SDXL on Sagemaker Studio Lab with InvokeAI

We've put together an installer script for bringing InvokeAI to Sagemaker Studio Lab and tested how the app runs on this platform. You can find a step-by-step walkthrough with the pros and cons in the video below.

Requirements

• A Sagemaker Studio Lab account
• Authentication token from ngrok.com
• 17 GB disk space in the home folder
• Empty temporary storage. It's empty by default each time the instance starts.
• For reasonably fast image generation, connect to a GPU runtime.

Installation and Starting Up

• Get the scripts from GitHub: https://github.com/wandaweb/InvokeAI-Sagemaker-Studio-Lab
  
• Open a new Terminal from the launcher
  
• Enter   
  sh install-dev.sh  
sh configure.sh  
sh start.sh
• The first time the app starts, paste your ngrok authentication token in the console input, and, optionally, the static domain name.  
• Visit your ngrok URL by clicking the link in the output.
   

Limitations

• Uses SDXL models in diffusers format, such as stabilityai/stable-diffusion-xl-base-1.0, Lykon/dreamshaper-xl-1-0 or Linaqruf/animagine-xl. These can be found on HuggingFace.
• Only one SDXL LoRA. Large LoRA might cause the app to crash. We've had success with TShirtDesignRedmond and DreamArt.