The watermarking should be on those things we want to verify as something that was not generated or manipulated. Something you'd add to, for instance, cameras. Putting them on the generated/manipulated is backwards as you can never get every model to watermark.
That model is equally bad though. Given that you're writing this in a discussion about gen AI watermarks, how in the world did you come up with the idea that Gen AI wouldn't be able to add a watermark?
Not that they "wouldn't be able".. that they wouldn't do it. For Gen AI watermarks to be useful all Gen AI systems need to add them and the incentives aren't there for that to happen. On the other hand the incentives are there for the non-generated sources to add it so they can differentiate themselves from the Gen AI media.
Perhaps not watermarks, but cameras could sign their pictures and put it in the extra data, that's not something that would be easy to add to fake pictures, or at least not the correct signature.
