Generate text using the Vertex AI Gemini API

In this guide, you send a text prompt request, and then a multimodal prompt and image request to the Vertex AI Gemini API and view the responses.

Prerequisites

To complete this guide, you must have a Google Cloud project with the Vertex AI API enabled. You can use the Vertex AI setup guide to complete these steps.

Add the Vertex AI client library as a dependency

The Vertex AI client library includes many features. Compiling all of them is relatively slow. To speed up compilation, you can just enable the features you need:

cargo add google-cloud-aiplatform-v1 --no-default-features --features prediction-service

Send a prompt to the Vertex AI Gemini API

First, initialize the client using the default settings:

    let client = PredictionService::builder().build().await?;

Then build the model name. For simplicity, this example receives the project ID as an argument and uses a fixed location (global) and the model is hard-coded to a specific version of gemini.

    const MODEL: &str = "gemini-2.5-flash";
    let model = format!("projects/{project_id}/locations/global/publishers/google/models/{MODEL}");

If you want to run this function in your own code, use the project id (without any projects/ prefix) of the project you selected while going through the prerequisites.

With the client initialized you can send the request:

    let response = client
        .generate_content()
        .set_model(&model)
        .set_contents([Content::new()
            .set_role("user")
            .set_parts([Part::new().set_text(
                "What's a good name for a flower shop that specializes in selling bouquets of dried flowers?",
            )])])
        .send()
        .await;

And then print the response. You can use the :#? format specifier to prettify the nested response objects:

    println!("RESPONSE = {response:#?}");

Review the complete code.

Send a prompt and an image to the Vertex AI Gemini API

As in the previous example, initialize the client using the default settings:

    let client = PredictionService::builder().build().await?;

And then build the model name:

    const MODEL: &str = "gemini-2.5-flash";
    let model = format!("projects/{project_id}/locations/global/publishers/google/models/{MODEL}");

The new request includes an image part:

            Part::new().set_file_data(
                FileData::new()
                    .set_mime_type("image/jpeg")
                    .set_file_uri("gs://generativeai-downloads/images/scones.jpg"),
            ),

And the prompt part:

            Part::new().set_text("Describe this picture."),

Send the full request:

    let response = client
        .generate_content()
        .set_model(&model)
        .set_contents([Content::new().set_role("user").set_parts([
            Part::new().set_file_data(
                FileData::new()
                    .set_mime_type("image/jpeg")
                    .set_file_uri("gs://generativeai-downloads/images/scones.jpg"),
            ),
            Part::new().set_text("Describe this picture."),
        ])])
        .send()
        .await;

As in the previous example, print the full response:

    println!("RESPONSE = {response:#?}");

Review the complete code.

Text prompt: complete code

use google_cloud_aiplatform_v1::client::PredictionService;
use google_cloud_aiplatform_v1::model::{Content, Part};

pub async fn sample(project_id: &str) -> anyhow::Result<()> {
    let client = PredictionService::builder().build().await?;

    const MODEL: &str = "gemini-2.5-flash";
    let model = format!("projects/{project_id}/locations/global/publishers/google/models/{MODEL}");

    let response = client
        .generate_content()
        .set_model(&model)
        .set_contents([Content::new()
            .set_role("user")
            .set_parts([Part::new().set_text(
                "What's a good name for a flower shop that specializes in selling bouquets of dried flowers?",
            )])])
        .send()
        .await;
    println!("RESPONSE = {response:#?}");

    Ok(())
}

Prompt and image: complete code

use google_cloud_aiplatform_v1::client::PredictionService;
use google_cloud_aiplatform_v1::model::{Content, FileData, Part};

pub async fn sample(project_id: &str) -> anyhow::Result<()> {
    let client = PredictionService::builder().build().await?;

    const MODEL: &str = "gemini-2.5-flash";
    let model = format!("projects/{project_id}/locations/global/publishers/google/models/{MODEL}");

    let response = client
        .generate_content()
        .set_model(&model)
        .set_contents([Content::new().set_role("user").set_parts([
            Part::new().set_file_data(
                FileData::new()
                    .set_mime_type("image/jpeg")
                    .set_file_uri("gs://generativeai-downloads/images/scones.jpg"),
            ),
            Part::new().set_text("Describe this picture."),
        ])])
        .send()
        .await;
    println!("RESPONSE = {response:#?}");

    Ok(())
}

Keyboard shortcuts