Model providers can build image protection into diffusion models, but open-weight deployment changes the picture. A new preprint argues that user-side defenses are not enough and that protection must be designed with the constraints of open-weight releases in mind.

The paper proposes four requirements for such protection: protection effectiveness, clean fidelity, compatibility, and tamper resistance. Together these span whether the protection actually works, whether it preserves the model's normal output, whether it fits into existing deployment scenarios, and whether it can withstand attempts to bypass it.

The abstract is truncated, so the authors' concrete approach for meeting these requirements is not described in the available text. The contribution, as presented, is the requirement framework itself.