Skip to content
/ img-studio Public

ImgStudio is a NextJS web app designed for easy deployment and user-friendly experience, streamlining access to the power of Google's GenAI model Imagen 3 to generate powerful images đŸ”„

License

MIT license
10 stars 17 forks Branches Tags Activity
Star
Notifications You must be signed in to change notification settings

aduboue/img-studio

BranchesTags

Repository files navigation

Infrastructure setup guide for ImgStudio

0\ Get access to Imagen models

  • In general, for Vertex: in the console
    • Go to Vertex AI > Enable all recommended APIs (they should include: Vertex AI API, Cloud Storage API)
    • Make sure the Vertex Service Account exists in your project service-PROJECT_NUMBER@gcp-sa-aiplatform.iam.gserviceaccount.com
  • For Imagen Generation:
    • Models are now in public GA, Imagen 3 Generate (imagen-3.0-generate-001) and Imagen 3 Generate Fast (imagen-3.0-fast-generate-001)
    • For people generation (adult and/ or children), you now need to fill out this form to get access.
  • For Imagen Editing & Customization
    • You can fill out this form to get access to the Preview feature (name: imagen-3.0-capability-001)
    • You will also need the Vertex Image Segmentation model when using editing, fill out this form to get access (name: image-segmentation-001)

1\ Create Cloud Storage buckets

  • Specifications: Regional in your desired region (ex: europe-west9 in Paris)
  • Create 3 buckets
    • Raw generated output content: YOUR_COMPANY-imgstudio-output
    • Shared content: YOUR_COMPANY-imgstudio-library
    • Configuration file bucket: YOUR_COMPANY-imgstudio-export-config
      • Here upload export-fields-options.json a configuration file specific to your usage that you can find an exemple of in the repository, its purpose is to setup the desired metadata you want to set for your generated content
      • In this file, for each fields (ex: contextAuthorTeam, contextTargetPlatform, contextAssociatedBrand, contextCollection), you can only change the ID of the field (ex: “contextAuthorTeam”), its label (ex: “In which team are you?”), its name (ex: “Associated team(s)”), its tag isMandatory (ex: true) and finally its options
      • Attention! The ID and the options’ values must only be letters, no spaces, no special characters, starting with a lowercase letter

2\ Setup Cloud Build trigger

  • Name: YOUR_COMPANY-imgstudio
  • Event: Manual invocation
  • Source:
    • Go to the public repository https://github.com/aduboue/img-studio
    • Monitor new releases to stay up-to-date: click on Watch > Custom > Releases > Apply
    • Use the top right button to setup a GitHub fork from the repository, which will create a copy repository in your own GitHub account
    • Back in Cloud Build, log into your GitHub account, then select the newly created repository
  • Configuration: Cloud Build configuration file (yaml or json)
    • Cloud Build configuration file location: /cloudbuild.yaml
  • Put in 7 substitution variables:
    • _NEXT_PUBLIC_EXPORT_FIELDS_OPTIONS_URI
      • The URI of the configuration JSON file in its bucket
      • Ex: gs://YOUR_COMPANY-imgstudio-export-config/export-fields-options.json
    • _NEXT_PUBLIC_GCS_BUCKET_LOCATION
      • The region selected for your GCS buckets
      • Ex: europe-west9
    • _NEXT_PUBLIC_VERTEX_API_LOCATION
      • The region you want to use for VertexAI APIs
      • Ex: europe-west9
    • _NEXT_PUBLIC_GEMINI_MODEL = gemini-1.5-flash-001
      • Don’t change this
    • _NEXT_PUBLIC_OUTPUT_BUCKET
      • The name of the raw generated output content bucket
      • Ex: YOUR_COMPANY-imgstudio-output
    • _NEXT_PUBLIC_TEAM_BUCKET
      • The name of the shared content bucket
      • Ex: YOUR_COMPANY-imgstudio-library
    • _NEXT_PUBLIC_PRINCIPAL_TO_USER_FILTERS
      • The sections of your users’ email address used to log in via IAP that will need to be removed in order to get their user ID, separated by commas
      • Ex: my email address is ‘admin-jdupont@company.com’, the value to set would be admin-,@company.com so that the user ID jdupont can be extracted
    • _NEXT_PUBLIC_EDIT_ENABLED
      • Allow to enable edit features, set it to false if you do not have access yet
    • _NEXT_PUBLIC_SEG_MODEL
      • Only mandatory if Edit is enabled
      • Service name for the Vertex Segmentation model, when you get access to it (see Step 0)
  • Service account: select the default already existing Cloud Build service account PROJECT_NUMBER-compute@developer.gserviceaccount.com
    • You may want to check in IAM it has the roles: Artifact Registry Writer and Logs Writer
  • > Save
  • Manualy run your first build !

3\ Enable IAP & configure oAuth Consent Screen

  • Go to Security > Identity Aware Proxy and enable the API
  • > Configure consent screen (oAuth)
    • User type can be either
      • Internal if you want to limit IAP users to your GCP org domain
      • External if you have some users on a different domain than your GCP org’s
    • > Create
    • Fill in
      • App Name: YOUR_COMPANY-imgstudio
      • User Support Email
      • Authorized Domain: YOUR_COMPANY_DOMAIN
      • Developer Contact Email
    • > Save and Continue
  • Add any needed scopes or > Save and Continue
  • Review the Summary
  • > Back to Dashboard

4\ Create application Service Account

  • Go to IAM > Service Accounts > Create Service Account
  • Provide name: YOUR_COMPANY-imgstudio-sa
  • Give roles:
    • Cloud Datastore User
    • Logs Writer
    • Secret Manager Secret Accessor
    • Service Account Token Creator
    • Storage Object Creator
    • Storage Object Viewer
    • Vertex AI User

5\ Deploy Cloud Run service

  • Deploy container > Service
  • Deploy one revision from an existing container image
  • Container image > Select from Artifact registry the latest image you just build in Cloud Build
  • Name: YOUR_COMPANY-imgstudio-app
  • Set your region (ex: europe-west9)
  • Authentication: Require authentication
  • Ingress Control > Internal > Allow traffic from external Application Load Balancers
  • Container(s) > Container port: 3000
  • Security > Service account: YOUR_COMPANY-imgstudio-sa
  • > Create
  • NB: if you try to access the published URL for the new service you should receive an error message stating “Error: Page Not Found”, this is due to the fact that we are only allowing ingress for external traffic from a Load Balancer

6\ Grant IAP Permissions on Cloud Run service

  • Create the IAP service account address
    • Go to the top right of the console > Shell icon “Activate Cloud Shell”
    • Wait for machine to setup
    • In the terminal, use this command and copy the output service account address
      • gcloud beta services identity create --service=iap.googleapis.com --project=PROJECT_ID
    • The format of the output you can copy should be service-PROJECT_NUMBER@gcp-sa-iap.iam.gserviceaccount.com
  • Go to Cloud Run, in the Services list, select the checkbox next to the name of your service
    • Click on Permissions > Add Principal
    • Grant the Cloud Run Invoker role to the previously created/ fetched IAP service account

7\ Create DNS Zone

  • Network Service > Cloud DNS
  • > Create zone
  • Complete the form
    • Zone Name: imgstudio
    • DNS Name: imgstudio.YOUR_COMPANY_DOMAIN
    • DNSSEC: Off
    • Cloud Logging: Off
  • Once DNS propagation is completed, verify the name servers of the DNS managed zone using the command below (it could take several hours to complete)
    • dig imgstudio.YOUR_COMPANY_DOMAIN NS +short

8\ Create Load Balancer & SSL Certificate

  • Network Service > Load Balancing
  • > Create Load Balancer
  • Select:
    • Type of Load Balancer: Application Load Balancer (HTTP/HTTPS)
    • Public Facing or Internal: Public Facing (external)
    • Global or single region: Global
    • Load Balancer Generation: Global external Application Load Balancer
  • > Configure
  • Load balancer name: YOUR_COMPANY-imgstudio-lb
  • Frontend configuration:
    • Protocol: HTTPS
    • IP address can be left Ephemeral (you could also configure a static IP)
    • Certificate dropdown > Create a New Certificate
      • Name: YOUR_COMPANY-imgstudio-cert
      • > Create Google-managed certificate
      • Domain 1: imgstudio.YOUR_COMPANY_DOMAIN
      • > Create
    • > Done
  • Backend Configuration
    • Backend Services & Backend Buckets > Create a Backend Service
    • Name: YOUR_COMPANY-imgstudio-back
    • Backend type: Serverless Network Endpoint Group > Done
    • Backends > New backend > Serverless Network Endpoint Groups dropdown > Create Serverless Network Endpoint Group
      • Name: YOUR_COMPANY-imgstudio-neg
      • Region: the region the Cloud Run service was deployed to (ex: europe-west9)
      • Serverless network endpoint group type > Cloud Run > Select service > YOUR_COMPANY-imgstudio-app > Create
      • Enable Cloud CDN > Off
      • > Create
  • Review the Load Balancer configuration
  • > Create

9\ Create DNS Record for Load Balancer frontend

  • Network Services > Load Balancing, select your load balancer YOUR_COMPANY-imgstudio-lb
  • Details > Frontend > IP-Port
  • Note the IP address
  • Network Services > Cloud DNS, select your DNS Zone imgstudio
  • Record Sets > Add Standard
  • Create record set
    • DNS name: imgstudio.YOUR_COMPANY_DOMAIN
    • IPv4 Address: your load balancer IP Address
    • > Create

10\ Enable IAP & grant user access

  • Security > Identity-Aware Proxy
  • Turn on IAP for your backend service YOUR_COMPANY-imgstudio-back
  • Select the checkbox next to your service, then > Add Principal
  • Enter the address of the user (or group) you want to allow access to imgstudio
  • Assign the role IAP-secured Web App User, > Save

11\ Firestore Database creation

  • Firestore > Create database
    • Firestore mode: Native mode, > Continue
    • Database ID: (default) (very important you keep it that way)
    • Location type: Region
    • Region: your desired region (ex: europe-west9 in Paris)
    • Secure rules: Production rules
  • Firestore > Indexes > Composite indexes > Create Index
    • Collection ID: metadata
    • Fields to index
      • Field path 1: combinedFilters, Index options 1: Array contains
      • Field path 2: timestamp, Index options 2: Descending
      • Field path 3: __name__, Index options 3: Descending
    • Query scope: Collection
    • > Create
    • Wait for the index to be successfully created!
  • Let’s setup security rules on your database, and only allow your Cloud Run service account to access it
    • In a new tab, go to
      • https://console.firebase.google.com/project/PROJECT_ID/firestore/databases/-default-/rules
      • If necessary, follow the steps to setup your Firebase project
      • Once in Firestore Database > Rules, go to the security rules editor
      • Write the following content, don’t forget to replace YOUR_COMPANY & PROJECT_ID in the Cloud Run service account rules_version = '2'; service cloud.firestore { match /databases/{database}/documents { match /{document=**} { allow read, get, list, create, update: if get(/databases/$(database)/documents/request.auth.uid).data.serviceAccount == 'YOUR_COMPANY-imgstudio-sa@PROJECT_ID.iam.gserviceaccount.com'; allow delete: if false; } } }
      • > Publish

About

ImgStudio is a NextJS web app designed for easy deployment and user-friendly experience, streamlining access to the power of Google's GenAI model Imagen 3 to generate powerful images đŸ”„

Topics

google-cloud image-generation imagen google-deepmind genai genai-usecase

Resources

Readme

License

MIT license
Activity

Stars

10 stars

Watchers

4 watching

Forks

17 forks
Report repository

Releases 1

ImgStudio for Imagen 3 Latest
Oct 1, 2024

Languages