Researchers develop a tool to freeze videos generated from images, blocking deepfake creation at the source

Computer scientists have created a method to add invisible changes to images that force AI video generators to produce static or nearly motionless videos instead of realistic ones. This prevents bad actors from using a single image to create convincing deepfake videos for fraud, harassment, or disinformation.

Until now, defenses against AI-generated video synthesis have been weak or incomplete. This work proposes a concrete technical approach: deliberately poison images so that when fed into video generation models, they jam up the motion-synthesis process itself. The practical effect is straightforward: an image protected this way becomes useless for malicious video creation, even if someone with the right tools finds it online. The open question is whether this holds up against determined attackers who might develop counter-techniques, and whether real image platforms will actually deploy protections like this before synthetic video forgery becomes as routine as deepfake photos.