Use open source tools and generative AI for asynchronous image generation
This Guidance helps you implement a scalable and low-cost Stable Diffusion (SD) web user interface (UI) inference architecture on AWS. SD is a popular open source project for generating images using generative AI. This Guidance shows how to use serverless and container services to build and deploy an end-to-end, low-cost, and asynchronous image generation architecture.
Please note: [Disclaimer]
Architecture Diagram
[Architecture diagram description]
Step 1
Users send prompts to an application running on AWS Fargate through an Application Load Balancing endpoint.
Step 2
The application sends the prompt to Amazon API Gateway that acts as an endpoint for the overall Guidance, including authentication. AWS Lambda validates the requests, publishes them to the designated Amazon Simple Notification Service (Amazon SNS) topic, and immediately returns a response.
Step 3
Amazon SNS publishes the message to Amazon Simple Queue Service (Amazon SQS) queues. Each message contains a Stable Diffusion (SD) model name attribute and will be delivered to the queues with matching SD model names.
Step 4
In the Amazon Elastic Kubernetes Service (Amazon EKS) cluster, the previously deployed open source Kubernetes Event Driven Auto-Scaler (KEDA) scales up new pods to process the incoming messages from SQS model processing queues.
Step 5
In the Amazon EKS cluster, Karpenter, an open source Kubernetes compute auto-scaler, launches new compute nodes based on GPU Amazon Elastic Compute Cloud (Amazon EC2) Spot instances (such as G4, G5, and P4) to schedule pending pods. The instances use pre-cached SD Runtime images and are based on Bottlerocket OS for fast boot.
Step 6
SD Runtimes load ML model files from Amazon Elastic File System (Amazon EFS) file system upon pod initializations.
Step 7
Queue agents (a program created for this Guidance) receive messages from SQS model processing queues and convert them to inputs for SD Runtime APIs calls.
Step 8
Queue agents call SD Runtime APIs, receive and decode responses, and save the generated images to Amazon Simple Storage Service (Amazon S3) buckets.
Step 9
Queue agents send notifications to the designated SNS topic from the pods and the application receives notifications from the corresponding SQS queue.
Step 10
An application running on Fargate downloads the generated images from the S3 bucket and renders them to users.
Step 1
Users send prompts to an application running on AWS Fargate through an Application Load Balancing endpoint.
Step 2
The application sends the prompt to Amazon API Gateway that acts as an endpoint for the overall Guidance, including authentication. AWS Lambda validates the requests, publishes them to the designated Amazon Simple Notification Service (Amazon SNS) topic, and immediately returns a response.
Step 3
Amazon SNS publishes the message to Amazon Simple Queue Service (Amazon SQS) queues. Each message contains a Stable Diffusion (SD) model name attribute and will be delivered to the queues with matching SD model names.
Step 4
In the Amazon Elastic Kubernetes Service (Amazon EKS) cluster, the previously deployed open source Kubernetes Event Driven Auto-Scaler (KEDA) scales up new pods to process the incoming messages from SQS model processing queues.
Step 5
In the Amazon EKS cluster, Karpenter, an open source Kubernetes compute auto-scaler, launches new compute nodes based on GPU Amazon Elastic Compute Cloud (Amazon EC2) Spot instances (such as G4, G5, and P4) to schedule pending pods. The instances use pre-cached SD Runtime images and are based on Bottlerocket OS for fast boot.
Step 6
SD Runtimes load ML model files from Amazon Elastic File System (Amazon EFS) file system upon pod initializations.
Step 7
Queue agents (a program created for this Guidance) receive messages from SQS model processing queues and convert them to inputs for SD Runtime APIs calls.
Step 8
Queue agents call SD Runtime APIs, receive and decode responses, and save the generated images to Amazon Simple Storage Service (Amazon S3) buckets.
Step 9
Queue agents send notifications to the designated SNS topic from the pods and the application receives notifications from the corresponding SQS queue.
Step 10
An application running on Fargate downloads the generated images from the S3 bucket and renders them to users.
Well-Architected Pillars
The AWS Well-Architected Framework helps you understand the pros and cons of the decisions you make when building systems in the cloud. The six pillars of the Framework allow you to learn architectural best practices for designing and operating reliable, secure, efficient, cost-effective, and sustainable systems. Using the AWS Well-Architected Tool, available at no charge in the AWS Management Console, you can review your workloads against these best practices by answering a set of questions for each pillar.
The architecture diagram above is an example of a Solution created with Well-Architected best practices in mind. To be fully Well-Architected, you should follow as many Well-Architected best practices as possible.
-
Operational ExcellenceRead the Operational Excellence whitepaper
AWS X-Ray traces a request from API Gateway to Lambda, Amazon SNS, Amazon SQS, Amazon EKS pods, and all the way to Amazon S3. This allows users to quickly troubleshoot requests. Additionally, X-Ray provides a service map that allows the user to identify the current system status with one glance.
AWS Distro for OpenTelemetry collects metrics from the EKS cluster and sends them to Amazon CloudWatch. Distro for OpenTelemetry and CloudWatch help you obtain critical metrics and receive automatic alerts when metrics exceed thresholds. CloudWatch Container Insights helps to visualize those metrics.
-
SecurityRead the Security whitepaper
AWS Identity and Access Management (IAM) provides access control for API Gateway, Lambda, Amazon SNS, Amazon SQS, and Amazon S3. IAM Roles for Service Accounts (IRSA) on EKS clusters provides fine-grained access control for pods running inside the EKS cluster. IAM and IRSA on EKS clusters enforce role-based access control and limit unauthorized access to resources.
-
ReliabilityRead the Reliability whitepaper
All the services in this Guidance are regional services with built-in high availability and fault tolerance against Availability Zone failure. In addition to the high availability of each individual service, this Guidance uses a loosely-coupled microservices architecture. Additionally, Amazon S3 and Amazon EFS provide highly reliable storage service to support uptime.
-
Performance EfficiencyRead the Performance Efficiency whitepaper
Amazon EKS provides flexible container scheduling and scalability. KEDA is an event-driven pod auto-scaler for Kubernetes. Karpenter simplifies Kubernetes infrastructure and automatically launches the right amount of compute resources to handle the cluster's applications. Combined with KEDA and Karpenter, Amazon EKS can scale up a new node in less than one minute. Bottlerocket is a container-optimized operating system and has a shorter start up time than Amazon Linux 2.
-
Cost OptimizationRead the Cost Optimization whitepaper
Spot instances offer a lower price than on-demand instances. Stable Diffusion Web UI in Amazon EKS requires GPU instances for inference, and using Spot instances could save up to 70% on costs. Because SQS queues store the inference tasks, you won’t lose data in the event of spot interruption.
Additionally, Amazon Elastic Container Service (Amazon ECS) on Fargate provides serverless infrastructure for running containerized applications. Amazon EKS provides a cost-efficient way to deploy, run, and manage containers through standard APIs. Amazon ECS on Fargate and Amazon EKS provide container orchestration with configurable infrastructure options to optimize costs.
-
SustainabilityRead the Sustainability whitepaper
Lambda automatically scales based on requests and does not charge for idle infrastructure. This reduces the compute usage for the frontend API. Karpenter has a container workload consolidation feature that periodically checks if the current workloads can be consolidated onto existing EKS compute nodes with a lesser resource footprint, which helps to reduce idle resources. Amazon ECS on Fargate provides a serverless infrastructure option to run containers with a minimal resource footprint.
Implementation Resources
A detailed guide is provided to experiment and use within your AWS account. Each stage of building the Guidance, including deployment, usage, and cleanup, is examined to prepare it for deployment.
The sample code is a starting point. It is industry validated, prescriptive but not definitive, and a peek under the hood to help you begin.
Related Content
[Title]
Disclaimer
The sample code; software libraries; command line tools; proofs of concept; templates; or other related technology (including any of the foregoing that are provided by our personnel) is provided to you as AWS Content under the AWS Customer Agreement, or the relevant written agreement between you and AWS (whichever applies). You should not use this AWS Content in your production accounts, or on production or other critical data. You are responsible for testing, securing, and optimizing the AWS Content, such as sample code, as appropriate for production grade use based on your specific quality control practices and standards. Deploying AWS Content may incur AWS charges for creating or using AWS chargeable resources, such as running Amazon EC2 instances or using Amazon S3 storage.
References to third-party services or organizations in this Guidance do not imply an endorsement, sponsorship, or affiliation between Amazon or AWS and the third party. Guidance from AWS is a technical starting point, and you can customize your integration with third-party services when you deploy the architecture.