Inspiration
Shopping for clothes online still forces customers to make decisions from flat product photos. Many shoppers hesitate because they cannot picture a garment on their own body, compare several angles, or understand how a selected size may differ from their usual size. SMART MIRROR was created as a mobile-first way to reduce that uncertainty while giving stores a future integration path.
What it does
SMART MIRROR is a live pre-commercial MVP. A user can create or select a persistent multi-view avatar, import a garment, generate a visual try-on, rotate through available views, save the result in a Studio, and optionally generate a short presentation video. The responsive web application is also installable as a PWA.
The AI Stylist answers questions about occasions, wardrobe combinations, possible size choices, and accessories. Its size guidance includes a confidence level and an explanation. Every result is presented as a visualisation, not a guarantee of exact fit, colour, size, fabric behaviour, or final appearance.
For stores, the product includes an early Business area explaining the proposed virtual try-on API, a working “Try it on me” demonstration interaction, a real partner-enquiry form, a waitlist, and Firebase-backed product metrics. Missing history is displayed as “No data yet” rather than replaced with invented traction.
How we built it
Gemini is used directly in the deployed product. Gemini image generation powers the avatar and virtual try-on workflow. Gemini text generation powers AI Stylist and the protected AI Operations workflow. AI Operations can classify requests, summarise feedback, and draft next actions, but a human must approve or reject every recommendation. It never sends external messages automatically.
Google Cloud provides the deployed foundation: Firebase Hosting serves the PWA, Cloud Functions protect server workflows and API credentials, Firebase Authentication identifies users, Cloud Firestore stores genuine business records and metrics, and Firebase Cloud Messaging supports notifications for long-running jobs. Gemini API keys remain server-side. Generated media is stored in Backblaze B2. Optional YouCam and Kling integrations support appearance analysis and short videos when their credentials and credits are available.
Challenges we ran into
The hardest challenges were preserving identity across multiple views, keeping avatars visually aligned, handling long-running AI tasks on mobile, avoiding stale PWA data, and separating a convincing demonstration from claims that the prototype cannot yet prove. We added persistent account data, media caching, background task states, notifications, cautious sizing language, and explicit Demo mode states for features that are not configured.
Accomplishments that we're proud of
We built and deployed an end-to-end mobile-first workflow rather than a static concept: authenticated avatar creation, visual try-on, a Studio history, optional presentation videos, an AI Stylist, a partner enquiry flow, real Firebase-backed metrics, and human-reviewed AI Operations. The same product can be tested on the web or installed as a PWA.
What we learned
A useful virtual try-on product needs more than image generation. It needs identity persistence, predictable recovery after refresh, honest uncertainty, privacy boundaries, merchant intent data, and human control over AI-assisted business actions. We also learned that “No data yet” is better product design than synthetic metrics when commercial validation has not happened.
What's next for SMART MIRROR
SMART MIRROR is an MVP in commercial launch preparation, not an established revenue-generating company. The deployed product, Gemini features, Google Cloud architecture, partner form, waitlist, event tracking, and human-reviewed operations flow are real. Production merchant integrations, subscription billing, verified revenue, paying-customer evidence, proven conversion improvement, and guaranteed sizing are not yet available.
At submission time, no verified revenue, paying-customer evidence, or sufficiently complete expense ledger is available. Missing financial records are not replaced with estimates. The next step is to validate the experience with shoppers and a small number of stores, document the first pilots, and measure which products users genuinely try most often.
Log in or sign up for Devpost to join the conversation.