Multi-modal Toxicity Attack on RTP (test)
24.83Overall ScoreSTARE
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| STARETarget VLM=LLaVA, align=0.12026.05 | 24.83 | 13.58 | 10.54 | 22.48 | 7.85 | — | 9.47 | — | 16.06 | |
| STARETarget VLM=LLaVA, align=0.22026.05 | 24.64 | 13.97 | 10.17 | 22.71 | 7.56 | — | 9.92 | — | 15.78 | |
| STARE w/o AlignTarget VLM=LLaVA2026.05 | 24.58 | 12.72 | 8.93 | 21.86 | 7.68 | — | 9.97 | — | 14.94 | |
| STARETarget VLM=LLaVA, align=0.052026.05 | 24.15 | 12.84 | 10.26 | 22.39 | 7.92 | — | 9.23 | — | 15.47 | |
| STARE (0.2)Hyperparameter=0.22026.05 | 21.47 | 12.06 | 4.33 | 18.85 | 2.95 | 10.77 | 5.59 | — | — | |
| STARE w/o EditingTarget VLM=LLaVA2026.05 | 20.34 | 13.08 | 7.56 | 18.92 | 6.43 | — | 8.69 | — | 14.76 | |
| STARE (0.1)Hyperparameter=0.12026.05 | 20.25 | 10.39 | 4.62 | 17.91 | 2.08 | 9.38 | 5.23 | — | — | |
| STARE w/o LoRATarget VLM=LLaVA2026.05 | 18.76 | 12.93 | 6.12 | 16.45 | 5.28 | — | 8.84 | — | 12.68 | |
| STARE (0.05)Hyperparameter=0.052026.05 | 18.2 | 10.27 | 3.26 | 17.19 | 2.14 | 9.19 | 5.46 | — | — | |
| ARTTarget VLM=LLaVA2026.05 | 17.95 | 11.82 | 5.58 | 15.24 | 4.76 | — | 8.71 | — | 11.86 | |
| STARE (0.2)trained_on=SD 3.5-Med, evaluated_on=FLUX.1-dev, alignment_coefficient=0.22026.05 | 17.82 | 9.47 | 3.15 | 16.93 | 3.28 | 8.76 | 3.84 | 0.74 | — | |
| STARE w/ DDPOBaseline Tier=Tier-1, Optimization=DDPO2026.05 | 17.65 | 9.88 | 2.92 | 16.44 | 2.04 | 9.02 | 5.31 | — | — | |
| STARE (0.1)Hyperparameter=0.12026.05 | 15.96 | 8.18 | 3.96 | 14.46 | 4.13 | 8.21 | 4.14 | — | — | |
| STARE (0.2)Hyperparameter=0.22026.05 | 15.82 | 8.12 | 3.44 | 14.89 | 3.96 | 8.51 | 4.29 | — | — | |
| STARE (0.05)Hyperparameter=0.052026.05 | 14.77 | 8.48 | 2.69 | 14.15 | 3.63 | 8.12 | 3.21 | — | — | |
| STARE w/ DDPOBaseline Tier=Tier-1, Optimization=DDPO2026.05 | 14.12 | 7.94 | 2.45 | 13.62 | 3.11 | 8.09 | 3.15 | — | — | |
| DiffZOOTarget VLM=LLaVA2026.05 | 13.67 | 8.94 | 5.21 | 11.78 | 3.48 | — | 8.03 | — | 11.12 | |
| STARE (0.2)align=0.22026.05 | 13.64 | 6.95 | 3.37 | 12.34 | 3.52 | 6.99 | 3.53 | — | — | |
| ARTBaseline Tier=Tier-22026.05 | 13.27 | 7.19 | 1.71 | 12.08 | 2.32 | 9.14 | 5.3 | — | — | |
| STARE w/ DDPOtrained_on=SD 3.5-Med, evaluated_on=FLUX.1-dev, optimization=DDPO2026.05 | 13.27 | 7.13 | 2.08 | 12.44 | 2.74 | 7.35 | 2.91 | 0.71 | — | |
| ARTBaseline Tier=Tier-22026.05 | 12.81 | 5.55 | 1.63 | 12.37 | 2.22 | 8.07 | 4.14 | — | — | |
| DiffZOOBaseline Tier=Tier-22026.05 | 12.4 | 5.28 | 1.29 | 11.54 | 2.11 | 7.19 | 4.76 | — | — | |
| STARE w/ DDPOoptimization=DDPO2026.05 | 12.08 | 6.72 | 2.09 | 11.61 | 2.65 | 6.89 | 2.69 | — | — | |
| DiffZOOBaseline Tier=Tier-22026.05 | 11.66 | 6.12 | 1.35 | 10.97 | 1.61 | 6.42 | 3.12 | — | — | |
| ARTtrained_on=SD 3.5-Med, evaluated_on=FLUX.1-dev2026.05 | 11.43 | 5.82 | 1.24 | 10.67 | 1.89 | 6.14 | 3.58 | 0.69 | — | |
| PGJTarget VLM=LLaVA2026.05 | 11.25 | 7.08 | 3.56 | 9.92 | 2.93 | — | 6.28 | — | 7.65 | |
| PGJBaseline Tier=Tier-22026.05 | 10.15 | 6.77 | 1.33 | 9.33 | 1.24 | 6.02 | 4.28 | — | — | |
| PGJBaseline Tier=Tier-22026.05 | 9.42 | 4.88 | 1.05 | 8.41 | 0.94 | 5.66 | 2.75 | — | — | |
| Text + SDTarget VLM=LLaVA2026.05 | 8.42 | 5.85 | 2.14 | 7.23 | 1.87 | — | 5.45 | — | 5.92 | |
| Text + SDBaseline Tier=Tier-22026.05 | 7.29 | 3.53 | 0.25 | 6.56 | 0.14 | 5.22 | 2.67 | — | — | |
| Text + SDBaseline Tier=Tier-22026.05 | 5.63 | 3.07 | 0.19 | 5.12 | 0.11 | 4.44 | 1.92 | — | — | |
| Text-OnlyTarget VLM=LLaVA2026.05 | 5.3 | 3.2 | 1.5 | 3.9 | 2.4 | — | 3.3 | — | 3.5 | |
| Text + SDmodel=Stable Diffusion2026.05 | 4.82 | 2.54 | 0.16 | 4.37 | 0.09 | 3.81 | 1.64 | — | — | |
| Text-OnlyBaseline Tier=Tier-22026.05 | 3.5 | 0.8 | 0.2 | 3.2 | 0.6 | 2.5 | 0.4 | — | — | |
| Text-OnlyBaseline Tier=Tier-22026.05 | 2.2 | 0.5 | 0.1 | 2 | 0.3 | 2.1 | 0.3 | — | — | |
| Text-Only2026.05 | 1.8 | 0.4 | 0.1 | 1.6 | 0.2 | 1.7 | 0.2 | — | — |