A tutorial explaining how to train and generate high-quality anime faces with StyleGAN 1+2 neural networks, and tips/scripts for effective StyleGAN use.
Discussion of how to modify existing images with GANs. There are several possibilities: train another NN to turn an image back into the original encoding; run blackbox search on encodings, repeatedly tweaking it to approximate a target face; or the whitebox approach, directly backpropagating through the model from the image to the encoding while holding the model fixed. All of these have been implemented for StyleGAN, and a combination works best. There are even GUIs for editing StyleGAN anime, pony, & furry faces!
- Examples
- Background
- FAQ
- Training Requirements
- Data Preparation
- Training
- Sampling
- Models
- Transfer Learning
- Reversing StyleGAN To Control & Modify Images
- StyleGAN 2
- Future Work
- BigGAN
- See Also
- External Links
- Appendix