If the synthetic images are not classified as forgeries, does that mean they're essentially generating model-specific fuzzed versions of the originals? So rathe

If the synthetic images are not classified as forgeries, does that mean they're essentially generating model-specific fuzzed versions of the originals? So rather than just adding noise to the pixels, it's able to pentrate the noise deeper? That sounds generally useful with deep models.