Fine-tuning the GLIDE model for ACG artwork generation

This is my first project in 2023. I try to use the same training technique introduced in a 2022 paper “Pretraining is all you need” for generating ACG artwork. It is tough because I need to preprocess the dataset by myself. I’m not sure if I can handle 200k images. I’m going to train another model to help me preprocess the large dataset. I hope a UNet can save me. However, the current process of my model using a smaller dataset is working well. I will update the result with the original dataset soon.

These are some good results I have: