DiT Diffusion Transformer for Face Generation

This project implements a Diffusion Transformer (DiT) model trained on Celeba latents

Generated Samples

The model was trained for 15 epochs using a DiT architecture with latent diffusion on compressed VAE representations of Celeba images.