Research map / Image generation / Text-to-image diffusion
StyleKeeper: Prevent Content Leakage using Negative Visual Query Guidance
Jaeseok Jeong, Junho Kim, Gayoung Lee, Yunjey Choi, Youngjung Uh
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 |
|---|---|---|---|
| add CFG with swapping self-attention (ablation) | StyleKeeper · T2I visual style prompting | text alignment / image quality (qualitative) | Improved |
| add negative visual query guidance (NVQG) (ablation) | StyleKeeper · T2I visual style prompting | content leakage (qualitative) | Improved |
| swap self-attention only in upblocks instead of all blocks (ablation) | StyleKeeper · T2I visual style prompting | content leakage (qualitative) | Improved |
| swap self-attention in bottleneck blocks (ablation) | StyleKeeper · T2I visual style prompting | content leakage (qualitative) | Hurt |
| start swapping in late upblock layers (24th layer onward) instead of early (ablation) | StyleKeeper · T2I visual style prompting | style / leakage trade-off (qualitative) | Improved |
| use stochastic encoding instead of DDIM inversion for real references (ablation) | StyleKeeper · real image reference | artifacts / style similarity (qualitative) | Improved |
| add color calibration (ablation) | StyleKeeper · real image reference | style similarity and text alignment (qualitative) | Improved |
| StyleKeeper vs StyleAligned | style similarity · Gram-matrix style similarity | style similarity (Gram): 0.759 -> 0.791 | Improved |
| StyleKeeper vs IP-Adapter | style similarity · Gram-matrix style similarity | style similarity (Gram): 0.768 -> 0.791 | Improved |
| StyleKeeper vs IP-Adapter | user preference · User study | vote share: 18.47% -> 58.15% | Improved |
| StyleKeeper vs StyleAligned | user preference · User study | vote share: 13.15% -> 58.15% | Improved |
| apply stochastic encoding and color calibration to other self-attention variants (StyleAligned, MasaCtrl) | existing methods · real image reference | artifacts / style (qualitative) | Improved |
Closest papers by meaning
- Less is More: Masking Elements in Image Condition Features Avoids Content Leakages in Style Transfer Diffusion Models (ICLR 2025 · similarity 0.88)
- CLeaR: A Unified Framework for Resolving the Leakage–Degradation Dilemma in Style Transfer (NeurIPS 2026 · similarity 0.86)
- Devil is in the Detail: Towards Injecting Fine Details of Image Prompt in Image Generation via Conflict-free Guidance and Stratified Attention (CVPR 2025 · similarity 0.86)
- StyleSSP: Sampling StartPoint Enhancement for Training-free Diffusion-based Method for Style Transfer (CVPR 2025 · similarity 0.86)
- StyleGuard: Preventing Text-to-Image-Model-based Style Mimicry Attacks by Style Perturbations (NeurIPS 2025 · similarity 0.86)
- SigStyle: Signature Style Transfer via Personalized Text-to-Image Models (AAAI 2025 · 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.
