FitMi

Inspiration

Travel often starts with a simple but stressful question: "What should I wear?" Before every trip, we find ourselves scrolling through social media for outfit inspiration. While many creators share beautiful travel looks, it is hard to know whether those outfits actually work for us -- our body shape, comfort needs, destination, weather, or the specific situations we will encounter. This gap between aspirational inspiration and practical, personal styling inspired us to build this project. We wanted to create an experience that helps real people dress confidently for real travel scenarios, rather than copying someone else's look.

What it does

We built an AI-powered travel styling app that generates personalized, scenario-based outfit recommendations. Using Stitch for UI prototyping and Google AI Studio (Gemini) for generation, the app takes into account: Individual attributes (body shape, height, style preferences) Travel context (destination, weather, daily activities) Social and situational norms (e.g. wedding guest etiquette, walking-heavy days) Instead of producing generic fashion suggestions, we designed the system to output wearable, realistic outfits that users can actually imagine themselves wearing. A key design principle was anti-modelization: the app avoids runway aesthetics or aspirational imagery and focuses on self-projection and everyday usability.

How we built it

The system combines structured rule-based logic with generative AI: We defined a set of hard fashion constraints, such as weather rules, body-shape considerations, and situational etiquette (for example, avoiding white outfits for wedding guests, or recommending waterproof shoes for rainy travel days). These rules are injected directly into the AI system prompt to constrain generation and reduce hallucination. We used A/B visual style quizzes to quickly learn user preferences, improving personalization without requiring long onboarding flows. To improve performance and user experience, we optimized generation with partial loading, caching, and progressive rendering. This hybrid approach allowed us to balance creativity with reliability.

Challenges we ran into

The biggest challenge was formalizing fashion knowledge into usable rules. Fashion decisions are rarely based on a single factor. A single outfit choice may need to consider: Weather, temperature, and walking distance Social norms and cultural expectations Body proportions and comfort Practical constraints like luggage size or repeat wear For example, rules like “wedding guests should not wear white” are simple on the surface, but quickly expand into edge cases depending on region, dress code, season, and formality level.

Accomplishments that we're proud of

We are especially proud that this project was built by two creators without a computer science background, using Gemini 3 through Google AI Studio and its underlying APIs to bring a complex, real-world idea to life. Rather than starting from a purely technical perspective, we approached the problem from lived experience and product intuition—then learned how to translate those ideas into a working AI system. Through this process, we proved that powerful generative models can enable non-engineers to build meaningful, functional applications when paired with the right tools. We are also proud that we successfully transformed subjective fashion knowledge into structured, enforceable styling rules. By encoding constraints around weather, body shape, comfort, and social etiquette, we guided the model to reason about context instead of producing generic or aspirational outputs. Finally, we intentionally designed the system to prioritize realism and self-projection. Instead of model-like imagery or idealized fashion, the app focuses on wearable, everyday outfits that users can realistically imagine themselves wearing. This balance between generative creativity and practical constraints is something we consider a core achievement of the project.

What we learned

Purely generative systems are not enough for decision-heavy domains like fashion. Clear constraints dramatically improve output quality and trust. Designing for realistic use matters more than producing visually impressive results. This project taught us how to translate subjective human knowledge into structured, AI-compatible logic—while still leaving room for personal expression.

What's next for FitMi

There are several directions we are excited to explore next: Expanding scenarios: supporting more use cases such as business travel, formal events, outdoor activities, and region-specific cultural norms. Cultural and media-inspired styling: offering optional outfit recommendations inspired by films, TV shows, and books connected to specific destinations—for example, suggesting looks inspired by iconic characters or scenes filmed at a location, allowing users to “style their trip” through cultural references. Shopping integration: providing optional purchase links for recommended items, allowing users to move seamlessly from inspiration to action. Virtual try-on: enabling users to preview outfits on their own avatars for better confidence and fit understanding. Closet-based styling: generating outfits based on items the user already owns, encouraging sustainability and reducing unnecessary purchases. Smarter personalization over time: learning continuously from user feedback and choices to refine recommendations. These extensions would move the app closer to becoming a long-term, personalized styling companion rather than a one-time planning tool.

Built With

• ai-studio
• gemini-3
• javascript
• react
• stitch
• typescript
• web