> I get that deepfakes are superimposing a non-consenting face on a (presumably) consenting body, but how does that matter? Is it that "visually lying" is a leg

> I get that deepfakes are superimposing a non-consenting face on a (presumably) consenting body, but how does that matter? Is it that "visually lying" is a legal loophole? That's not how deepfakes work, since nothing is superimposed. Models like Stable Diffusion build a training set out of several images of a face (e.g. with something like LORA), and reduce it to a set of algorithms. That algorithm is a sophisticated way of expressing "green shirt, black hair, large chin, etc" It's not actually storing photos of faces. It then unmasks a new face out noise using those blueprints. You can change the blueprints, for example, I could ask SD to give me an image with gigantic rabbit ears. To a layperson it might look the same since the technology is so new, but it's not in any way analogous to a photograph, it's analogous to a painter who saw someone, wrote down some notes on a notepad, went to their studio and painted them. The painting might look realistic but it's not the same as the original photo.