GANs are behind pretty much every "deep fake" technology out there. There's some fairly impressive stuff like that generating feasible selfie-style photos of people that don't exist, which is mostly used for generating fake social media profiles. Overall though, like the guy said, any decent results require a highly constrained training set, and thus can only generate a highly constrained output set. People freak out about GANs like they are gonna make it impossible to tell what's real or not, but anyone who works with them can you that is very much not the case, and beyond several very spefic use cases, the generative results have artifacts that betray them as obviously fake.
GANs are behind pretty much every "deep fake" technology out there. There's some fairly impressive stuff like that generating feasible selfie-style photos of people that don't exist, which is mostly used for generating fake social media profiles. Overall though, like the guy said, any decent results require a highly constrained training set, and thus can only generate a highly constrained output set. People freak out about GANs like they are gonna make it impossible to tell what's real or not, but anyone who works with them can you that is very much not the case, and beyond several very spefic use cases, the generative results have artifacts that betray them as obviously fake.