2026.03 | 3.8 | 3.8 | 6.1 | 5.5 | 1.7 | 10.1 | 5.8 | 5.2 |