IMPORTANT UPDATE 28.07.2023 The domain for this app expired and I've decided to delete the AWS account to prevent any additional costs. All user data will be deleted and no backups of any kind were done. If you access to some images that were generated by the system, I might be able to restore it by cancelling the account deletion, but there's no guarantee that it will work. Please contact me before 3 months if you need me to do this.
This project was built as MVP to understand the landscape of image generation with Stable Diffusion. Ultimately, after about 3 weeks of development, it was realized this is not a good product for ongoing development, for the following reasons:
- Most people are instead interested in running Stable Diffusion quickly and cheaply.
- In the cloud, you cannot do it cheaply and quickly, e.g, you need to over-provision GPU resources which are expensive.
- Many people execute it locally, and a wide range of open source repositories have emerged. Anyone that owns a GPU with about 4GB of RAM can use a fork of the stable diffusion web UI to run it locally. Here's an example repository with tons of features: https://github.com/AUTOMATIC1111/stable-diffusion-webui
This was nonetheless a very interesting project to develop, which provided good experience in many ways.
If you're interested, you can see the original system design and planning here:
https://pacific-drip-6f9.notion.site/Openimagegenius-f5eda2bd24b745a9a38f7124327fa08e
You can also access the playground to use AWS Lambda to generate images (at medium quality, currently 32 inference steps):
https://auth.openimagegenius.com/
The UI is very raw, and is just a Google OAuth screen:
It requires you to share your e-mail address.
Generating the first image takes about 3 minutes. The following will take between 1-2 minutes. They will be stored in a public S3 bucket that is behind a CloudFront / and Route 53 subdomain.
If you think this project is worth continuing or forking, feel free to get in touch :)