GAN Image Generation
Comments speculate on and compare a new image generation or style transfer technique to GANs, StyleGAN, Stable Diffusion, GauGAN, and related methods, discussing training schemes, novelty, and potential overfitting.
Activity Over Time
Top Contributors
Keywords
Sample Comments
Could this be used to generate unique images from a training set?
What happens I someone trains his image generation models with AI generated images? :-D
Sounds like GauGAN.[1][2][1]: http://nvidia-research-mingyuliu.com/gaugan/[2]: https://github.com/NVlabs/SPADE
Is this done with a GAN or manual?
An AI that recognizes AI generated images would be a great tool for training an AI that produces images that won't be recognized as AI generated images via GAN, no?
Looks yawn. Just feed all the textures into a GAN already.
It's GANs all the way down.
It's a GAN with an unorthodox (and rather slow) training scheme!
Interesting how pretty much all the example images look like renders/paintings as opposed to photographs. Maybe that's what it's trained on?
What is the point of this? It's hard to see how this is useful. Maybe it's just an exercise to show what a diffusion model can do?