Research map / Adversarial robustness and security / Attacks on text and graphs
Reconstructing Template-Memorized Images from Natural Prompts
Sol Yarkoni, Mahmood Sharif, Roi Livni
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 |
|---|---|---|---|
| use short benign collocation prompts instead of training-set captions | extraction attack · SD 1.4 | number of prompts needed: 350{,}000 -> 108 | Improved |
| apply BE-PRSS prompt-variation mitigation | SD 1.4 with template collocations · 43 memorizing collocations | collocations still producing template image: 28 -> 22 | No clear effect |
| use BE-PRSS detection threshold 1.0 on template memorization | BE-PRSS detector · template-memorized prompts | accuracy: 0.595 | Hurt |
| use attack against newer models (SD 3.5) | template attack · Stable Diffusion 3.5 Medium | template reconstruction | Hurt |
| use collocations from SD 1.4 on real-user prompts (Midjourney) | template attack · Midjourney prompts | prompts producing template image: 8 out of 20 | Improved |
| use collocations from SD 1.4 on real-user prompts (DiffusionDB) | template attack · DiffusionDB prompts | prompts producing template image: 7 out of 15 | Improved |
Closest papers by meaning
- Detecting, Explaining, and Mitigating Memorization in Diffusion Models (ICLR 2024 · similarity 0.86)
- Memories of Forgotten Concepts (CVPR 2025 · similarity 0.86)
- ConceptPrune: Concept Editing in Diffusion Models via Skilled Neuron Pruning (ICLR 2025 · 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)
- Prompting Hard or Hardly Prompting: Prompt Inversion for Text-to-Image Diffusion Models (CVPR 2024 · similarity 0.85)
- Enhancing Privacy-Utility Trade-offs to Mitigate Memorization in Diffusion Models (CVPR 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.
