Media resolution

The media_resolution parameter controls how the Gemini API processes media inputs like images, videos, and PDF documents by determining the maximum number of tokens allocated for media inputs, allowing you to balance response quality against latency and cost. For different settings, default values and how they correspond to tokens, see the Token counts section.

You can configure media resolution in two ways:

  • Per part (Gemini 3 only)

  • Globally for an entire generateContent request (all multimodal models)

Per-part media resolution (Gemini 3 only)

Gemini 3 allows you to set media resolution for individual media objects within your request, offering fine-grained optimisation of token usage. You can mix resolution levels in a single request. For example, using high resolution for a complex diagram and low resolution for a simple contextual image. This setting overrides any global configuration for a specific part. For default settings, see Token counts section.

Python

from google import genai
from google.genai import types

# The media_resolution parameter for parts is currently only available in the v1alpha API version. (experimental)
client = genai.Client(
  http_options={
      'api_version': 'v1alpha',
  }
)

# Replace with your image data
with open('path/to/image1.jpg', 'rb') as f:
    image_bytes_1 = f.read()

# Create parts with different resolutions
image_part_high = types.Part.from_bytes(
    data=image_bytes_1,
    mime_type='image/jpeg',
    media_resolution=types.MediaResolution.MEDIA_RESOLUTION_HIGH
)

model_name = 'gemini-3-pro-preview'

response = client.models.generate_content(
    model=model_name,
    contents=["Describe these images:", image_part_high]
)
print(response.text)

Javascript

// Example: Setting per-part media resolution in JavaScript
import { GoogleGenAI, MediaResolution, Part } from '@google/genai';
import * as fs from 'fs';
import { Buffer } from 'buffer'; // Node.js

const ai = new GoogleGenAI({ httpOptions: { apiVersion: 'v1alpha' } });

// Helper function to convert local file to a Part object
function fileToGenerativePart(path, mimeType, mediaResolution) {
    return {
        inlineData: { data: Buffer.from(fs.readFileSync(path)).toString('base64'), mimeType },
        mediaResolution: { 'level': mediaResolution }
    };
}

async function run() {
    // Create parts with different resolutions
    const imagePartHigh = fileToGenerativePart('img.png', 'image/png', Part.MediaResolutionLevel.MEDIA_RESOLUTION_HIGH);
    const model_name = 'gemini-3-pro-preview';
    const response = await ai.models.generateContent({
        model: model_name,
        contents: ['Describe these images:', imagePartHigh]
        // Global config can still be set, but per-part settings will override
        // config: {
        //   mediaResolution: MediaResolution.MEDIA_RESOLUTION_MEDIUM
        // }
    });
    console.log(response.text);
}
run();

REST

# Replace with paths to your images
IMAGE_PATH="path/to/image.jpg"

# Base64 encode the images
BASE64_IMAGE1=$(base64 -w 0 "$IMAGE_PATH")

MODEL_ID="gemini-3-pro-preview"

echo '{
    "contents": [{
      "parts": [
        {"text": "Describe these images:"},
        {
          "inline_data": {
            "mime_type": "image/jpeg",
            "data": "'"$BASE64_IMAGE1"'",
          },
          "media_resolution": {"level": "MEDIA_RESOLUTION_HIGH"}
        }
      ]
    }]
  }' > request.json

curl -s -X POST \
  "https://generativelanguage.googleapis.com/v1alpha/models/${MODEL_ID}:generateContent" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d @request.json

Global media resolution

You can set a default resolution for all media parts in a request using the GenerationConfig. This is supported by all multimodal models. If a request includes both global and per-part settings, the per-part setting takes precedence for that specific item.

Python

from google import genai
from google.genai import types

client = genai.Client()

# Prepare standard image part
with open('image.jpg', 'rb') as f:
    image_bytes = f.read()
image_part = types.Part.from_bytes(data=image_bytes, mime_type='image/jpeg')

# Set global configuration
config = types.GenerateContentConfig(
    media_resolution=types.MediaResolution.MEDIA_RESOLUTION_HIGH
)

response = client.models.generate_content(
    model='gemini-2.5-flash',
    contents=["Describe this image:", image_part],
    config=config
)
print(response.text)

Javascript

import { GoogleGenAI, MediaResolution } from '@google/genai';
import * as fs from 'fs';

const ai = new GoogleGenAI({ });

async function run() {
   // ... (Image loading logic) ...

   const response = await ai.models.generateContent({
      model: 'gemini-2.5-flash',
      contents: ["Describe this image:", imagePart],
      config: {
         mediaResolution: MediaResolution.MEDIA_RESOLUTION_HIGH
      }
   });
   console.log(response.text);
}
run();

REST

# ... (Base64 encoding logic) ...

curl -s -X POST \
  "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:generateContent" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [...],
    "generation_config": {
      "media_resolution": "MEDIA_RESOLUTION_HIGH"
    }
  }'

Available resolution values

The Gemini API defines the following levels for media resolution:

  • MEDIA_RESOLUTION_UNSPECIFIED: The default setting. The token count for this level varies significantly between Gemini 3 and earlier Gemini models.
  • MEDIA_RESOLUTION_LOW: Lower token count, resulting in faster processing and lower cost, but with less detail.
  • MEDIA_RESOLUTION_MEDIUM: A balance between detail, cost, and latency.
  • MEDIA_RESOLUTION_HIGH: Higher token count, providing more detail for the model to work with, at the expense of increased latency and cost.
  • (Coming soon) MEDIA_RESOLUTION_ULTRA_HIGH: Highest token count required for specific use cases such as computer use.

The exact number of tokens generated for each of these levels depends on both the media type (Image, Video, PDF) and the model version.

Token counts

The tables below summarize the approximate token counts for each media_resolution value and media type per model family.

Gemini 3 Models

MediaResolution Image Video PDF
MEDIA_RESOLUTION_UNSPECIFIED (Default) 1120 70 560
MEDIA_RESOLUTION_LOW 280 70 280 + Native Text
MEDIA_RESOLUTION_MEDIUM 560 70 560 + Native Text
MEDIA_RESOLUTION_HIGH 1120 280 1120 + Native Text

Gemini 2.5 models

MediaResolution Image Video PDF (Scanned) PDF (Native)
MEDIA_RESOLUTION_UNSPECIFIED (Default) 256 + Pan & Scan (~2048) 256 256 + OCR 256 + Native Text
MEDIA_RESOLUTION_LOW 64 64 64 + OCR 64 + Native Text
MEDIA_RESOLUTION_MEDIUM 256 256 256 + OCR 256 + Native Text
MEDIA_RESOLUTION_HIGH 256 + Pan & Scan 256 256 + OCR 256 + Native Text

Choosing the right resolution

  • Default (UNSPECIFIED): Start with the default. It's tuned for a good balance of quality, latency, and cost for most common use cases.
  • LOW: Use for scenarios where cost and latency are paramount, and fine-grained detail is less critical.
  • MEDIUM / HIGH: Increase the resolution when the task requires understanding intricate details within the media. This is often needed for complex visual analysis, chart reading, or dense document comprehension.
  • Per-part control (Gemini 3): Leverage this to optimize token usage. For example, in a prompt with multiple images, use HIGH for a complex diagram and LOW or MEDIUM for simpler contextual images.

Recommended settings

The following lists the recommended media resolution settings for each supported media type.

Media Type Recommended Setting Max Tokens Usage Guidance
Images MEDIA_RESOLUTION_HIGH 1120 Recommended for most image analysis tasks to ensure maximum quality.
PDFs MEDIA_RESOLUTION_MEDIUM 560 Optimal for document understanding; quality typically saturates at medium. Increasing to high rarely improves OCR results for standard documents.
Video (General) MEDIA_RESOLUTION_LOW (or MEDIA_RESOLUTION_MEDIUM) 70 (per frame) Note: For video, low and medium settings are treated identically (70 tokens) to optimize context usage. This is sufficient for most action recognition and description tasks.
Video (Text-heavy) MEDIA_RESOLUTION_HIGH 280 (per frame) Required only when the use case involves reading dense text (OCR) or small details within video frames.

Always test and evaluate the impact of different resolution settings on your specific application to find the best trade-off between quality, latency, and cost.

Version compatibility summary

  • The MediaResolution enum is available for all models supporting media input.
  • The token counts associated with each enum level differ between Gemini 3 models and earlier Gemini versions.
  • Setting media_resolution on individual Part objects is exclusive to Gemini 3 models.

Next steps