Multimodal embeddings API

import vertexai
from vertexai.vision_models import Image, MultiModalEmbeddingModel

# TODO(developer): Update & uncomment line below
# PROJECT_ID = "your-project-id"
vertexai.init(project=PROJECT_ID, location="us-central1")

model = MultiModalEmbeddingModel.from_pretrained("multimodalembedding@001")
image = Image.load_from_file(
    "gs://cloud-samples-data/vertex-ai/llm/prompts/landmark1.png"
)

embeddings = model.get_embeddings(
    image=image,
    contextual_text="Colosseum",
    dimension=1408,
)
print(f"Image Embedding: {embeddings.image_embedding}")
print(f"Text Embedding: {embeddings.text_embedding}")
# Example response:
# Image Embedding: [-0.0123147098, 0.0727171078, ...]
# Text Embedding: [0.00230263756, 0.0278981831, ...]

/**
 * TODO(developer): Uncomment these variables before running the sample.\
 * (Not necessary if passing values as arguments)
 */
// const project = 'YOUR_PROJECT_ID';
// const location = 'YOUR_PROJECT_LOCATION';
// const bastImagePath = "YOUR_BASED_IMAGE_PATH"
// const textPrompt = 'YOUR_TEXT_PROMPT';
const aiplatform = require('@google-cloud/aiplatform');

// Imports the Google Cloud Prediction service client
const {PredictionServiceClient} = aiplatform.v1;

// Import the helper module for converting arbitrary protobuf.Value objects.
const {helpers} = aiplatform;

// Specifies the location of the api endpoint
const clientOptions = {
  apiEndpoint: 'us-central1-aiplatform.googleapis.com',
};
const publisher = 'google';
const model = 'multimodalembedding@001';

// Instantiates a client
const predictionServiceClient = new PredictionServiceClient(clientOptions);

async function predictImageFromImageAndText() {
  // Configure the parent resource
  const endpoint = `projects/${project}/locations/${location}/publishers/${publisher}/models/${model}`;

  const fs = require('fs');
  const imageFile = fs.readFileSync(baseImagePath);

  // Convert the image data to a Buffer and base64 encode it.
  const encodedImage = Buffer.from(imageFile).toString('base64');

  const prompt = {
    text: textPrompt,
    image: {
      bytesBase64Encoded: encodedImage,
    },
  };
  const instanceValue = helpers.toValue(prompt);
  const instances = [instanceValue];

  const parameter = {
    sampleCount: 1,
  };
  const parameters = helpers.toValue(parameter);

  const request = {
    endpoint,
    instances,
    parameters,
  };

  // Predict request
  const [response] = await predictionServiceClient.predict(request);
  console.log('Get image embedding response');
  const predictions = response.predictions;
  console.log('\tPredictions :');
  for (const prediction of predictions) {
    console.log(`\t\tPrediction : ${JSON.stringify(prediction)}`);
  }
}

await predictImageFromImageAndText();

import com.google.cloud.aiplatform.v1beta1.EndpointName;
import com.google.cloud.aiplatform.v1beta1.PredictResponse;
import com.google.cloud.aiplatform.v1beta1.PredictionServiceClient;
import com.google.cloud.aiplatform.v1beta1.PredictionServiceSettings;
import com.google.gson.Gson;
import com.google.gson.JsonObject;
import com.google.protobuf.InvalidProtocolBufferException;
import com.google.protobuf.Value;
import com.google.protobuf.util.JsonFormat;
import java.io.IOException;
import java.nio.charset.StandardCharsets;
import java.nio.file.Files;
import java.nio.file.Paths;
import java.util.ArrayList;
import java.util.Base64;
import java.util.HashMap;
import java.util.List;
import java.util.Map;

public class PredictImageFromImageAndTextSample {

  public static void main(String[] args) throws IOException {
    // TODO(developer): Replace this variable before running the sample.
    String project = "YOUR_PROJECT_ID";
    String textPrompt = "YOUR_TEXT_PROMPT";
    String baseImagePath = "YOUR_BASE_IMAGE_PATH";

    // Learn how to use text prompts to update an image:
    // https://cloud.google.com/vertex-ai/docs/generative-ai/image/edit-images
    Map<String, Object> parameters = new HashMap<String, Object>();
    parameters.put("sampleCount", 1);

    String location = "us-central1";
    String publisher = "google";
    String model = "multimodalembedding@001";

    predictImageFromImageAndText(
        project, location, publisher, model, textPrompt, baseImagePath, parameters);
  }

  // Update images using text prompts
  public static void predictImageFromImageAndText(
      String project,
      String location,
      String publisher,
      String model,
      String textPrompt,
      String baseImagePath,
      Map<String, Object> parameters)
      throws IOException {
    final String endpoint = String.format("%s-aiplatform.googleapis.com:443", location);
    final PredictionServiceSettings predictionServiceSettings =
        PredictionServiceSettings.newBuilder().setEndpoint(endpoint).build();

    // Initialize client that will be used to send requests. This client only needs to be created
    // once, and can be reused for multiple requests.
    try (PredictionServiceClient predictionServiceClient =
        PredictionServiceClient.create(predictionServiceSettings)) {
      final EndpointName endpointName =
          EndpointName.ofProjectLocationPublisherModelName(project, location, publisher, model);

      // Convert the image to Base64
      byte[] imageData = Base64.getEncoder().encode(Files.readAllBytes(Paths.get(baseImagePath)));
      String encodedImage = new String(imageData, StandardCharsets.UTF_8);

      JsonObject jsonInstance = new JsonObject();
      jsonInstance.addProperty("text", textPrompt);
      JsonObject jsonImage = new JsonObject();
      jsonImage.addProperty("bytesBase64Encoded", encodedImage);
      jsonInstance.add("image", jsonImage);

      Value instanceValue = stringToValue(jsonInstance.toString());
      List<Value> instances = new ArrayList<>();
      instances.add(instanceValue);

      Gson gson = new Gson();
      String gsonString = gson.toJson(parameters);
      Value parameterValue = stringToValue(gsonString);

      PredictResponse predictResponse =
          predictionServiceClient.predict(endpointName, instances, parameterValue);
      System.out.println("Predict Response");
      System.out.println(predictResponse);
      for (Value prediction : predictResponse.getPredictionsList()) {
        System.out.format("\tPrediction: %s\n", prediction);
      }
    }
  }

  // Convert a Json string to a protobuf.Value
  static Value stringToValue(String value) throws InvalidProtocolBufferException {
    Value.Builder builder = Value.newBuilder();
    JsonFormat.parser().merge(value, builder);
    return builder.build();
  }
}

import (
	"context"
	"encoding/json"
	"fmt"
	"io"

	aiplatform "cloud.google.com/go/aiplatform/apiv1beta1"
	aiplatformpb "cloud.google.com/go/aiplatform/apiv1beta1/aiplatformpb"
	"google.golang.org/api/option"
	"google.golang.org/protobuf/encoding/protojson"
	"google.golang.org/protobuf/types/known/structpb"
)

// generateForTextAndImage shows how to use the multimodal model to generate embeddings for
// text and image inputs.
func generateForTextAndImage(w io.Writer, project, location string) error {
	// location = "us-central1"
	ctx := context.Background()
	apiEndpoint := fmt.Sprintf("%s-aiplatform.googleapis.com:443", location)
	client, err := aiplatform.NewPredictionClient(ctx, option.WithEndpoint(apiEndpoint))
	if err != nil {
		return fmt.Errorf("failed to construct API client: %w", err)
	}
	defer client.Close()

	model := "multimodalembedding@001"
	endpoint := fmt.Sprintf("projects/%s/locations/%s/publishers/google/models/%s", project, location, model)

	// This is the input to the model's prediction call. For schema, see:
	// https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/multimodal-embeddings-api#request_body
	instance, err := structpb.NewValue(map[string]any{
		"image": map[string]any{
			// Image input can be provided either as a Google Cloud Storage URI or as
			// base64-encoded bytes using the "bytesBase64Encoded" field.
			"gcsUri": "gs://cloud-samples-data/vertex-ai/llm/prompts/landmark1.png",
		},
		"text": "Colosseum",
	})
	if err != nil {
		return fmt.Errorf("failed to construct request payload: %w", err)
	}

	req := &aiplatformpb.PredictRequest{
		Endpoint: endpoint,
		// The model supports only 1 instance per request.
		Instances: []*structpb.Value{instance},
	}

	resp, err := client.Predict(ctx, req)
	if err != nil {
		return fmt.Errorf("failed to generate embeddings: %w", err)
	}

	instanceEmbeddingsJson, err := protojson.Marshal(resp.GetPredictions()[0])
	if err != nil {
		return fmt.Errorf("failed to convert protobuf value to JSON: %w", err)
	}
	// For response schema, see:
	// https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/multimodal-embeddings-api#response-body
	var instanceEmbeddings struct {
		ImageEmbeddings []float32 `json:"imageEmbedding"`
		TextEmbeddings  []float32 `json:"textEmbedding"`
	}
	if err := json.Unmarshal(instanceEmbeddingsJson, &instanceEmbeddings); err != nil {
		return fmt.Errorf("failed to unmarshal JSON: %w", err)
	}

	imageEmbedding := instanceEmbeddings.ImageEmbeddings
	textEmbedding := instanceEmbeddings.TextEmbeddings

	fmt.Fprintf(w, "Text embedding (length=%d): %v\n", len(textEmbedding), textEmbedding)
	fmt.Fprintf(w, "Image embedding (length=%d): %v\n", len(imageEmbedding), imageEmbedding)
	// Example response:
	// Text embedding (length=1408): [0.0023026613 0.027898183 -0.011858357 ... ]
	// Image embedding (length=1408): [-0.012314269 0.07271844 0.00020170923 ... ]

	return nil
}

import vertexai

from vertexai.vision_models import MultiModalEmbeddingModel, Video
from vertexai.vision_models import VideoSegmentConfig

# TODO(developer): Update & uncomment line below
# PROJECT_ID = "your-project-id"
vertexai.init(project=PROJECT_ID, location="us-central1")

model = MultiModalEmbeddingModel.from_pretrained("multimodalembedding@001")

embeddings = model.get_embeddings(
    video=Video.load_from_file(
        "gs://cloud-samples-data/vertex-ai-vision/highway_vehicles.mp4"
    ),
    video_segment_config=VideoSegmentConfig(end_offset_sec=1),
)

# Video Embeddings are segmented based on the video_segment_config.
print("Video Embeddings:")
for video_embedding in embeddings.video_embeddings:
    print(
        f"Video Segment: {video_embedding.start_offset_sec} - {video_embedding.end_offset_sec}"
    )
    print(f"Embedding: {video_embedding.embedding}")

# Example response:
# Video Embeddings:
# Video Segment: 0.0 - 1.0
# Embedding: [-0.0206376351, 0.0123456789, ...]

import (
	"context"
	"encoding/json"
	"fmt"
	"io"
	"time"

	aiplatform "cloud.google.com/go/aiplatform/apiv1beta1"
	aiplatformpb "cloud.google.com/go/aiplatform/apiv1beta1/aiplatformpb"
	"google.golang.org/api/option"
	"google.golang.org/protobuf/encoding/protojson"
	"google.golang.org/protobuf/types/known/structpb"
)

// generateForVideo shows how to use the multimodal model to generate embeddings for video input.
func generateForVideo(w io.Writer, project, location string) error {
	// location = "us-central1"

	// The default context timeout may be not enough to process a video input.
	ctx, cancel := context.WithTimeout(context.Background(), 15*time.Second)
	defer cancel()

	apiEndpoint := fmt.Sprintf("%s-aiplatform.googleapis.com:443", location)
	client, err := aiplatform.NewPredictionClient(ctx, option.WithEndpoint(apiEndpoint))
	if err != nil {
		return fmt.Errorf("failed to construct API client: %w", err)
	}
	defer client.Close()

	model := "multimodalembedding@001"
	endpoint := fmt.Sprintf("projects/%s/locations/%s/publishers/google/models/%s", project, location, model)

	// This is the input to the model's prediction call. For schema, see:
	// https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/multimodal-embeddings-api#request_body
	instances, err := structpb.NewValue(map[string]any{
		"video": map[string]any{
			// Video input can be provided either as a Google Cloud Storage URI or as base64-encoded
			// bytes using the "bytesBase64Encoded" field.
			"gcsUri": "gs://cloud-samples-data/vertex-ai-vision/highway_vehicles.mp4",
			"videoSegmentConfig": map[string]any{
				"startOffsetSec": 1,
				"endOffsetSec":   5,
			},
		},
	})
	if err != nil {
		return fmt.Errorf("failed to construct request payload: %w", err)
	}

	req := &aiplatformpb.PredictRequest{
		Endpoint: endpoint,
		// The model supports only 1 instance per request.
		Instances: []*structpb.Value{instances},
	}
	resp, err := client.Predict(ctx, req)
	if err != nil {
		return fmt.Errorf("failed to generate embeddings: %w", err)
	}

	instanceEmbeddingsJson, err := protojson.Marshal(resp.GetPredictions()[0])
	if err != nil {
		return fmt.Errorf("failed to convert protobuf value to JSON: %w", err)
	}
	// For response schema, see:
	// https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/multimodal-embeddings-api#response-body
	var instanceEmbeddings struct {
		VideoEmbeddings []struct {
			Embedding      []float32 `json:"embedding"`
			StartOffsetSec float64   `json:"startOffsetSec"`
			EndOffsetSec   float64   `json:"endOffsetSec"`
		} `json:"videoEmbeddings"`
	}
	if err := json.Unmarshal(instanceEmbeddingsJson, &instanceEmbeddings); err != nil {
		return fmt.Errorf("failed to unmarshal json: %w", err)
	}
	// Get the embedding for our single video segment (`.videoEmbeddings` object has one entry per
	// each processed segment).
	videoEmbedding := instanceEmbeddings.VideoEmbeddings[0]

	fmt.Fprintf(w, "Video embedding (seconds: %.f-%.f; length=%d): %v\n",
		videoEmbedding.StartOffsetSec,
		videoEmbedding.EndOffsetSec,
		len(videoEmbedding.Embedding),
		videoEmbedding.Embedding,
	)
	// Example response:
	// Video embedding (seconds: 1-5; length=1408): [-0.016427778 0.032878537 -0.030755188 ... ]

	return nil
}

import vertexai

from vertexai.vision_models import Image, MultiModalEmbeddingModel, Video
from vertexai.vision_models import VideoSegmentConfig

# TODO(developer): Update & uncomment line below
# PROJECT_ID = "your-project-id"
vertexai.init(project=PROJECT_ID, location="us-central1")

model = MultiModalEmbeddingModel.from_pretrained("multimodalembedding@001")

image = Image.load_from_file(
    "gs://cloud-samples-data/vertex-ai/llm/prompts/landmark1.png"
)
video = Video.load_from_file(
    "gs://cloud-samples-data/vertex-ai-vision/highway_vehicles.mp4"
)

embeddings = model.get_embeddings(
    image=image,
    video=video,
    video_segment_config=VideoSegmentConfig(end_offset_sec=1),
    contextual_text="Cars on Highway",
)

print(f"Image Embedding: {embeddings.image_embedding}")

# Video Embeddings are segmented based on the video_segment_config.
print("Video Embeddings:")
for video_embedding in embeddings.video_embeddings:
    print(
        f"Video Segment: {video_embedding.start_offset_sec} - {video_embedding.end_offset_sec}"
    )
    print(f"Embedding: {video_embedding.embedding}")

print(f"Text Embedding: {embeddings.text_embedding}")
# Example response:
# Image Embedding: [-0.0123144267, 0.0727186054, 0.000201397663, ...]
# Video Embeddings:
# Video Segment: 0.0 - 1.0
# Embedding: [-0.0206376351, 0.0345234685, ...]
# Text Embedding: [-0.0207006838, -0.00251058186, ...]

import (
	"context"
	"encoding/json"
	"fmt"
	"io"
	"time"

	aiplatform "cloud.google.com/go/aiplatform/apiv1beta1"
	aiplatformpb "cloud.google.com/go/aiplatform/apiv1beta1/aiplatformpb"
	"google.golang.org/api/option"
	"google.golang.org/protobuf/encoding/protojson"
	"google.golang.org/protobuf/types/known/structpb"
)

// generateForImageTextAndVideo shows how to use the multimodal model to generate embeddings for
// image, text and video data.
func generateForImageTextAndVideo(w io.Writer, project, location string) error {
	// location = "us-central1"

	// The default context timeout may be not enough to process a video input.
	ctx, cancel := context.WithTimeout(context.Background(), 15*time.Second)
	defer cancel()

	apiEndpoint := fmt.Sprintf("%s-aiplatform.googleapis.com:443", location)
	client, err := aiplatform.NewPredictionClient(ctx, option.WithEndpoint(apiEndpoint))
	if err != nil {
		return fmt.Errorf("failed to construct API client: %w", err)
	}
	defer client.Close()

	model := "multimodalembedding@001"
	endpoint := fmt.Sprintf("projects/%s/locations/%s/publishers/google/models/%s", project, location, model)

	// This is the input to the model's prediction call. For schema, see:
	// https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/multimodal-embeddings-api#request_body
	instance, err := structpb.NewValue(map[string]any{
		"text": "Domestic cats in natural conditions",
		"image": map[string]any{
			// Image and video inputs can be provided either as a Google Cloud Storage URI or as
			// base64-encoded bytes using the "bytesBase64Encoded" field.
			"gcsUri": "gs://cloud-samples-data/generative-ai/image/320px-Felis_catus-cat_on_snow.jpg",
		},
		"video": map[string]any{
			"gcsUri": "gs://cloud-samples-data/video/cat.mp4",
		},
	})
	if err != nil {
		return fmt.Errorf("failed to construct request payload: %w", err)
	}

	req := &aiplatformpb.PredictRequest{
		Endpoint: endpoint,
		// The model supports only 1 instance per request.
		Instances: []*structpb.Value{instance},
	}

	resp, err := client.Predict(ctx, req)
	if err != nil {
		return fmt.Errorf("failed to generate embeddings: %w", err)
	}

	instanceEmbeddingsJson, err := protojson.Marshal(resp.GetPredictions()[0])
	if err != nil {
		return fmt.Errorf("failed to convert protobuf value to JSON: %w", err)
	}
	// For response schema, see:
	// https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/multimodal-embeddings-api#response-body
	var instanceEmbeddings struct {
		ImageEmbeddings []float32 `json:"imageEmbedding"`
		TextEmbeddings  []float32 `json:"textEmbedding"`
		VideoEmbeddings []struct {
			Embedding      []float32 `json:"embedding"`
			StartOffsetSec float64   `json:"startOffsetSec"`
			EndOffsetSec   float64   `json:"endOffsetSec"`
		} `json:"videoEmbeddings"`
	}
	if err := json.Unmarshal(instanceEmbeddingsJson, &instanceEmbeddings); err != nil {
		return fmt.Errorf("failed to unmarshal JSON: %w", err)
	}

	imageEmbedding := instanceEmbeddings.ImageEmbeddings
	textEmbedding := instanceEmbeddings.TextEmbeddings
	// Get the embedding for our single video segment (`.videoEmbeddings` object has one entry per
	// each processed segment).
	videoEmbedding := instanceEmbeddings.VideoEmbeddings[0].Embedding

	fmt.Fprintf(w, "Image embedding (length=%d): %v\n", len(imageEmbedding), imageEmbedding)
	fmt.Fprintf(w, "Text embedding (length=%d): %v\n", len(textEmbedding), textEmbedding)
	fmt.Fprintf(w, "Video embedding (length=%d): %v\n", len(videoEmbedding), videoEmbedding)
	// Example response:
	// Image embedding (length=1408): [-0.01558477 0.0258355 0.016342038 ... ]
	// Text embedding (length=1408): [-0.005894961 0.008349559 0.015355394 ... ]
	// Video embedding (length=1408): [-0.018867437 0.013997682 0.0012682161 ... ]

	return nil
}