Google Cloud Virtual Try-On API - Comprehensive Documentation

Overview

Virtual Try-On is a Google Cloud Vertex AI service that generates realistic images of people wearing clothing products. It leverages the Imagen image generation model (virtual-try-on-001) to create synthetic try-on images from two inputs: a person image and a product image.

Key Purpose

• Enable e-commerce platforms to generate virtual try-on images automatically
• Create personalized product previews without physical samples
• Improve customer experience for fashion and apparel retail

Core Capabilities

Primary Features

• Virtual Try-On Image Generation: Generate images of people wearing provided clothing products
• Multiple Output Support: Generate 1-4 images per request
• Digital Watermarking: Apply verification watermarks to generated images
• Content Credentials (C2PA): Embed digital provenance information in images
• User-Configurable Safety Settings: Control content filters and safety parameters
• Person Generation: Support for synthetic person images in try-on scenarios
• Provisioned Throughput: Support for guaranteed capacity reservations

Limitations (Not Supported)

• Generic image generation (use Imagen instead)
• Custom model training
• Batch processing without API calls

Model Details: virtual-try-on-001

Model Information

• Model ID: virtual-try-on-001
• Service: Generative AI on Vertex AI
• Published By: Google

Supported Input/Output

| Type | Details | |——|———| | Inputs | Images (PNG, JPEG) | | Outputs | Images (PNG, JPEG) |

Technical Specifications

Image Constraints

| Specification | Details | |—|—| | Maximum Image Size | 10 MB | | Maximum Output Images | 4 images per request | | Supported Aspect Ratios | Same as input image | | Supported Resolutions | Same as input image | | Supported MIME Types | image/png, image/jpeg |

Performance & Quotas

| Metric | Value | |——–|——-| | Regional Online Prediction Requests | 50 per minute (per region, per base model) | | Prompt Languages | English |

API Endpoints & Parameters

REST API Endpoint

POST https://REGION-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/REGION/publishers/google/models/virtual-try-on-001:predict

Required Parameters

• REGION: The GCP region where your project is located (must be a supported region)
• PROJECT_ID: Your Google Cloud project ID
• BASE64_PERSON_IMAGE: Base64-encoded image of the person
• BASE64_PRODUCT_IMAGE: Base64-encoded image of the product
• IMAGE_COUNT: Number of output images (range: 1-4)
• GCS_OUTPUT_PATH: Cloud Storage path for output images

Request Structure

Request JSON Body

{
  "instances": [
    {
      "personImage": {
        "image": {
          "bytesBase64Encoded": "BASE64_PERSON_IMAGE"
        }
      },
      "productImages": [
        {
          "image": {
            "bytesBase64Encoded": "BASE64_PRODUCT_IMAGE"
          }
        }
      ]
    }
  ],
  "parameters": {
    "sampleCount": IMAGE_COUNT,
    "storageUri": "GCS_OUTPUT_PATH"
  }
}

Response Structure

Response JSON

{
  "predictions": [
    {
      "mimeType": "image/png",
      "bytesBase64Encoded": "BASE64_IMG_BYTES"
    },
    {
      "mimeType": "image/png",
      "bytesBase64Encoded": "BASE64_IMG_BYTES"
    }
  ]
}

Authentication & Setup

Prerequisites

1. Google Cloud Account
   • Create account or sign in to existing account
   • Free tier: $300 in credits for new customers
2. Project Setup
   • Select or create a Google Cloud project
   • Required role: roles/resourcemanager.projectCreator
3. Enable Vertex AI API
   • Required role: roles/serviceusage.serviceUsageAdmin
   • Enable API
4. Billing
   • Verify billing is enabled on your project

Authentication Methods

Python SDK

pip install --upgrade google-genai

Set environment variables:

export GOOGLE_CLOUD_PROJECT=YOUR_PROJECT_ID
export GOOGLE_CLOUD_LOCATION=global
export GOOGLE_GENAI_USE_VERTEXAI=True

Initialize gcloud CLI:

gcloud auth application-default login

REST API

Obtain access token:

gcloud auth print-access-token

Code Examples

Python Example

import base64
import os
from google.genai import client

# Initialize client
genai = client.Client(
    api_key=os.environ.get("GOOGLE_API_KEY")
)

# Load images
with open("person.png", "rb") as f:
    person_image = base64.standard_b64encode(f.read()).decode()

with open("product.png", "rb") as f:
    product_image = base64.standard_b64encode(f.read()).decode()

# Generate virtual try-on
response = genai.models.predict(
    model="virtual-try-on-001",
    instances=[
        {
            "personImage": {
                "image": {
                    "bytesBase64Encoded": person_image
                }
            },
            "productImages": [
                {
                    "image": {
                        "bytesBase64Encoded": product_image
                    }
                }
            ]
        }
    ],
    parameters={
        "sampleCount": 2,
        "storageUri": "gs://my-bucket/output/"
    }
)

# Process results
for prediction in response.predictions:
    print(f"Generated image: {prediction['mimeType']}")

REST API Example (curl)

curl -X POST \
  -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json; charset=utf-8" \
  -d @request.json \
  "https://us-central1-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/us-central1/publishers/google/models/virtual-try-on-001:predict"

request.json:

{
  "instances": [
    {
      "personImage": {
        "image": {
          "bytesBase64Encoded": "BASE64_ENCODED_PERSON_IMAGE"
        }
      },
      "productImages": [
        {
          "image": {
            "bytesBase64Encoded": "BASE64_ENCODED_PRODUCT_IMAGE"
          }
        }
      ]
    }
  ],
  "parameters": {
    "sampleCount": 2,
    "storageUri": "gs://my-bucket/output/"
  }
}

PowerShell Example

$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://us-central1-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/us-central1/publishers/google/models/virtual-try-on-001:predict" | `
  Select-Object -Expand Content

Pricing

Virtual Try-On uses the Imagen pricing model for Vertex AI.

Reference: See Imagen Pricing in the Vertex AI pricing documentation for current rates.

Pricing is typically based on:

• Number of images generated
• Image resolution/size
• Region of deployment

Use Cases

E-Commerce & Retail

1. Apparel Try-On: Generate images of customers wearing clothing items
2. Footwear Visualization: Show shoes/sneakers on various feet types
3. Accessory Preview: Display hats, sunglasses, jewelry on modeled faces
4. Size/Style Variants: Create multiple looks for different product options

Fashion & Beauty

1. Virtual Styling: Show outfits on diverse body types
2. Color Variations: Display same product in different colors on a person
3. Makeup Try-On: Visualize cosmetics on various skin tones
4. Personalized Recommendations: Generate tailored product visuals

Customer Experience

1. Reduced Return Rates: Accurate product visualization before purchase
2. Mobile Shopping: Lightweight try-on without augmented reality complexity
3. Accessibility: Enable virtual try-on for diverse customer bases
4. Conversion Optimization: Increase confidence in purchase decisions

Limitations & Considerations

Technical Limitations

• Input Images: Must be PNG or JPEG, max 10 MB each
• Output Limit: Maximum 4 images per request
• Aspect Ratio: Output maintains input image aspect ratio
• Rate Limiting: 50 requests per minute per region
• Prompt Language: English only

Content Considerations

• Supports both real and synthetic person images
• Includes person generation capability for creative scenarios
• User-configurable safety filters available
• Digital watermarking available for verification

Not Supported

• Generic image generation (use Imagen model instead)
• Real-time streaming
• Custom model fine-tuning
• Interactive/iterative editing (single request per try-on)

Related Resources

• Colab Notebook: Virtual Try-On Getting Started
• Supported Regions: Generative AI on Vertex AI Locations
• Image Generation Guide: Generate Images with Imagen
• Safety & Responsible AI: Imagen Responsible AI Guide
• Content Credentials: C2PA Support
• Python SDK: SDK Reference Documentation
• IAM Roles: Granting Roles & Permissions
• Authentication: Set Up Application Default Credentials

Summary

The Google Cloud Virtual Try-On API (virtual-try-on-001) is a production-ready service for generating realistic try-on images. It integrates seamlessly with Vertex AI, supports multiple programming languages, and includes enterprise features like watermarking and safety controls. With a quota of 50 requests per minute and support for up to 4 outputs per request, it’s suitable for e-commerce platforms seeking to improve customer experience through personalized product visualization.

Documentation Last Updated: 2026-03-05 API Model: virtual-try-on-001 Service: Google Cloud Vertex AI - Generative AI