DALL-E usage with DAGsHub Storage

This notebook demonstrates how to use OpenAI's DALL-E project, automatically downloading the models from DAGsHub storage for faster access.

Dali meet WALL-E meet world

Recently OpenAI released a public PyTorch package for the discrete VAE used for DALL-E. To make it easier to work with the model this DAGsHub repository enables you to use this package to easily access, iterate and version control your experiments with integration to DVC and Git

Blog: https://openai.com/blog/dall-e/Code: https://github.com/openai/DALL-E

DALL-E is a generative transformer language model that can accept both text and images as input to generate images from text. Trained on a 12-billion parameter version of GPT-3 it can receive both text and image as a single stream containing up to 1280 tokens —256 for the text and 1024 for the image.

In general some of the capabilities of DALL-E include;

• Create plausible images with a sentence
• Support for varied sentence structures
• Modify attributes of an existing image. For example input a illustration and convert the same image to a sketch.
• Change attributes like color, or image composition like adding a hat to an image of a dog.
• Generate images of multiple objects. For example, a stack of blocks, with a penguin on top wearing a red hat.
• Play with perspective and 3D style
• Work with internal and external structure, for example the cross section of an egg.

DALL-E offers a green field with multiple practical and commercial applications ranging from design, architecture, illustration, eCommerce to fashion. We hope by offering easier access to this model we can inspire the community to develop interesting use cases.

Image from the OpenAI blog on DALL-E