Getting Started

This demo shows three Generative AI use-cases integrated into single solution on Miro board. It turns Python notebooks into dynamic interactive experience, where several team members can brainstorm, explore, exchange ideas empowered by privately hosted Sagemaker generative AI models. This demo can be easily extended by adding use-cases to demonstrate new concepts and solutions.

Usage instructions:

Use case	How it looks like	SageMaker JumpStart models	Details
1. Image generation		Stable Diffusion 2.1	To generate new image: Select one or several yellow stickers with prompts, then run the app.
2. Image inpainting		Stable Diffusion 2 Inpainting	To transform a part of image: Define changing part of image using round shape, add a sticky note with change prompt, connect image and sticky note, then select all 4 items and run the app.
3. Image trasformation		Stable Diffusion 2 Depth FP16	To transform image: Select image and sticky note with transformation prompt connected by line, then run the app.

Start from brainstorming and then develop your visual idea step-by-step.

💡 Tips: you can use resulted image from previous step as an input for the next.

Architecture overview

• Miro application is running on the board. Loaded from S3 bucket, accessed via CloudFront distribution. Written on TypeScript.
• Authorization and AIML proxy lambdas. Accessed via APIGateway deployed behind CloudFront. Written on Python.
  • Authorization function authorize provide access to backend functions only for authorized Miro application. It's used to protect organization data and generated content in AWS account.
  • AIML proxy function mlInference is required to handle API call from application and redirect it to correct Sagemaker inference endpoint. It also can be used for more complex use-cases, when several AIML functions work.
• Sagemaker inference endpoints. Run inference.

The demo could be extended in two ways:

1. by adding new AIML use cases.
2. by changing/empowering interface on Miro board or Web-interface.

In both cases existing environment can be used as boilerplate. More details here

Deployment

Prerequisites

1. AWS account with access to create
   • IAM roles
   • Lambda functions
   • API Gateway endpoints
   • S3 buckets
   • CloudFront distributions
2. AWS CLI installed and configured
3. NodeJS installed
4. NPM installed
5. AWS CDK installed (min. version 2.94.x is required)
6. Docker installed

To begin setup Generative AI demo in your AWS account, follow these steps:

Start with creating Miro application

1. Familiarize yourself with Miro's Developer Platform:

Visit the Miro Developer Platform documentation (https://developers.miro.com/docs) to learn about the available APIs, SDKs, and other resources that can help you build your app.

2. Create Miro Developer Team

💡 If you already have Miro Developer Team in you account, skip this step.

3. Go to the Miro App management Dashboard (https://miro.com/app/settings/user-profile/apps/)

and click "Create new app".

Fill in the necessary information about your app, such as its name, select Developer team. Note: you don't need to check the "Expire user authorization token" checkbox. Click "Create app" to create your app.

4. Copy client secret on app creation page

Now everything ready to set up backend

5. Configure CLI access to AWS account via profile or environment variables

👇 Demo operator user/role policies

(steps below developed and tested in Cloud9 and Sagemaker, role with following policies)

IAMFullAccess, AmazonS3FullAccess, AmazonSSMFullAccess, CloudWatchLogsFullAccess,
CloudFrontFullAccess, AmazonAPIGatewayAdministrator, AWSCloudFormationFullAccess, 
AWSLambda_FullAccess, AmazonSageMakerFullAccess

6. Bootstrap CDK stack in the target account:

cdk bootstrap aws://<account_id>/<region>

7. Docker buildx is required to build Graviton2 Lambda images on x86 platform.

It could be either used from Docker Desktop package - no need in steps 7.1 and 7.2 in this case; or installed separately (steps below developed and tested on AWS Cloud9):

1. Binary installation manual

2. On x86 platform to enable multiarch building capability launch

   docker run --rm --privileged multiarch/qemu-user-static --reset -p yes

8. Configure Miro application client secret

Edit deployment-config.json to authorize Miro application to access backend. Find the following parameter and put as value secret string you received in step 4.

"clientAppSecret": "ADD_MIRO_APP_CLIENT_SECRET_HERE",

9. Deploy backend

Run npm run deploy from the project root folder. You will be requested to provide application client secret to continue installation. When installation is completed, all the necessary resources are deployed as CloudFormation DeployStack in the target account. Write down CloudFront HTTPS distrubution address:

    DeployStack.DistributionOutput = xyz123456.cloudfront.net

10. Return to Miro application creation dialog to complete app configuration

Please enter the CloudFront URL that you obtained on the previous step.

11. Add necessary permission.

12. Install the app to the team.

Back to the Miro Developer Dashboard, click "More apps" icon on application bar, find your just installed app in the list and start working.

Sagemaker endpoints

You need to run dedicated Sagemaker endpoint for each use case.

• 1-create_image image generation with Stable diffusion 2.1 in SageMaker JumpStart
• 2-inpaint_image image inpainting with Stable diffusion 2 Inpainting fp16 in SageMaker JumpStart
• 3-modify_image image modification with Stable Diffusion 2 Depth FP16 in SageMaker JumpStart

[Legacy]

You need to run dedicated Sagemaker endpoint for each use case. Each use-case is supported by a separate Jupyter notebook in ./ml_services/<use_case> subdirectory:

• 1-create_image image generation (Stable diffusion 2.1), based on this example
• 2-inpaint_image image inpainting (Stable diffusion 2 Inpainting fp16), based on this example
• 3-modify_image image pix2pix modification (Huggingface instruct pix2pix), based on this example

Starting Sagemaker endpoints

• Access SageMaker Studio in your account
• Navigate to SageMaker JumpStart
• Search and Select one of Models [Stable Diffusion 2.1, Stable Diffusion 2 Inpainting, Stable Diffusion 2 Depth FP16]
• In Deploy section click deploy button.

[Legacy]

• Go to ./ml_services/<use_case> directory and run one-by-one all three Sagemaker notebooks.
• After endpoint started and successfully tested in notebook, go to Miro board, select required items and run use-case.

Demo extension with additional use-cases

🛸 Extension guidance --TBD-- ⏳

License

This library is licensed under the MIT-0 license. For more details, please see LICENSE file

Legal disclaimer

Sample code, software libraries, command line tools, proofs of concept, templates, or other related technology are provided as AWS Content or Third-Party Content under the AWS Customer Agreement, or the relevant written agreement between you and AWS (whichever applies). You should not use this AWS Content or Third-Party Content in your production accounts, or on production or other critical data. You are responsible for testing, securing, and optimizing the AWS Content or Third-Party Content, such as sample code, as appropriate for production grade use based on your specific quality control practices and standards. Deploying AWS Content or Third-Party Content may incur AWS charges for creating or using AWS chargeable resources, such as running Amazon EC2 instances or using Amazon S3 storage.