Training custom Ai generative models

Want to train your own Ai models to achieve a unique look or generate content that the public models aren't very good at? It may seem out of reach when you look at the big models trained on hundreds of millions of images with hundreds of GPUs for months. But there is good news! You can start with an existing pre-trained model and "fine tune" it on your own dataset. It takes a tiny fraction of the time and data but still yields fantastic results. Here are some notes and links to get you going. Please note this is definitely more involved than just using the Ai art notebooks. There are often missing pieces and gaps in the pipeline requiring some scripting or manual work. I've trained all but VQGAN and LDM now so hit me up (@pharmapsychotic) if you have issues or want to chat! 

Dataset tools



OpenAI Diffusion

ESRGAN (superresolution)