Figment Fusion is a Jupyter notebook for AI-based image generation. It is powered by the Diffusers library and is pre-configured for the Stable Diffusion and Waifu Diffusion models.
The project is primarily aimed at technology enthusiasts who may not have a background in machine learning but are eager to get started. The goal is to create a tool that encourages playing with source code and assists in understanding how to use diffusion models in your projects.
The simplest way to get started with Figment Fusion is to open the notebook file directly from this repository in Colaboratory, a Google service. Try clicking the “Open in Colab” button!
Each step is explained right in the file, and additional documentation can be found on the Figment Fusion wiki.
These are the most relevant features that should be added to future Figment Fusion releases.
-
Schedulers
Figment Fusion uses the schedulers provided by Diffusers, which is missing some well-known options such as Euler. This is being worked on by contributors of the library (issue #277 at the Diffusers project). -
Image-to-image and inpainting
Not yet implemented because it has not been decided how to build a practical image-to-image and inpainting user interface that can be embedded in the notebook. Suggestions are welcome (issue #1).
-
Jupyter-first
Figment Fusion was designed from the start to be used as a notebook, rather than being converted as an afterthought, resulting in a better user experience. Furthermore, it was built for Jupyter, with no Colab-specific features required for it to function out-of-the-box, enabling portability to more generic environments such as the ones provided by servers on RunPod and Vast.ai. -
Simplicity
“Do one thing and do it well” is a Unix philosophy principle that Figment Fusion follows. As a utility designed solely for image generation, it does not include features such as facial enhancement and upscaling. This implies that it is not a complete solution for creating artwork, as the use of complementary tools is required to extract the highest quality possible from the generated work, but it does have the advantage of making the code more maintainable and the setup steps run much faster. -
Openness
Figment Fusion is released in the public domain, which means that anyone can freely copy, modify, and share its code. This should be beneficial to those who wish to use it as a starting point for their own projects and thus contribute to the growth of the AI artwork ecosystem.