Fit scanner

Watching a movie and wishing you dress just as good as the actors but don't know how? Scan their fit and get shopping recommendations in a few taps!

Comment

Inspiration

Some of us tend to admire how others dress—whether it's a stranger on the street or someone online with a great sense of style. But when it comes to dressing well ourselves, it can be overwhelming. With the countless fashion styles, trends, and stores out there, knowing what to wear or how to recreate a look is tough. That sparked an idea: what if you could just point your phone at someone’s outfit—on the street or your screen—and instantly get information on how to recreate that look?

What it does

Fit Scanner is an app that helps users identify and replicate outfits they see in real life or online. You can upload a photo or point your camera at someone (with consent), and the app will analyze the outfit, describe each clothing piece in detail, and recommend similar items available online. Additionally, the app shows a curated feed of the “Top Fits of the Day,” scraped from fashion blogs and community posts, to keep users inspired with trending looks.

How we built it

We built Fit Scanner using the following tech stack:

  • Frontend: Flutter for cross-platform mobile & web app development.
  • Backend: FastAPI to handle API endpoints.
  • Database: Dockerized PostgreSQL for persistent storage.

AI Integration:

  • We explored image-capable LLMs like Gemini to describe clothing from images.
  • Experimented with prompt engineering to get structured outputs from LLMs.
  • Attempted image-to-text analysis followed by text-to-image generation for visualization.
  • Web Scraping: Our backend crawls multiple fashion sites to gather daily outfit inspirations.

Challenges we ran into

  • Not all large language models (LLMs) support image inputs.
  • Most good generative models (especially for realistic image generation) are expensive or hard to run locally.
  • Prompt engineering for accurate multi-item outfit breakdowns was much harder than expected.
  • We had difficulty generating AI images that faithfully reflected described outfits.

Accomplishments that we're proud of

  • Built a working prototype that lets users upload images and view detailed clothing descriptions.
  • Successfully created a backend system to scrape and serve trending outfits.
  • Integrated Flutter with FastAPI and Docker, enabling a smooth flow from user input to AI-powered results.
  • Gained hands-on experience with AI tools and prompt tuning.

What we learned

  • Image-processing LLMs are powerful but difficult to use without proper infrastructure or access.
  • Prompt engineering is both a science and an art—results vary a lot based on how you phrase things.
  • Realistic text-to-image generation is still a challenging area, especially when it comes to styling multiple items.

What’s next for Fit Scanner

  • Add a social layer, allowing users to share and vote on community-submitted fits.
  • Seek funding/support to run or access powerful GenAI models needed for production use.

Built With

Share this project:

Updates