Research map / Image generation / Autoregressive and latent generation
QRShield: Exploiting Vulnerabilities of Latent Diffusion Models for Preventing AI Art Plagiarism
Xunyue Mo, Weibin Wu, Qingrui Tu, Hang Wang, Junxi He, Zibin Zheng
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 |
|---|---|---|---|
| raise anchor frequency from none to high (ablation) | QRShield · Full U-Net fine-tuning | FID: 153.11 -> 306.64 | Improved |
| use 2x2 periodic anchor tiling instead of 1x1 (ablation) | QRShield · Full U-Net fine-tuning | FID: 178.52 -> 306.64 | Improved |
| use finer tiling (4x4, 8x8) instead of 2x2 (ablation) | QRShield · Full U-Net fine-tuning | FID: 306.64 -> 226.90 (4x4), 188.82 (8x8) | Hurt |
| share one anchor across all VAE latent channels (ablation) | QRShield · Full U-Net fine-tuning | FID: 274.83 -> 306.64 | Improved |
| QRShield vs Mist on full U-Net fine-tuning | QRShield vs Mist · Full U-Net fine-tuning | FID: 186.09 -> 306.64 | Improved |
| QRShield vs Mist on LoRA with DreamBooth | QRShield vs Mist · LoRA with DreamBooth | FID: 327.94 -> 382.69 | Improved |
| QRShield vs best baseline on Textual Inversion | QRShield vs Sim-AC · Textual Inversion | FID: 269.48 -> 306.49 | Improved |
| QRShield vs Mist on protection compute | QRShield vs Mist · Protection efficiency (RTX 4090) | time and VRAM: ~58s -> ~11s; ~16G -> ~3G | Improved |
| transfer adversarial examples to other Stable Diffusion versions | QRShield · Cross-model transfer (full U-Net) | FID: 306.64 -> 286.06 (SD1.5), 262.55 (SD1.4) | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Anti-Diffusion — Anti-Diffusion: Preventing Abuse of Modifications of Diffusion-Based Models (AAAI 2025) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Adversarial Perturbations Cannot Reliably Protect Artists From Generative AI (ICLR 2025 · similarity 0.89)
- DiffusionGuard: A Robust Defense Against Malicious Diffusion-based Image Editing (ICLR 2025 · similarity 0.88)
- Toward effective protection against diffusion-based mimicry through score distillation (ICLR 2024 · similarity 0.87)
- Towards Black-Box Membership Inference Attack for Diffusion Models (ICML 2025 · similarity 0.87)
- AdvPaint: Protecting Images from Inpainting Manipulation via Adversarial Attention Disruption (ICLR 2025 · similarity 0.87)
- Nearly Zero-Cost Protection Against Mimicry by Personalized Diffusion Models (CVPR 2025 · similarity 0.87)
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.
