Guidance For Asynchronous Inference With Stable Diffusion On Aws

[Guidance for Asynchronous Inference with Stable Diffusion on AWS](https://aws.amazon.com/solutions/guidance/asynchronous-image-generation-with-stable-diffusion-on-aws/)

Implementing a fast scaling and low cost Stable Diffusion inference solution with serverless and containers on AWS

[Stable Diffusion](https://aws.amazon.com/what-is/stable-diffusion/) is a popular open source project for generating images using Generative AI. Building a scalable and cost efficient Machine Learning (ML) Inference solution is a common challenge that many AWS customers are facing. This project shows how to use serverless architecture and container services to build an end-to-end low cost, rapidly scaling asynchronous image generation architecture. This repo contains the sample code and CDK deployment scripts that will help you to deploy this solution in a few steps.

Features

• Asynchronous API and Serverless Event-Driven Architecture
• Image Generation with open source Stable Diffusion runtimes running on Amazon EKS (Kubernetes 1.35)
• Automatic Amazon SQS queue length based scaling with KEDA
• Automatic provisioning of EC2 instances for Amazon EKS compute Nodes with Karpenter v1 (stable CRDs)
• SOCI Parallel Pull with NVMe instance store for fast container startup (up to 60% faster image pulls)
• Scaling up new Amazon EKS nodes within 2 minutes to run Inference tasks
• Saving up to 70% with AWS GPU spot EC2 instances
• Bottlerocket OS with IMDSv2 enforced for enhanced security

Deployment Documentation

Please see detailed Implementation Guides here:

• [English](https