Watermarks vs. Perturbations for Preventing AI-based Style Editing

Qiuyu Tang, Aparna Bharati

Track: tiny / short paper (3-5 pages)

Keywords: Diffusion, watermark, asset protection, style editing

TL;DR: We assess the feasibility of watermarks and perturbations to prevent style editing by evaluating their visual impact, perceptual effects, and task alignment.

Abstract: The remarkable image editing capabilities of generative models have led to growing concerns regarding unauthorized editing of multimedia. To mitigate against such misuse, artists and creators can utilize traditional image watermarking and more recent adversarial perturbation-based protection techniques to protect media assets. Watermarks generally protect the origin by establishing ownership, but can be easily removed. However, perturbation-based protection is aimed at disrupting editing and is harder to remove. In this paper, we evaluate the effectiveness of the two methods against Stable Diffusion in preventing the generation of usable edits.

Presenter: ~Qiuyu_Tang1

Format: Maybe: the presenting author will attend in person, contingent on other factors that still need to be determined (e.g., visa, funding).

Funding: No, the presenting author of this submission does *not* fall under ICLR’s funding aims, or has sufficient alternate funding.

Anonymization: This submission has been anonymized for double-blind review via the removal of identifying information such as names, affiliations, and identifying URLs.

Submission Number: 38