We'll definitely need open models here, because proprietary AI services can (and I imagine will) be set to detect invisible watermarks in the ads and refuse to filter them out.
(It's almost obvious extension of current AI image fingerprinting efforts. If a robust invisible watermark can be used to tell the image was AI generated, the same scheme can also be used for general DRM.)
(It's almost obvious extension of current AI image fingerprinting efforts. If a robust invisible watermark can be used to tell the image was AI generated, the same scheme can also be used for general DRM.)