Quan Wang, Zeyu Dong, Tong Qiao · ACM International Conference on Multimedia (ACM MM) 2026 · 2026
DOI: 10.1145/3767308.3835684
Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).
In recent years, Diffusion Models (DMs) have achieved remarkable success in style transfer and AI-driven artistic creation. However, they have also exacerbated the risk of unauthorized style imitation of original artworks. Existing protection methods, such as Mist and Glaze, mitigate infringement risks by embedding adversarial perturbations into images to disrupt the fine-tuning process, especially for prevalent techniques like DreamBooth and LoRA. Nevertheless, these denfenses exhibit clear limitations when facing training-free diffusion-based style transfer methods, such as StyleID. To address this challenge, we propose Anti-Training-Free Style Transfer in Diffusion Models(AFSD), an active defense framework tailored for training-free style transfer. Specifically, AFSD introduces adaptive perturbations in high-frequency regions and imposes constraints at both the latent representation and self-attention feature levels, thereby effectively disrupting the propagation of style information while preserving perceptual quality. Extensive experiments demonstrate that AFSD significantly degrades the performance of training-free diffusion-based style transfer methods. Our work provides an effective proactive defense strategy for protecting the copyright of artistic works.
ACM International Conference on Multimedia (ACM MM) 2026 · 0 citations
ACM International Conference on Multimedia (ACM MM) 2026 · 0 citations
ACM International Conference on Multimedia (ACM MM) 2026 · 0 citations
No comments yet — start the discussion below.