Research map / Image generation / Text-to-image diffusion
Rickrolling the Artist: Injecting Backdoors into Text Encoders for Text-to-Image Synthesis
Lukas Struppek, Dominik Hintersdorf, Kristian Kersting
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 backdoor loss with sufficient weight (beta 0.1 vs 0.0) (ablation) | TPA text encoder backdoor · Stable Diffusion v1.4, single TPA | z-score: 0.10 -> 101.94 | Improved |
| add the utility loss (ablation) | TPA text encoder backdoor · FID / ImageNet-V2 top-1 | FID: 155.49 -> 17.54 | Improved |
| raise beta from 0.01 to 0.1 (ablation) | TPA text encoder backdoor · Stable Diffusion v1.4, single TPA | z-score: 81.07 -> 101.94 | Improved |
| increase beta to 10 (ablation) | TPA text encoder backdoor · ImageNet-V2 zero-shot | top-1 accuracy: 69.24% -> 8.75% | Hurt |
| increase beta to 1 (ablation) | TPA text encoder backdoor · MS-COCO FID | FID: 17.54 -> 16.85 | Improved |
| increase number of poisoned training samples (ablation) | TPA · Stable Diffusion v1.4 | z-score | Improved |
| increase number of poisoned samples (ablation) | TPA · Stable Diffusion v1.4 | FID / Sim_target | No clear effect |
| inject more TPA backdoors (up to 32) (ablation) | TPA · Stable Diffusion v1.4 | z-score / Sim_clean | Hurt |
| inject more TAA backdoors (ablation) | TAA · Stable Diffusion v1.4 | evaluation metrics | No clear effect |
| replace cosine distance with MAE (ablation) | TPA · Stable Diffusion v1.4 | z-score: 101.94 -> 91.55 | Hurt |
| replace cosine distance with MSE or Poincare (ablation) | TPA · Stable Diffusion v1.4 | z-score: 101.94 -> 101.63 / 100.88 | No clear effect |
| train on MS-COCO 2014 prompts instead of LAION-Aesthetics (ablation) | TPA · Stable Diffusion v1.4 | z-score: 101.94 -> 101.37 | No clear effect |
Closest papers by meaning
- Personalization as a Shortcut for Few-Shot Backdoor Attack against Text-to-Image Diffusion Models (AAAI 2024 · similarity 0.90)
- Token by Token, Compromised: Backdoor Vulnerabilities in Unified Autoregressive Models (NeurIPS 2026 · similarity 0.88)
- ART: Automatic Red-teaming for Text-to-Image Models to Protect Benign Users (NeurIPS 2024 · similarity 0.88)
- BackdoorLLM: A Comprehensive Benchmark for Backdoor Attacks and Defenses on Large Language Models (NeurIPS 2025 · similarity 0.87)
- ColJailBreak: Collaborative Generation and Editing for Jailbreaking Text-to-Image Deep Generation (NeurIPS 2024 · similarity 0.87)
- Distraction is All You Need: Memory-Efficient Image Immunization against Diffusion-Based Image Editing (CVPR 2024 · 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.
