Research map / Image generation / Autoregressive and latent generation
Replication in Visual Diffusion Models: A Survey and Outlook
Wenhao Wang, Yifan Sun, Zongxin Yang, Zhengdong Hu, Zhentao Tan, Yi Yang
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 classifier-free guidance scale from 1.0 to 15.0 (ablation) | Stable Diffusion v1.5 · CFG sweep | duplication rate at tau=0.9: 2.2% -> 85.6% | Hurt |
| raise classifier-free guidance scale from 1.0 to 15.0 (ablation) | Stable Diffusion v1.5 · CFG sweep | CLIP text-image score: 0.274 -> 0.327 | Improved |
| raise CFG scale from 10.0 to 15.0 (ablation) | Stable Diffusion v1.5 · CFG sweep | CLIP text-image score: 0.327 -> 0.327 | No clear effect |
| name an artist in the prompt instead of a generic prompt | Stable Diffusion v1.5 · Intra-artist style consistency | Gram: 0.581 -> 0.734 | Improved |
| name an artist in the prompt instead of a generic prompt | Stable Diffusion v1.5 · Intra-artist style consistency | CLIP: 0.688 -> 0.799 | Improved |
| apply semantic deduplication (SemDeDup) to training data | Stable Diffusion · Mitigation comparison | replication rate: 1.9% -> 0.8% | Improved |
| train with DP-SGD at eps=1 instead of no mitigation | Stable Diffusion · Mitigation comparison | FID: 12.4 -> 28.7 | Hurt |
| tighten DP budget from eps=10 to eps=1 | DP-SGD · Mitigation comparison | replication rate / FID: 0.4% -> 0.1%; FID 18.2 -> 28.7 | Mixed |
| apply ESD concept unlearning | Stable Diffusion · Mitigation comparison | replication rate: 1.9% -> 0.3% | Improved |
| use SalUn instead of ESD | unlearning · Mitigation comparison | replication rate: 0.3% -> 0.2% | Improved |
Closest papers by meaning
- Image Copy Detection for Diffusion Models (NeurIPS 2024 · similarity 0.90)
- CommonCanvas: Open Diffusion Models Trained on Creative-Commons Images (CVPR 2024 · similarity 0.87)
- UnlearnCanvas: Stylized Image Dataset for Enhanced Machine Unlearning Evaluation in Diffusion Models (NeurIPS 2024 · similarity 0.86)
- Disguised Copyright Infringement of Latent Diffusion Models (ICML 2024 · similarity 0.86)
- The Emergence of Reproducibility and Consistency in Diffusion Models (ICML 2024 · similarity 0.86)
- Anti-Diffusion: Preventing Abuse of Modifications of Diffusion-Based Models (AAAI 2025 · similarity 0.86)
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.
