Research map / Image generation / Text-to-image diffusion
VisiLock: Authorizing Instruction-based Image editing with Dual Score Distillation
Van Thanh Le, Yun Fu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| remove the key from the input (unauthorized use) | VisiLock · MagicBrush (all turns) | DINO: 0.726 -> 0.072 | Hurt |
| remove the key from the input | VisiLock · MagicBrush (all turns) | CLIP-I: 0.821 -> 0.481 | Hurt |
| use blur degraded teacher instead of fixed target (ablation) | VisiLock · MagicBrush | unauthorized DINO: 0.072 -> 0.326 (weaker lock) | Hurt |
| use noise degraded teacher instead of fixed target (ablation) | VisiLock · MagicBrush | unauthorized DINO: 0.072 -> 0.617 (weaker lock) | Hurt |
| use noise degraded teacher instead of fixed target (ablation) | VisiLock · MagicBrush | authorized DINO: 0.726 -> 0.784 | Improved |
| increase margin repulsion m from 0.5 to 1.0 (fixed target) (ablation) | VisiLock · MagicBrush | authorized CLIP-I: 0.801 -> 0.821 | Improved |
| increase margin repulsion m from 0.0 to 1.0 (fixed target) (ablation) | VisiLock · MagicBrush | unauthorized CLIP-I: 0.485 -> 0.481 | Improved |
| increase margin repulsion m for blur teacher (ablation) | VisiLock · MagicBrush | unauthorized DINO: 0.324 -> 0.326 | No clear effect |
| increase trigger size from 16px to 128px (ablation) | VisiLock · MagicBrush | delta CLIP-I: -0.320 -> -0.347 | No clear effect |
| initialize student from degraded teacher instead of original IP2P or random (ablation) | VisiLock · MagicBrush | lock strength (qualitative) | Improved |
| add contrastive FMLock loss instead of dual distillation | naive fine-tuning · InstructPix2Pix | training stability | Hurt |
| run 500-step adversarial self-distillation unlock attack | VisiLock · MagicBrush | unauthorized DINO | Mixed |
Closest papers by meaning
- Semantically Aligned Gradient-Driven Context-Preserving Image Editing (ECCV 2026 · similarity 0.89)
- Balancing Preservation and Modification: A Region and Semantic Aware Metric for Instruction-Based Image Editing (ICML 2025 · similarity 0.88)
- Multi-Reward as Condition for Instruction-based Image Editing (ICLR 2025 · similarity 0.88)
- InsightEdit: Towards Better Instruction Following for Image Editing (CVPR 2025 · similarity 0.87)
- EdiVal-Agent: An Object-Centric Framework for Automated, Fine-Grained Evaluation of Multi-Turn Editing (ICLR 2026 · similarity 0.87)
- When the Prompt Becomes Visual: Vision-Centric Jailbreak Attacks for Large Image Editing Models (ICML 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
