Boost Image Resolution with Imagen on Vertex AI

Unlock the power of high-resolution imagery by using Imagen on Vertex AI to increase the size of your images without sacrificing quality. This guide will walk you through the process, whether you prefer using the graphical interface or a programmatic approach via the REST API.

Model Version Compatibility

It's important to note that the upscaling feature is specific to certain model versions. Currently, it is only available for the original Imagen model.

Feature	Imagen (v.002)	Imagen 2 (v.005)	Imagen 2 (v.006)
Upscaling	✔	Not supported	Not supported

Below are the code samples and steps to upscale an existing, generated, or edited image.

Upscaling via the Console

For a straightforward, user-friendly approach, you can use the Vertex AI console:

1. First, follow the instructions to generate an image with text.
2. Select the image you wish to enhance.
3. Click Upscale/export.
4. From the options, select Upscale images.
5. Choose your desired Scale factor (2x or 4x).
6. Click Export to save your new, high-resolution image.

Upscaling via the REST API

For developers looking to integrate this feature into their applications, the REST API provides a powerful method. For detailed information on imagegeneration model requests, refer to the model API reference.

To activate upscaling, you must include the "mode": "upscale" parameter and the upscaleConfig object in your JSON request.

Before you begin, replace the following placeholders in the request:

• LOCATION: The region for your project, such as us-central1. See the list of available regions.
• PROJECT_ID: Your unique Google Cloud project ID.
• B64_BASE_IMAGE: The image you want to upscale, provided as a base64-encoded byte string (10 MB limit).
• IMAGE_SOURCE: Alternatively, the Cloud Storage URI of your source image, e.g., gs://output-bucket/source-photos/photo.png.
• UPSCALE_FACTOR: Optional. The factor to scale the image by. Available values are x2 or x4.

API Endpoint

Use the following POST request URL:

http POST https://LOCATION-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION/publishers/google/models/imagegeneration@002:predict

Request Body

Construct your request with the following JSON structure:

{ "instances": [ { "prompt": "", "image": { // Use one of the following to specify the image "bytesBase64Encoded": "B64_BASE_IMAGE", "gcsUri": "IMAGE_SOURCE" } } ], "parameters": { "sampleCount": 1, "mode": "upscale", "upscaleConfig": { "upscaleFactor": "UPSCALE_FACTOR" } } }

Sending the Request

To execute your request, you can use one of the following command-line tools.

Using cURL

Save the request body to a file named request.json and run the following command:

bash curl -X POST
-H "Authorization: Bearer $(gcloud auth print-access-token)"
-H "Content-Type: application/json; charset=utf-8"
-d @request.json
"https://LOCATION-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION/publishers/google/models/imagegeneration@002:predict"

Using PowerShell

Save the request body to a file named request.json and execute this command:

powershell $cred = gcloud auth print-access-token $headers = @{ "Authorization" = "Bearer $cred" }

Invoke-WebRequest -Method POST -Headers $headers -ContentType: "application/json; charset=utf-8" -InFile request.json ` -Uri "https://LOCATION-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION/publishers/google/models/imagegeneration@002:predict" | Select-Object -Expand Content

Sample API Response

A successful request will return a JSON response containing the upscaled image encoded in base64:

{ "predictions": [ { "mimeType": "image/png", "bytesBase64Encoded": "iVBOR..[base64-encoded-upscaled-image]...YII=" } ] }

What's Next?

To learn more about the latest advancements in Google's generative AI, explore these resources:

• A developer's guide to getting started with Imagen 3 on Vertex AI
• New generative media models and tools, built with and for creators
• New in Gemini: Custom Gems and improved image generation with Imagen 3
• Google DeepMind: Imagen 3 - Our highest quality text-to-image model