Big Sleep

Big Sleep is an open source command line tool for text to image generation, published on GitHub by lucidrains under the MIT license. It combines OpenAI's CLIP with the generator from a BigGAN, a technique originally created by Ryan Murdock, and wraps the work so that anyone with a GPU can use it. The repository is a Python package, and the vendor lists it as a simple tool rather than a hosted service. A user installs it with pip and runs a single command in the terminal, giving a text prompt such as a pyramid made of ice. The GAN then dreams up images that match the phrase, saving them to the folder where the command was run. It can also be called from Python code through an Imagine class, with settings such as learning rate and how often progress images are saved. Several phrases can be trained together using a delimiter, and the repository links to notebooks, including a simplified version and a user made one with bug fixes and Google Drive integration. The tool suits researchers, hobbyists and developers who want to experiment with early CLIP guided image synthesis on their own hardware. The software is free under the MIT license and the vendor charges no price for it. Running it requires a GPU, so the only cost is the user's own hardware or cloud compute.