Research map / Image generation / Text-to-image diffusion
DeLeaker: Dynamic Inference-Time Reweighting For Semantic Leakage Mitigation in Text-to-Image Models
Mor Ventura, Michael Toker, Or Patashnik, Yonatan Belinkov, Roi Reichart
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 |
|---|---|---|---|
| apply DeLeaker attention reweighting vs entity description prompt | FLUX.1-Dev · SLIM pair subsets | major mitigation rate (%): 35.60% -> 46.07% | Improved |
| add entity descriptions to DeLeaker | DeLeaker · SLIM pair subsets | major mitigation rate (%): 46.07% -> 53.57% | No clear effect |
| add self-identity (image-text) strengthening (ablation) | DeLeaker · SLIM pair subsets | mitigation ratio vs full: 1.00 -> 0.54 when removed | Improved |
| add cross-entity image-text suppression (ablation) | DeLeaker · SLIM pair subsets | improvement (major+minor): 29% reduction when omitted | Improved |
| add text-text suppression (ablation) | DeLeaker · SLIM pair subsets | mitigation: 9% to 20% degradation when removed | Improved |
| weaken image-image interactions (ablation) | DeLeaker · SLIM pair subsets | mitigation | No clear effect |
| use layout-based methods (3DIS, RPF, RAG-Diffusion) vs DeLeaker | FLUX.1-Dev · SLIM pair subsets | degradation rate (%): 12.98% (DeLeaker) vs 64.91% (RAG-Diffusion) major degradation | Hurt |
| use CLIP/BLIP scores as proxy for leakage labels (ablation) | evaluation · SLIM pair subset (N=821) | Spearman rho: rho -0.027 to 0.027 | No clear effect |
| increase prompt complexity (pairs to interactions + style) | DeLeaker · SLIM subsets | total mitigation rate (%): 42.4 -> 66.4 | Improved |
Closest papers by meaning
- Semantic Surgery: Zero-Shot Concept Erasure in Diffusion Models (NeurIPS 2025 · similarity 0.87)
- Enhancing Multi-Image Understanding through Delimiter Token Scaling (ICLR 2026 · similarity 0.86)
- MemLeak: Diagnosing Information Leaks in Multimodal Agent Memory (NeurIPS 2026 · similarity 0.86)
- Fine-Grained Erasure in Text-to-Image Diffusion-based Foundation Models (CVPR 2025 · similarity 0.85)
- MACE: Mass Concept Erasure in Diffusion Models (CVPR 2024 · similarity 0.85)
- Semantic Routing: Exploring Multi-Layer LLM Feature Weighting for Diffusion Transformers (NeurIPS 2026 · similarity 0.85)
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.
