Guide · Agent skill

How to use the Virtual Try-On Agent skill (retail-virtual-tryon)

Updated 2026-10-29 · written with AI from the project's own documentation and checked against our catalogue data

The Virtual Try-On Agent skill (retail-virtual-tryon) teaches an AI agent to set up and build a virtual try-on application, where a shopper uploads a photo and sees clothing rendered on them as a still image or a short catwalk-style video. This guide is for developers who may be new to agent skills and want to install the skill, trigger it, and run a local sandbox or export a web app. By the end you will know what to install, what Google Cloud resources it needs, and which steps deserve a review before anything goes public.

Virtual Try-On Agentgoogle/adk-recipes · Apache-2.0 Download the skill

Quick start

1Download and unzip the skill. Download the ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder, keeping the folder name retail-virtual-tryon. Alternatively, upload the ZIP under Settings → Capabilities → Skills in the Claude apps.
2Meet the Google Cloud requirements. Confirm you have a Google Cloud project with billing, access to Gemini image models and Veo, Python 3.11+, and the gcloud CLI installed.
3Ask for a try-on agent. Make a natural request such as 'build a virtual try-on demo on Google Cloud' and wait for the setup menu.
4Answer the setup questions. Choose Quick start or export mode, then answer one question at a time. Press Enter to accept each default.
5Run bootstrap and setup. Run the bootstrap script from the shell, then run setup.py against design-spec.md using .venv/bin/python.
6Test in the sandbox. Open http://localhost:8080 in your browser after the sandbox starts and try a fitting.

What the skill makes the agent do

An agent skill is a folder of instructions and scripts that an AI agent loads when your request matches the skill's description. This one turns the agent into a builder for virtual try-on (VTO) agents on Google Cloud. It uses Gemini image models for still try-on images and Veo for catwalk-style video animations.

The skill walks you through a setup interview, prepares Google Cloud resources such as storage buckets and enabled APIs, and then either runs a local sandbox or exports a containerized web app for Cloud Run deployment.

Image try-on: a composite image of the user photo and a product.
Video try-on: catwalk-style mp4 animations built from the composite image.
Local sandbox: a test page you open at http://localhost:8080.
Export: a standalone containerized codebase with a GCS-hosted product catalog and Cloud Run configuration.

Install the retail-virtual-tryon skill

Download the ZIP for this skill from this page and unzip it into one of two locations. For all your projects, use ~/.claude/skills/. For a single project, put it in that project's .claude/skills/ folder. The folder name must be retail-virtual-tryon so the agent can find SKILL.md.

In the Claude apps, you can instead upload the ZIP under Settings → Capabilities → Skills.

Personal install: ~/.claude/skills/retail-virtual-tryon/
Project install: .claude/skills/retail-virtual-tryon/ inside your project.

Check the requirements before you start

You need a Google Cloud project with billing turned on and access to Gemini image models and Veo. Image and video generation will not run without this. You also need Python 3.11 or newer and the gcloud CLI, which the skill uses for project configuration and Cloud Run deployment.

The skill installs its pip packages through its own pyproject.toml during bootstrap. The packages it names are google-genai, google-cloud-storage, pillow, pyyaml and google-adk.

Python 3.11+
Google Cloud project with billing and access to Gemini image models and Veo
gcloud CLI
pip packages: google-genai, google-cloud-storage, pillow, pyyaml, google-adk

Trigger the skill with a natural request

Ask the agent in plain language, for example: 'I want to build a virtual try-on demo on Google Cloud.' The skill then shows a setup menu first. It does not run setup or deployment scripts on a general request. Your first reply should be a menu choice.

Choose 1 for Quick start, which sets up a local testing sandbox through a short interview. Choose 2 to export a web app and set up GCS catalog sync. Pressing Enter uses the default, which is Quick start.

Quick start (1): local sandbox, roughly 90 seconds according to the skill's menu.
Export Web App & GCS Catalog Sync (2): standalone containerized code, roughly 3 minutes according to the skill's menu.
The interview asks one question at a time and shows the default, for example: Q: Try-on mode? [default: 2].

Run the setup and open the sandbox

The skill sets up a workspace with a .venv, installs itself in editable mode, and writes a design-spec.md file that records your answers. You run the bootstrap script once from a shell. The skill's own commands then run against the workspace config. Use .venv/bin/python rather than plain python, as the skill instructs.

Quick start asks for your project ID, try-on mode, region, catalog path, and whether to upload a local catalog to GCS. The defaults are image_and_video, us-west1, and the bundled demo catalog. When setup succeeds, the sandbox starts and you open http://localhost:8080 in your browser.

Try-on mode 1 is image_only and 2 is image_and_video.
Catalog path accepts demo, a local folder, or a gs:// URI.
Export mode asks for a GCS catalog bucket (default {project_id}-tryon-catalog) and a target directory (default ./vto-retail-app).

SKILL_DIR=$(for d in ~/.claude/skills ~/.agents/skills ~/.gemini/skills ~/.cursor/skills; do
  [ -f "$d/retail-virtual-tryon/SKILL.md" ] && echo "$d/retail-virtual-tryon" && break
done)
bash "$SKILL_DIR/scripts/bootstrap.sh"
.venv/bin/python "$SKILL_DIR/scripts/setup.py" --config ./design-spec.md

Review the safety notes before deploying

This skill runs scripts and uses the network. It creates cloud buckets and enables APIs. Its exported Cloud Run deploy script makes the service publicly accessible through the --allow-unauthenticated flag. Treat the skill as a prototype tool, not a production deployment tool, unless you have reviewed it.

Before you run anything, read the scripts in the skill folder, check the project ID the agent writes into design-spec.md, and confirm that a public endpoint is what you want. Remove or change the public access setting if the demo should not be open to everyone.

Scripts run on your machine, including bootstrap.sh and setup.py.
Setup creates buckets and enables APIs in your Google Cloud project.
deploy_cloudrun.sh uses --allow-unauthenticated, so the service is public.
Generation calls use your billing account.

Tips for a smooth first run

Start with the demo catalog and image_only mode to confirm your setup works before you add video. Video generation is the more expensive path because it adds a Veo step on top of image generation. Keep your design-spec.md file, since it records the answers that the scripts read.

If you want a standalone deployable cloud agent with Agent Engine, Gemini Enterprise App, Cloud Run and Agent Garden, the README points to a separate contrib project in the same repository.

Use .venv/bin/python, not bare python.
Keep the bootstrap and setup steps in order and confirm each one succeeded.
Export the web app only after you have tested the sandbox.

What people use it for

A retailer's developer builds a clothing try-on demo to show stakeholders how shoppers could see outfits on themselves.
A team creates a starting codebase for a Cloud Run try-on web app with a product catalog hosted in a GCS bucket.
A developer tests try-on image generation locally with the bundled demo catalog before adding their own product photos.
A team exports a containerized web app and reviews its deployment script before deciding on access controls.

Frequently asked questions

Is the Virtual Try-On Agent skill free?

The skill is licensed under Apache-2.0. Running it needs a Google Cloud project with billing turned on, because image and video generation use Gemini image models and Veo. The source does not list prices, so check your Google Cloud billing for current costs.

What do I need to install before using this skill?

You need Python 3.11 or newer, a Google Cloud project with billing and model access, and the gcloud CLI. The skill installs its pip packages through its pyproject.toml, which covers google-genai, google-cloud-storage, pillow, pyyaml and google-adk.

Where do I put the skill folder for Claude?

Unzip the downloaded ZIP into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder for one project. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.

Why does the agent ask me questions instead of running setup right away?

The skill is built to show a setup menu first and wait for your choice before it runs any setup or deployment script. You pick Quick start or export mode, then answer one question at a time. This prevents it from creating cloud resources on a general request.

Is it safe to deploy the exported app to Cloud Run?

Review it first. The exported deploy script uses --allow-unauthenticated, which makes the Cloud Run service publicly accessible. The skill is not described as ready for production use without review, so check the access settings and scripts before you deploy.

Can I use this skill without a Google Cloud account?

Not for the try-on generation itself. The skill needs a Google Cloud project with billing and access to Gemini image models and Veo, and it sets up cloud resources. Without that, you can read the skill's files and code but cannot run image or video generation.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use this skill if you are a retail developer who wants a working virtual try-on prototype on Google Cloud and is comfortable with cloud billing and running setup scripts. Pick something else if you want a ready-made consumer shopping app with no cloud setup, or if you need a production deployment without reviewing the public Cloud Run configuration.

Download the skill