Open Stables is ML infrastructure code for Stable Diffusion based web services. The initial deploy environment is AWS only, with more cloud provider deployment options to follow. Open Stables is architectured for auto-scaling, Dockerized cloud-deployments. All Open Stables code is permissiviely open sourced.
Goal
The main goal of this project is to enable folks to easily use Stable Diffusion inside their own private cloud accounts. For example AWS rents out NVIDIA A10G GPUs with 24 GB of VRAM, for about one dollar per hour. That is a spiffy GPU for running Stable Diffusion. Open Stables enables private experimentation with production qaulity Stable Diffusion infrastructure, without the up-front cost of purchasing a high-end GPU.
Status
Currently the following popular Stable Diffusion based services have been made to work on Open Stables:
- StableStudio
- InvokeAI
- Automatic1111
- ComfyUI
Each of the above four codebases is a web application that runs Stable Diffusion in a web server with a GPU. The Open Stables project packages all of those applications to run interoperably on AWS. This includes web UI front ends and API service backends. In this way a single project can be used to provision all the Stables. In other words, Open Stables is open code for all the Stables.
Architecture
The Open Stable architecture is very Docker centric. Initial releases of Open Stable are targeted specifically for AWS. The long term goal is to have Open Stables run well on all the major cloud providers, as well as stock Kubernetes on private metal data centers. The planned development roadmap is:
- Amazon
- Microsoft
- Kubernetes more broadly
License
This is the main repository of the Open Stables project. All Open Stables code is Apache-2.0 licensed. Some of the above listed codebases that can be deployed via Open Stables code have different licenses. Specific applications packaged to run atop Open Stables are hosted in separate repositories.
For more information, see the wiki.
