Toward Effective Protection against Diffusion-Based Mimicry through Score Distillation

We identify the encoder as a key vulnerability in latent diffusion models and use score distillation to craft protective perturbations with roughly twice the speed and half the memory. A complementary strategy minimizes semantic loss to produce more natural perturbations and improve protection against diffusion-based image mimicry.

Recommended citation: Xue, H., Liang, C.*, Wu, X.*, and Chen, Y. (2024). "Toward Effective Protection against Diffusion-Based Mimicry through Score Distillation." ICLR 2024.
Download Paper