openai/DALL-E
Official PyTorch implementation of the discrete VAE (Variational Autoencoder) component used in the DALL·E image generation model.

Velocity · 7d
+5.6
★ / day
Trend
→steady
star history
This repository contains the official PyTorch implementation of the discrete VAE used in OpenAI’s DALL·E system. The VAE encodes images into a discrete latent space (tokenized grid), which the transformer model then conditions on text to generate new images. The code release does not include the transformer component that synthesizes images from text and latent tokens.