Fashion with Antigravity & Gemini

Written by Margaret Maynard-Reid & Jason Davernport

This blog post showcases how we can quickly prototype an AI fashion app using Antigravity, and highlight new features in Antigravity 2.0 and Gemini. Whether you are a builder or experienced developer, you will find this blog post useful.

SOTA on Google Cloud

Here is a high level architecture of the app using the SOTA models and tools on Google Cloud:

Antigravity 2.0

Google Antigravity was released in late 2025 and upgraded it to 2.0 earlier this year.

It is an autonomous AI developer platform designed for the "agent-first" era. Here are some highlights of its features:

• An agent first approach to the development lifecycle
• Accesses the latest models from Google and other companies
• Executes complex, spec driven workflows with the user’s level of comfort in task delegation and privilege
• Learns from user behavior and lets users use tools and skills to refine agent behavior
• Supports MCP connections to many products including Google Cloud services.
• Meets you on the surface of your choice in CLI, IDE, or in Agent Manager
• Creates an agent team for you to get your development done
• Is powered by the latest Gemini models

And you can get started with Antigravity today by downloading it, or using this Codelab as a great walkthrough.

Gemini Core & Flash Models

Gemini 3.1 Pro is great for multi-step tasks and deep thinking while Gemini 3.x Flash and Flash Lite models are great for speed and coding tasks. These models are used for coding in Antigravity 2.0.

Nano Banana Image Models

Nano Banana Pro, 2 and 2 Lite are state of the art image generation and editing models. Our app showcases some of the new features from the state-of-the-art AI models and deploy to Google Cloud.

Nano Banana Pro has features such as:

• Search grounding and thinking
• Enhanced realism great for rendering intricate fabric textures in fashion images.
• Character and style consistency
• Studio-quality camera and lighting controls and editing Image resolutions of 1K, 2K, 4K with a wide range of aspect ratios

A2UI Spec

A2UI is a specification that allows agents to define the interfaces that should be rendered to a user dynamically. In this project, we used A2UI to dynamically render chat information the user wanted to display to the agent. You can read more about the spec here and see some additional examples using A2UI with Flutter for rendering here.

Building the App

How the app is built

Many ‘getting started’ app experiences focus primarily on the application’s look and feel, commonly associated with the frontend of an application. While this is great for a quick idea or a ‘vibe app’, it often is not sufficient as soon as you add additional data or access requirements.

With the Antigravity agent and Gemini 3, we coded an ‘almost three-tier’ app. Three tier applications are common in the cloud web application space. They consist of:

• A frontend service: this is typically the interface the user sees and is publicly exposed.
• A backend service: this contains the connections to other applications, services, and gateways. A basic principle is to use this and limit the ability to use the services to only the frontend.
• A database: this contains a layer to store user state, reference data the application needs, and other structural elements that may be required including configurations.

We used the Antigravity agent to code the frontend and backend, and used a simpler JSON file for our ‘database’. You can use Antigravity to help provision a local container to run a database such as PostgreSQL or provision a cloud database like CloudSQL. These databases give you more flexibility for the types of information you would store and the ability for easier reads and writes.

For an app framework, we used the Flutter framework. Flutter is built on top of the Dart programming language, and gives us the ability to deploy a common code base as a web app, a compiled binary for a target operating system, or even a mobile application.

And the app uses Gemini 3 Pro Image to help with modifying the mood of a specific option in the storefront.

Deploy the app

You can deploy the app to a service like Google Cloud Run, or run it locally.

Prerequisites:

• A Gemini API key with access to the Gemini models
• A Google Cloud project if you wish to deploy the app

Steps:

1. Clone the application repo: https://github.com/davenportjw/agentic-fashion
2. Read the readme to install Flutter and other required dependencies
3. Build the app
4. Then there are 2 options to deploy: to Google Cloud Run or run locally

Option 1: Deploy to Google Cloud Run

To deploy the full application to Google Cloud, ensure you are authenticated and have your project configured:

gcloud auth login
gcloud config set project <your_project_id>
./deploy.sh

Once deployment finishes, access the app using the Frontend Cloud Run URL printed in your terminal.

Option 2: Run Locally

If you prefer to run everything locally without deploying, clone the repository and run run_local.sh (this spins up the FastAPI backend, the ADK AI Stylist agent, and the Flutter web frontend).

Using a Gemini API Key:

./run_local.sh --GEMINI_API_KEY <your_api_key>

Or using your Google Cloud Project:

export GOOGLE_CLOUD_PROJECT=<your_project_id>
./run_local.sh

The script will automatically launch Chrome to display the frontend.

App Demo: UI Walkthrough

Once you follow the steps above to clone the code from our GitHub repo and have the app up and running, let’s take a look at what the app does.

At a high level, here are the features:

1. A store front with a list of fashion images
2. Select an image to enter the customization process.
3. The user can change an outfit’s style and color.
4. Save the outfit by downloading it.
5. Page through the edit history
6. Add to chopping cart
7. An agent the user can chat with to find dresses and customize them in an agentic (chat) interface.

Step by step let’s walk through the app UI demo.

First you are greeted with the store front, click on “Enter the store”,

Select one fashion dress, and click on “Make the design mine”,

Now we can choose a style (default to Classic) and or color to re-design the dress.
Let’s choose Avant-Garde and a Red color and “Generate with Gemini”, here we are using Nano Banana Pro to make this change.

You can download the image by pressing the download button, or page through your edits.

If you choose to use the chat option, you can ask the agent questions about different styles to find the one that you want, and also customize. This is powered by A2UI, where the agent gets to pick how it displays interfaces to you based on designs you set, and rendered by the front end UI.

What’s next?

We’ve built a complex, three-tier application with powerful SOTA AI integration in a fraction of the time, all thanks to Antigravity and Gemini. Start using Antigravity for your development. The future of coding is agentic. In addition, Antigravity 2.0 is great for building multi-agents agentic apps as an example mentioned in our blog above. Check out the SOTA image model, Nano Banana Pro, and see how easy it is to integrate cutting-edge generative AI into your app.

Project inspiration

Margaret Maynard-Reid - I work at the intersection of AI, art, and fashion design, incorporating AI into creative workflows. As a Google Developer Expert (GDE) in Cloud AI, I create tech content and contribute to open-source projects. Jason and I co-presented a short talk "Fashion with Antigravity" at DevFest Seattle 2025, on the day when Antigravity was launched. So we’ve decided to collaborate to demonstrate how to quickly build a production-grade fashion app using Antigravity, extending the short talk to a project with sample code, detailed blog post and YouTube videos (upcoming).

Jason Davenport - Jason is an Area Technical Lead at Google Cloud. He is the overall Tech Lead for information experience, developer relations, and frameworks and languages. He focuses on applied generative AI in coding and software development, including AI in the SDLC, integration with AI in business transformation, and forward looking trends in AI. His background is in data engineering, software engineering, and devops. Prior to his current role at Google, Jason has held various roles in technical leadership and management, including as a practice director for a regional consultancy.