Use prompt rewriter

Try image generation (Vertex AI Studio)

Try Imagen in a Colab

Imagen on Vertex AI offers an LLM-based prompt rewriting tool, also known as a prompt rewriter. The prompt rewriter helps you obtain higher quality output images by adding more detail to your prompt.

If you disable the prompt rewriter, the quality of the images and how well the output resembles the prompt that you supplied may be impacted. This feature is enabled by default for the following model versions:

  • imagen-4.0-generate-preview-06-06 (Preview)
  • imagen-4.0-fast-generate-preview-06-06 (Preview)
  • imagen-4.0-ultra-generate-preview-06-06 (Preview)
  • imagen-3.0-generate-002

The rewritten prompt is delivered by API response only if the original prompt is fewer than 30 words long.

Use the prompt rewriter

To use the prompt rewriter, do the following:

Console

  1. In the Google Cloud console, go to the Vertex AI > Media Studio page.

    Go to Media Studio

  2. Click Imagen. The Imagen Media Studio image generation page is displayed.

  3. In the Settings panel, adjust the following options:

    • Model: Choose a model from the available options.

      For more information about available models, see Imagen models

    • Number of results: Adjust the slider or enter a value between 1 and 4.

    • In the Negative prompt box, enter a prompt that describes what you don't want generated in the image.

  4. In the Write your prompt box, click Help me write.

    The Enhance my prompt window is displayed.

  5. In the Current prompt box, write your prompt, and then click Enhance.

    The rewritten prompt is displayed in the Enhanced prompt box. You can edit the enhanced prompt or use it as displayed.

  6. Click Insert to use the displayed prompt.

    The prompt is inserted into the Write your prompt box.

  7. Click Generate.

REST

Before using any of the request data, make the following replacements:

  • PROJECT_ID: Your Google Cloud project ID.
  • MODEL_VERSION: The image generation model version to use.

    For more information about model versions and features, see model versions.

  • LOCATION: Your project's region. For example, us-central1, europe-west2, or asia-northeast3. For a list of available regions, see Generative AI on Vertex AI locations.
  • TEXT_PROMPT: The text prompt that guides what images the model generates. Before images are generated, this base prompt is enhanced with more detail and descripitive language using the LLM-based prompt rewriting tool.
  • IMAGE_COUNT: The number of generated images. Accepted integer values: 1-4. Default value: 4.
  • enhancePrompt - A boolean to enable LLM-based prompt enhancement. By default, this value is set to true.

HTTP method and URL: 

POST https://LOCATION-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION/publishers/google/models/MODEL_VERSION:predict

Request JSON body: 

{
  "instances": [
    {
      "prompt": "TEXT_PROMPT"
    }
  ],
  "parameters": {
    "sampleCount": IMAGE_COUNT,
    "enhancePrompt": true
  }
}

To send your request, choose one of these options:

curl

Save the request body in a file named request.json, and execute the following command: 

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/MODEL_VERSION:predict"

PowerShell

Save the request body in a file named request.json, and execute the following command: 

$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/MODEL_VERSION:predict" | Select-Object -Expand Content
With prompt enhancement enabled, the response includes an additional prompt field that shows the enhanced prompt and its associated generated image: 
  {
    "predictions": [
      {
        "mimeType": "MIME_TYPE",
        "prompt": "ENHANCED_PROMPT_1",
        "bytesBase64Encoded": "BASE64_IMG_BYTES_1"
      },
      {
        "mimeType": "MIME_TYPE",
        "prompt": "ENHANCED_PROMPT_2",
        "bytesBase64Encoded": "BASE64_IMG_BYTES_2"
      }
    ]
  }

For example, the following sample response is for a request with "sampleCount": 2 and "prompt": "A raccoon wearing formal clothes, wearing a top hat. Oil painting in the style of Vincent Van Gogh.". The response returns two prediction objects, each with their enhanced prompt and the generated image bytes base64-encoded. 

{
  "predictions": [
    {
      "mimeType": "image/png",
      "prompt": "An oil painting in the style of Vincent van Gogh, depicting a raccoon adorned
        in a finely tailored tuxedo, complete with a crisp white shirt and a bow tie. The raccoon
        also sports a classic top hat, perched jauntily on its head. The painting uses thick,
        swirling brushstrokes characteristic of van Gogh, with vibrant hues of blue, yellow, and
        green in the background, contrasting with the dark tones of the raccoon's attire. The light
        source is subtly placed, casting a dramatic shadow of the raccoon's attire onto the surface
        it sits upon, further enhancing the depth and dimensionality of the composition. The
        overall impression is one of a whimsical and sophisticated character, a raccoon elevated to
        a higher class through its formal attire, rendered in van Gogh's iconic style.",
      "bytesBase64Encoded": "BASE64_IMG_BYTES"
    },
    {
      "mimeType": "image/png",
      "prompt": "An oil painting in the style of Vincent van Gogh featuring a raccoon in a
        dapper suit, complete with a black jacket, crisp white shirt, and a black bow tie. The
        raccoon is wearing a black top hat, adding a touch of elegance to its ensemble. The
        painting is rendered with characteristic van Gogh brushwork, utilizing thick, impasto
        strokes of color. The background is a swirl of blues, greens, and yellows, creating a
        vibrant yet slightly chaotic atmosphere that contrasts with the raccoon's formal attire.
        The lighting is dramatic, casting sharp shadows and highlighting the textures of the fabric
        and the raccoon's fur, enhancing the sense of realism within the fantastical scene. The
        composition focuses on the raccoon's proud posture, highlighting the whimsical contrast of
        a wild animal dressed in formal attire, captured in the unique artistic language of van
        Gogh. ",
      "bytesBase64Encoded": "BASE64_IMG_BYTES"
    }
  ]
}

What's next