Document understanding

You can add PDFs to Gemini requests to perform tasks that involve understanding the contents of the included PDFs. This page shows you how to add PDFs to your requests to Gemini in Vertex AI by using the Google Cloud console and the Vertex AI API.

Supported models

The following table lists the models that support document understanding:

Model	PDF modality details
Gemini 1.5 Flash Go to the Gemini 1.5 Flash model card	Maximum pages per prompt: 300 Maximum PDF file size: 30 MB
Gemini 1.5 Pro Go to the Gemini 1.5 Pro model card	Maximum pages per prompt: 300 Maximum PDF file size: 30 MB
Gemini 1.0 Pro Vision Go to the Gemini 1.0 Pro Vision model card	Maximum pages per prompt: 16 Maximum PDF file size: 30 MB

For a list of languages supported by Gemini models, see model information Google models. To learn more about how to design multimodal prompts, see Design multimodal prompts. If you're looking for a way to use Gemini directly from your mobile and web apps, see the Google AI SDKs for Android, Swift, and web.

Add documents to a request

You can add a single PDF in your requests to Gemini.

Single PDF

The following tab shows you how to include a PDF in a prompt request using the Python SDK. This PDF sample works with all Gemini multimodal models.

Python

To learn how to install or update the Vertex AI SDK for Python, see Install the Vertex AI SDK for Python. For more information, see the Vertex AI SDK for Python API reference documentation.

Streaming and non-streaming responses

You can choose whether the model generates streaming responses or non-streaming responses. For streaming responses, you receive each response as soon as its output token is generated. For non-streaming responses, you receive all responses after all of the output tokens are generated.

For a streaming response, use the stream parameter in generate_content.

  response = model.generate_content(contents=[...], stream = True)
  

For a non-streaming response, remove the parameter, or set the parameter to False.

Sample code

import vertexai

from vertexai.generative_models import GenerativeModel, Part

# TODO(developer): Update and un-comment below lines
# project_id = "PROJECT_ID"

vertexai.init(project=project_id, location="us-central1")

model = GenerativeModel(model_name="gemini-1.5-flash-001")

prompt = """
You are a very professional document summarization specialist.
Please summarize the given document.
"""

pdf_file_uri = "gs://cloud-samples-data/generative-ai/pdf/2403.05530.pdf"
pdf_file = Part.from_uri(pdf_file_uri, mime_type="application/pdf")
contents = [pdf_file, prompt]

response = model.generate_content(contents)
print(response.text)

Java

Before trying this sample, follow the Java setup instructions in the Vertex AI quickstart. For more information, see the Vertex AI Java SDK for Gemini reference documentation.

To authenticate to Vertex AI, set up Application Default Credentials. For more information, see Set up authentication for a local development environment.

Streaming and non-streaming responses

You can choose whether the model generates streaming responses or non-streaming responses. For streaming responses, you receive each response as soon as its output token is generated. For non-streaming responses, you receive all responses after all of the output tokens are generated.

For a streaming response, use the generateContentStream method.

  public ResponseStream generateContentStream(Content content)
  

For a non-streaming response, use the generateContent method.

  public GenerateContentResponse generateContent(Content content)
  

Sample code

import com.google.cloud.vertexai.VertexAI;
import com.google.cloud.vertexai.api.GenerateContentResponse;
import com.google.cloud.vertexai.generativeai.ContentMaker;
import com.google.cloud.vertexai.generativeai.GenerativeModel;
import com.google.cloud.vertexai.generativeai.PartMaker;
import com.google.cloud.vertexai.generativeai.ResponseHandler;
import java.io.IOException;

public class PdfInput {

  public static void main(String[] args) throws IOException {
    // TODO(developer): Replace these variables before running the sample.
    String projectId = "your-google-cloud-project-id";
    String location = "us-central1";
    String modelName = "gemini-1.5-flash-001";

    pdfInput(projectId, location, modelName);
  }

  // Analyzes the given video input.
  public static String pdfInput(String projectId, String location, String modelName)
      throws IOException {
    // 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 (VertexAI vertexAI = new VertexAI(projectId, location)) {
      String pdfUri = "gs://cloud-samples-data/generative-ai/pdf/2403.05530.pdf";

      GenerativeModel model = new GenerativeModel(modelName, vertexAI);
      GenerateContentResponse response = model.generateContent(
          ContentMaker.fromMultiModalData(
              "You are a very professional document summarization specialist.\n"
                  + "Please summarize the given document.",
              PartMaker.fromMimeTypeAndData("application/pdf", pdfUri)
          ));

      String output = ResponseHandler.getText(response);
      System.out.println(output);
      return output;
    }
  }
}

Node.js

Before trying this sample, follow the Node.js setup instructions in the Generative AI quickstart using the Node.js SDK. For more information, see the Node.js SDK for Gemini reference documentation.

To authenticate to Vertex AI, set up Application Default Credentials. For more information, see Set up authentication for a local development environment.

Streaming and non-streaming responses

You can choose whether the model generates streaming responses or non-streaming responses. For streaming responses, you receive each response as soon as its output token is generated. For non-streaming responses, you receive all responses after all of the output tokens are generated.

For a streaming response, use the generateContentStream method.

  const streamingResp = await generativeModel.generateContentStream(request);
  

For a non-streaming response, use the generateContent method.

  const streamingResp = await generativeModel.generateContent(request);
  

Sample code

const {VertexAI} = require('@google-cloud/vertexai');

/**
 * TODO(developer): Update these variables before running the sample.
 */
async function analyze_pdf(projectId = 'PROJECT_ID') {
  const vertexAI = new VertexAI({project: projectId, location: 'us-central1'});

  const generativeModel = vertexAI.getGenerativeModel({
    model: 'gemini-1.5-flash-001',
  });

  const filePart = {
    file_data: {
      file_uri: 'gs://cloud-samples-data/generative-ai/pdf/2403.05530.pdf',
      mime_type: 'application/pdf',
    },
  };
  const textPart = {
    text: `
    You are a very professional document summarization specialist.
    Please summarize the given document.`,
  };

  const request = {
    contents: [{role: 'user', parts: [filePart, textPart]}],
  };

  const resp = await generativeModel.generateContent(request);
  const contentResponse = await resp.response;
  console.log(JSON.stringify(contentResponse));
}

Go

Before trying this sample, follow the Go setup instructions in the Vertex AI quickstart. For more information, see the Vertex AI Go SDK for Gemini reference documentation.

To authenticate to Vertex AI, set up Application Default Credentials. For more information, see Set up authentication for a local development environment.

Streaming and non-streaming responses

You can choose whether the model generates streaming responses or non-streaming responses. For streaming responses, you receive each response as soon as its output token is generated. For non-streaming responses, you receive all responses after all of the output tokens are generated.

For a streaming response, use the GenerateContentStream method.

  iter := model.GenerateContentStream(ctx, genai.Text("Tell me a story about a lumberjack and his giant ox. Keep it very short."))
  

For a non-streaming response, use the GenerateContent method.

  resp, err := model.GenerateContent(ctx, genai.Text("What is the average size of a swallow?"))
  

Sample code

import (
	"context"
	"errors"
	"fmt"
	"io"

	"cloud.google.com/go/vertexai/genai"
)

// pdfPrompt is a sample prompt type consisting of one PDF asset, and a text question.
type pdfPrompt struct {
	// pdfPath is a Google Cloud Storage path starting with "gs://"
	pdfPath string
	// question asked to the model
	question string
}

// generateContentFromPDF generates a response into the provided io.Writer, based upon the PDF
// asset and the question provided in the multimodal prompt.
func generateContentFromPDF(w io.Writer, prompt pdfPrompt, projectID, location, modelName string) error {
	// prompt := pdfPrompt{
	// 	pdfPath: "gs://cloud-samples-data/generative-ai/pdf/2403.05530.pdf",
	// 	question: `
	// 		You are a very professional document summarization specialist.
	// 		Please summarize the given document.
	// 	`,
	// }
	// location := "us-central1"
	// modelName := "gemini-1.5-flash-001"
	ctx := context.Background()

	client, err := genai.NewClient(ctx, projectID, location)
	if err != nil {
		return fmt.Errorf("unable to create client: %w", err)
	}
	defer client.Close()

	model := client.GenerativeModel(modelName)

	part := genai.FileData{
		MIMEType: "application/pdf",
		FileURI:  prompt.pdfPath,
	}

	res, err := model.GenerateContent(ctx, part, genai.Text(prompt.question))
	if err != nil {
		return fmt.Errorf("unable to generate contents: %w", err)
	}

	if len(res.Candidates) == 0 ||
		len(res.Candidates[0].Content.Parts) == 0 {
		return errors.New("empty response from model")
	}

	fmt.Fprintf(w, "generated response: %s\n", res.Candidates[0].Content.Parts[0])
	return nil
}

C#

Before trying this sample, follow the C# setup instructions in the Vertex AI quickstart. For more information, see the Vertex AI C# reference documentation.

To authenticate to Vertex AI, set up Application Default Credentials. For more information, see Set up authentication for a local development environment.

Streaming and non-streaming responses

You can choose whether the model generates streaming responses or non-streaming responses. For streaming responses, you receive each response as soon as its output token is generated. For non-streaming responses, you receive all responses after all of the output tokens are generated.

For a streaming response, use the StreamGenerateContent method.

  public virtual PredictionServiceClient.StreamGenerateContentStream StreamGenerateContent(GenerateContentRequest request)
  

For a non-streaming response, use the GenerateContentAsync method.

  public virtual Task<GenerateContentResponse> GenerateContentAsync(GenerateContentRequest request)
  

For more information on how the server can stream responses, see Streaming RPCs.

Sample code

using Google.Cloud.AIPlatform.V1;
using System;
using System.Threading.Tasks;

public class PdfInput
{
    public async Task<string> SummarizePdf(
        string projectId = "your-project-id",
        string location = "us-central1",
        string publisher = "google",
        string model = "gemini-1.5-flash-001")
    {

        var predictionServiceClient = new PredictionServiceClientBuilder
        {
            Endpoint = $"{location}-aiplatform.googleapis.com"
        }.Build();

        string prompt = @"You are a very professional document summarization specialist.
Please summarize the given document.";

        var generateContentRequest = new GenerateContentRequest
        {
            Model = $"projects/{projectId}/locations/{location}/publishers/{publisher}/models/{model}",
            Contents =
            {
                new Content
                {
                    Role = "USER",
                    Parts =
                    {
                        new Part { Text = prompt },
                        new Part { FileData = new() { MimeType = "application/pdf", FileUri = "gs://cloud-samples-data/generative-ai/pdf/2403.05530.pdf" }}
                    }
                }
            }
        };

        GenerateContentResponse response = await predictionServiceClient.GenerateContentAsync(generateContentRequest);

        string responseText = response.Candidates[0].Content.Parts[0].Text;
        Console.WriteLine(responseText);

        return responseText;
    }
}

REST

After you set up your environment, you can use REST to test a text prompt. The following sample sends a request to the publisher model endpoint.

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

• LOCATION: The region to process the request. Enter a supported region. For the full list of supported regions, see Available locations.

  Click to expand a partial list of available regions

  • us-central1
  • us-west4
  • northamerica-northeast1
  • us-east4
  • us-west1
  • asia-northeast3
  • asia-southeast1
  • asia-northeast1
• PROJECT_ID: Your project ID.
• FILE_URI: The Cloud Storage URI of the file to include in the prompt. The bucket object must either be publicly readable or reside in the same Google Cloud project that's sending the request. You must also specify the media type (mimeType) of the file.

  If you don't have a PDF file in Cloud Storage, then you can use the following publicly available file: gs://cloud-samples-data/generative-ai/pdf/2403.05530.pdf with a mime type of application/pdf. To view this PDF, open the sample PDF file.

• MIME_TYPE: The media type of the file specified in the data or fileUri fields. Acceptable values include the following:

  Click to expand MIME types

  • application/pdf
  • audio/mpeg
  • audio/mp3
  • audio/wav
  • image/png
  • image/jpeg
  • text/plain
  • video/mov
  • video/mpeg
  • video/mp4
  • video/mpg
  • video/avi
  • video/wmv
  • video/mpegps
  • video/flv
• TEXT: The text instructions to include in the prompt. For example, You are a very professional document summarization specialist. Please summarize the given document.

To send your request, choose one of these options:

curl

Note: The following command assumes that you have logged in to the gcloud CLI with your user account by running gcloud init or gcloud auth login , or by using Cloud Shell, which automatically logs you into the gcloud CLI . You can check the currently active account by running gcloud auth list.

Save the request body in a file named request.json. Run the following command in the terminal to create or overwrite this file in the current directory:

cat > request.json << 'EOF'
{
  "contents": {
    "role": "USER",
    "parts": [
      {
        "fileData": {
          "fileUri": "FILE_URI",
          "mimeType": "MIME_TYPE"
        }
      },
      {
        "text": "TEXT"
      }
    ]
  }
}
EOF

Then execute the following command to send your REST request:

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/gemini-1.5-flash:generateContent"

PowerShell

Note: The following command assumes that you have logged in to the gcloud CLI with your user account by running gcloud init or gcloud auth login . You can check the currently active account by running gcloud auth list.

Save the request body in a file named request.json. Run the following command in the terminal to create or overwrite this file in the current directory:

@'
{
  "contents": {
    "role": "USER",
    "parts": [
      {
        "fileData": {
          "fileUri": "FILE_URI",
          "mimeType": "MIME_TYPE"
        }
      },
      {
        "text": "TEXT"
      }
    ]
  }
}
'@  | Out-File -FilePath request.json -Encoding utf8

Then execute the following command to send your REST request:

$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/gemini-1.5-flash:generateContent" | Select-Object -Expand Content

You should receive a JSON response similar to the following.

Response

{
  "candidates": [
    {
      "content": {
        "role": "model",
        "parts": [
          {
            "text": "This report presents Gemini 1.5 Pro, the first model release in the Gemini 1.5
              family, a novel mixture-of-experts multimodal model capable of recalling and reasoning
              over extremely long contexts of information, up to 10 million tokens. This surpasses
              existing models, which are typically constrained to 200,000 tokens.\n\nGemini 1.5 Pro
              is a highly compute-efficient model that improves on Gemini 1.0 Pro's performance
              across a range of benchmarks, even surpassing Gemini 1.0 Ultra in many categories
              while requiring less training compute. In particular, the model excels in handling
              long-context retrieval tasks, achieving near-perfect recall for text, audio and video,
              and even demonstrating in-context learning capabilities by learning to translate a new
              language from just one book.\n\nThe report discusses the new long-context capabilities
              of Gemini 1.5 Pro, including its novel architecture and training infrastructure, and
              showcases qualitative examples of the model's ability to handle long, mixed-modality
              inputs. It then explores quantitative evaluations of the model's performance in
              several categories, including perplexity over long sequences, needle-in-a-haystack
              retrieval tasks, and realistic multimodal benchmarks like long-document QA and
              long-context audio understanding. The report also addresses the important aspects of
              responsible deployment, outlining the model's impact assessment, evaluation approach,
              and mitigation efforts.\n\nOverall, Gemini 1.5 Pro represents a significant
              advancement in multimodal language modeling, pushing the boundaries of long-context
              understanding and showcasing the potential for large models to handle complex,
              mixed-modality information at scale. \n"
          }
        ]
      },
      "finishReason": "STOP",
      "safetyRatings": [
        {
          "category": "HARM_CATEGORY_HATE_SPEECH",
          "probability": "NEGLIGIBLE",
          "probabilityScore": 0.13273923,
          "severity": "HARM_SEVERITY_NEGLIGIBLE",
          "severityScore": 0.08819004
        },
        {
          "category": "HARM_CATEGORY_DANGEROUS_CONTENT",
          "probability": "NEGLIGIBLE",
          "probabilityScore": 0.1046602,
          "severity": "HARM_SEVERITY_NEGLIGIBLE",
          "severityScore": 0.0996453
        },
        {
          "category": "HARM_CATEGORY_HARASSMENT",
          "probability": "NEGLIGIBLE",
          "probabilityScore": 0.15987214,
          "severity": "HARM_SEVERITY_NEGLIGIBLE",
          "severityScore": 0.098946586
        },
        {
          "category": "HARM_CATEGORY_SEXUALLY_EXPLICIT",
          "probability": "NEGLIGIBLE",
          "probabilityScore": 0.056966383,
          "severity": "HARM_SEVERITY_NEGLIGIBLE",
          "severityScore": 0.075721376
        }
      ]
    }
  ],
  "usageMetadata": {
    "promptTokenCount": 19882,
    "candidatesTokenCount": 336,
    "totalTokenCount": 20218
  }
}

Note the following in the URL for this sample:

• Use the generateContent method to request that the response is returned after it's fully generated. To reduce the perception of latency to a human audience, stream the response as it's being generated by using the streamGenerateContent method.
• The multimodal model ID is located at the end of the URL before the method (for example, gemini-1.5-flash or gemini-1.0-pro-vision). This sample may support other models as well.

Console

To send a multimodal prompt by using the Google Cloud console, do the following:

1. In the Vertex AI section of the Google Cloud console, go to the Vertex AI Studio page.

   Go to Vertex AI Studio

2. Under Prompt design (single turn), click Open.

3. Configure the model and parameters:

   • Model: Select a model.
   • Region: Select the region that you want to use.

   • Temperature: Use the slider or textbox to enter a value for temperature.

     The temperature is used for sampling during response generation, which occurs when topP and topK are applied. Temperature controls the degree of randomness in token selection. Lower temperatures are good for prompts that require a less open-ended or creative response, while higher temperatures can lead to more diverse or creative results. A temperature of 0 means that the highest probability tokens are always selected. In this case, responses for a given prompt are mostly deterministic, but a small amount of variation is still possible.

     If the model returns a response that's too generic, too short, or the model gives a fallback response, try increasing the temperature.

   • Token limit: Use the slider or textbox to enter a value for the max output limit.

     Maximum number of tokens that can be generated in the response. A token is approximately four characters. 100 tokens correspond to roughly 60-80 words.

     Specify a lower value for shorter responses and a higher value for potentially longer responses.

   • Add stop sequence (optional): Enter a stop sequence, which is a series of characters (including spaces) that stops response generation if the model encounters it. The sequence is not included as part of the response. You can add up to five stop sequences.
4. Optional: To configure advanced parameters, click Advanced and configure as follows:

Click to expand advanced configurations

• Top-K: Use the slider or textbox to enter a value for top-K. (not supported for Gemini 1.5).

  Top-K changes how the model selects tokens for output. A top-K of 1 means the next selected token is the most probable among all tokens in the model's vocabulary (also called greedy decoding), while a top-K of 3 means that the next token is selected from among the three most probable tokens by using temperature.

  For each token selection step, the top-K tokens with the highest probabilities are sampled. Then tokens are further filtered based on top-P with the final token selected using temperature sampling.

  Specify a lower value for less random responses and a higher value for more random responses.

• Top-P: Use the slider or textbox to enter a value for top-P. Tokens are selected from most probable to the least until the sum of their probabilities equals the value of top-P. For the least variable results, set top-P to 0.
• Enable Grounding: Grounding isn't supported for multimodal prompts.
5. To upload media, such as a PDF file, do the following:
1. Click Insert Media, and select a source. If you choose Google Drive as your source, you must choose an account and give consent to Vertex AI Studio to access your account the first time you select this option. You can upload multiple images that have a total size of up to 10 MB. A single file can't exceed 7 MB.
2. Click the file that you want to add.
3. Click Select. The file thumbnail displays in the Prompt pane.
6. Enter your text prompt in the Prompt pane. The model uses previous messages as context for new responses.
7. Click Submit, and the response is generated.
8. Optional: To save your prompt to My prompts, click Save.
9. Optional: To get the Python code or a curl command for your prompt, click Get code.
10. Optional: To clear all previous messages, click Clear conversation.

Set model parameters

The following model parameters can be set on multimodal models:

Top-P

Top-P changes how the model selects tokens for output. Tokens are selected from the most (see top-K) to least probable until the sum of their probabilities equals the top-P value. For example, if tokens A, B, and C have a probability of 0.3, 0.2, and 0.1 and the top-P value is 0.5, then the model will select either A or B as the next token by using temperature and excludes C as a candidate.

Specify a lower value for less random responses and a higher value for more random responses.

Top-K

Top-K changes how the model selects tokens for output. A top-K of 1 means the next selected token is the most probable among all tokens in the model's vocabulary (also called greedy decoding), while a top-K of 3 means that the next token is selected from among the three most probable tokens by using temperature.

For each token selection step, the top-K tokens with the highest probabilities are sampled. Then tokens are further filtered based on top-P with the final token selected using temperature sampling.

Specify a lower value for less random responses and a higher value for more random responses.

Temperature

The temperature is used for sampling during response generation, which occurs when topP and topK are applied. Temperature controls the degree of randomness in token selection. Lower temperatures are good for prompts that require a less open-ended or creative response, while higher temperatures can lead to more diverse or creative results. A temperature of 0 means that the highest probability tokens are always selected. In this case, responses for a given prompt are mostly deterministic, but a small amount of variation is still possible.

If the model returns a response that's too generic, too short, or the model gives a fallback response, try increasing the temperature.

Valid parameter values

Parameter	Gemini 1.0 Pro Vision	Gemini 1.5 Pro	Gemini 1.5 Flash
Top-K	1 - 40 (default 32)	Not supported	Not supported
Top-P	0 - 1.0 (default 1.0)	0 - 1.0 (default 0.95)	0 - 1.0 (default 0.95)
Temperature	0 - 1.0 (default 0.4)	0 - 2.0 (default 1.0)	0 - 2.0 (default 1.0)

Document requirements

The required MIME type for a PDF is application/pdf.

PDF best practices

When using PDFs, use the following best practices and information for the best results:

• PDFs are treated as images, so a single page of a PDF is treated as one image.
  • The number of pages supported is limited to the number of images a model can support. For Gemini 1.0 Pro Vision, the limit is 16. For Gemini 1.5 Pro and Gemini 1.5 Flash the limit is 300. If you have a long document, consider splitting it into multiple PDFs to process it.
  • When using PDFs as an input, the cost follows Gemini image pricing. For example, if you include a two page PDF in a Gemini API call, you incur an input fee of processing two images.
• If your prompt contains a single PDF, place the PDF before the text prompt.
• Use PDFs created with text rendered as text instead of using text in scanned images. This format ensures text is machine-readable so that it's easier for the model to edit, search, and manipulate compared to scanned image PDFs. This practice provides optimal results when working with text-heavy documents like contracts.

Limitations

While Gemini multimodal models are powerful in many multimodal use cases, it's important to understand the limitations of the models:

• Spatial reasoning: The models aren't precise at locating text or objects in PDFs. They might only return the approximated counts of objects.
• Accuracy: The models might hallucinate when interpreting handwritten text in PDF documents.

What's next

• Start building with Gemini multimodal models - new customers get $300 in free Google Cloud credits to explore what they can do with Gemini.
• Learn how to send chat prompt requests.
• Learn about responsible AI best practices and Vertex AI's safety filters.