‘Protected’ Images Are Easier, Not More Difficult, to Steal With AI
SMRTR summary
Recent research reveals that watermarking tools designed to prevent AI image edits may have the opposite effect. Instead of blocking changes, some protections actually help AI models like Stable Diffusion follow editing prompts more accurately, making unwanted manipulations easier. The study tested various protection methods on natural and artistic images, finding that added perturbations often increased image-text alignment scores, allowing for more precise edits.
SMRTR provides this summary for quick context. The original article belongs to Unite AI.
Read the original article