This is a submission for the Google AI Studio Multimodal Challenge
What I Built
AI Fashion Playground is a web platform that reimagines the personal wardrobe experience. It allows users to mix and match outfits from their own closet or e-commerce sites on a realistic, AI-generated avatar of themselves.
The applet solves several key problems for fashion consumers:
- The "Try Before You Buy" Gap in E-commerce: It provides a hyper-realistic virtual try-on, allowing users to see how clothes actually look on them before buying, which can drastically reduce costly and wasteful returns.
- The "Full Closet, Nothing to Wear" Paradox: An AI stylist feature rediscovers and curates new outfits from the user's existing wardrobe, promoting sustainable fashion choices and maximizing the value of clothes they already own.
- The Inaccessibility of Personal Styling: It acts as a free, on-demand personal stylist that understands a user's unique wardrobe, style preferences, and daily needs to provide expert guidance for any occasion.
Demo
- Live Application: https://aistudio.google.com/apps/drive/1Mj9xQQgZyyGPMzJiVtlOpKAKUyQw5MRV
- Google Cloud Run: https://ai-fashion-playground-569679889124.us-west1.run.app
- Video Demo: https://youtu.be/eiwI9AjeQks
Screenshots
Landing Page
Freestyle AI Playground
Get Ready With Me using AI
How I Used Google AI Studio
Google AI Studio was not just a tool for testing; it was the central workbench where the AI-driven soul of this application was forged from scratch. Every core feature was born from an iterative cycle of prototyping, prompt engineering, and code generation within the AI Studio environment.
-
Engineering the Virtual Try-On Engine: The most complex feature is generating a realistic image of the user wearing different clothes. I used AI Studio to master the
gemini-2.5-flash-image-preview
model.- Prototyping: I started by uploading a sample avatar and a few clothing images directly into the Studio's multimodal prompt interface.
- Prompt Engineering: My critical challenge was ensuring the model preserved the user's identity. Through dozens of iterations in AI Studio, I engineered the core instruction found in
services/geminiService.ts
:**Crucially, do not alter the person's face, hair, or body shape. They must remain identical to the first image.**
This precise instruction was the key to achieving realistic, true-to-life results.
-
Building the AI Stylist's Brain with Structured Output: The "Get Ready With Me" feature required the AI to think like a stylist and provide structured, usable data.
- Schema Definition: I used AI Studio's structured output feature to design a precise JSON schema for the
gemini-2.5-flash
model. This is visible in theselectOutfitFromWardrobe
function, where the AI is required to return a{ selection: string[], reasoning: string, affirmation: string }
object. Prototyping this in the Studio meant I never had to write fragile string-parsing logic in my app. - Persona Crafting: I fine-tuned the stylist's personality directly in the AI Studio prompt, crafting the persona of a "fun, encouraging, and stylish best friend" and testing its conversational output until it felt just right.
- Schema Definition: I used AI Studio's structured output feature to design a precise JSON schema for the
-
Automating the Digital Wardrobe (Multimodal Vision): To make the wardrobe "smart," I needed to analyze and tag every clothing item.
- Image-to-JSON: I leveraged AI Studio to build a prompt for
gemini-2.5-flash
that takes a clothing image as input and returns a structured JSON object with itscategory
,color
,season
, andstyle
. - Seamless Workflow: This entire workflow—from experimenting with the image-and-text prompt in AI Studio to using "Get Code" to implement the
analyzeClothingItem
function—allowed me to build a powerful, automated feature in a fraction of the time.
- Image-to-JSON: I leveraged AI Studio to build a prompt for
In essence, AI Studio provided a seamless bridge from creative ideation to production-ready code. It allowed me to visually build and validate complex multimodal prompts and then translate them directly into the application's service layer, making it the indispensable core of my development process.
Multimodal Features
AI Fashion Playground is built on a foundation of several interconnected multimodal features that create a seamless and intelligent user experience.
Multimodal Virtual Try-On (Image + Text → Image)
This is the app's centerpiece. It takes multiple images as input (a photo of the user, photos of clothing items) and a text prompt (describing the mood, scene, or style) to generate a single, new, coherent image. This powerfulimage-and-text-to-image
capability is what allows users to visualize themselves in new outfits.Automated Wardrobe Intelligence (Image → Structured Text)
When a user uploads a piece of clothing, the app usesgemini-2.5-flash
to analyze the image and output a structured JSON object containing its category, color, style, and ideal season. Thisimage-to-text
functionality transforms a simple image library into a smart, filterable, and machine-readable wardrobe, which is the knowledge base for the AI stylist.-
Conversational Styling Pipeline (Text → JSON → Image)
The "Get Ready With Me" feature is a multi-step multimodal workflow:- Step 1 (Text → JSON): The user provides a natural language text prompt describing an event (e.g., "a first date at a cozy cafe"). The AI processes this text, cross-references it with the user's wardrobe data, and outputs a JSON object containing the names of the selected clothing items and a text-based rationale for its choice.
- Step 2 (Images + Text → Image): The application then programmatically gathers the images for the selected items, the user's avatar image, and the AI's styling advice as a text prompt. It feeds all of this into the image generation model to create the final styled look. This pipeline creates a magical, conversational experience that feels like collaborating with a real fashion expert.
Iterative Image Refinement (Image + Text → Image)
After an image has been generated, users aren't locked in. They can provide additional text prompts (e.g., "change the background to a park at sunset") to modify the existing image. This gives users fine-grained creative control and makes the process interactive and fun.
By integrating these multimodal features, AI Fashion Playground moves beyond a simple utility and becomes a creative partner, helping users unlock the full potential of their wardrobe and express their unique style with confidence.
💖 Built for Everyone
AI Fashion Playground believes fashion should be fun, inclusive, and accessible to everyone. Whether you're exploring your personal style, planning outfits for special occasions, or just having creative fun – this platform is designed to celebrate your unique identity and boost your confidence.
Made with ❤️ for the fashion-forward and the style-curious alike.
Top comments (0)