min(DALL·E)

 

This is a minimal implementation of Boris Dayma's DALL·E Mini in PyTorch. It has been stripped to the bare essentials necessary for doing inference. The only third party dependencies are numpy and torch.

It currently takes 7.4 seconds to generate an image with DALL·E Mega with PyTorch on a standard GPU runtime in Colab

The flax model, and the code for coverting it to torch, have been moved here.

Install

$ pip install min-dalle

Usage

Use the python script image_from_text.py to generate images from the command line.

$ python image_from_text.py --text='artificial intelligence' --seed=7

$ python image_from_text.py --text='court sketch of godzilla on trial' --mega

To load a model once and generate multiple times, initialize MinDalleTorch, then call generate_image with some text and a seed.

from min_dalle import MinDalleTorch

model = MinDalleTorch(
    is_mega=True, 
    is_reusable=True,
    models_root='./pretrained'
)

image = model.generate_image("a comfy chair that looks like an avocado")
display(image)

image = model.generate_image("trail cam footage of gollum eating watermelon", seed=1)
display(image)

Model parameters will be downloaded as needed to the directory specified. The models can also be manually downloaded here.