Overview

This is a custom image generation model build from scratch using Pytorch. The architecture is based of the foundational paper behind Stable Diffusion which can be found here.

Architecture Planning

1. Variational Autoencoder (VAE)
   • Encoder
   • Decoder
2. Diffusion UNet
   • FiLM (optional)
   • Time Embeddings
3. CLIP Text Encoder

Training Planning

1. Train using DDPM