Research map / Image generation / Fast diffusion sampling
VillanDiffusion: A Unified Backdoor Attack Framework for Diffusion Models
Sheng-Yen Chou, Pin-Yu Chen, Tsung-Yi Ho
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 |
|---|---|---|---|
| backdoor with VillanDiffusion at 90% poison rate versus clean model (ablation) | DDPM with 7 samplers · CIFAR10, Hat trigger | FID: about 10\% | Hurt |
| backdoor at poison rate below 70% (ablation) | DDPM with 7 samplers · CIFAR10, Hat trigger | FID: 70\% | Improved |
| increase poison rate to about 10% (ablation) | DDPM backdoor · CIFAR10, Hat trigger | MSE to target: about 10\% | Improved |
| backdoor with 20% poison rate | DDPM · CelebA-HQ | FID: 25\% $\sim$ 85\% | Hurt |
| use DDIM sampler randomness eta=1 instead of eta=0 with BadDiffusion (ablation) | BadDiffusion · CIFAR10 | backdoor success (MSE) | Mixed |
| replace BadDiffusion objective with VillanDiffusion on ODE samplers (ablation) | backdoor loss · CIFAR10 ODE samplers | backdoor success | Improved |
| apply inference-time clipping range [-1, 1] to all samplers (ablation) | backdoored DMs · CIFAR10 | FID and MSE | No clear effect |
| higher caption similarity between trigger and clean caption (ablation) | caption-trigger backdoor on Stable Diffusion · CelebA-HQ-Dialog and Pokemon Caption | MSE and MSE threshold | Hurt |
Later papers that compared with this method
2 later accepted papers report VillanDiffusion as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- STEDiff: Revealing the Spatial and Temporal Redundancy of Backdoor Attacks in Text-to-Image Diffusion Models (ICLR 2026 · better on 6 of 6 shared results)
- Watch Your Step: Information Injection in Diffusion Models via Shadow Timestep Embedding (ICML 2026 · better on 4 of 4 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
- Elijah: Eliminating Backdoors Injected in Diffusion Models via Distribution Shift (AAAI 2024 · similarity 0.92)
- TERD: A Unified Framework for Safeguarding Diffusion Models Against Backdoors (ICML 2024 · similarity 0.90)
- UIBDiffusion: Universal Imperceptible Backdoor Attack for Diffusion Models (CVPR 2025 · similarity 0.90)
- UFID: A Unified Framework for Black-box Input-level Backdoor Detection on Diffusion Models (AAAI 2025 · similarity 0.89)
- BackdoorDM: A Comprehensive Benchmark for Backdoor Learning on Diffusion Model (NeurIPS 2025 · similarity 0.88)
- The Stronger the Diffusion Model, the Easier the Backdoor: Data Poisoning to Induce Copyright BreachesWithout Adjusting Finetuning Pipeline (ICML 2024 · similarity 0.88)
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.
