Inspiration
Each of us has felt the grind of online shopping: "Oh that looks great on that fasion model, but will it look good on me?" With the latest breakthroughs in vision models, we imagined a better path. We sought out to creat an app that brings the certainty of an in-store fitting room to the web.
What it does
Our app turns every shopper into a fashion model. We built a marketplace where each listing is styled by you, that way sellers can showcase real fits on real bodies. Users can also upload photos, try on multiple pieces from their own closet, then mix in new items they do not own yet to create complete looks.
How we built it
We built a FastAPI backend that pairs with Google’s NanoBanana model to create a photorealistic image of the user wearing any piece of clothing while preserving the original photo’s lighting, background, and pose. On the front end, a React app delivers a clean, fast experience with intuitive uploads, instant previews, and simple sharing.
Challenges we ran into
- Inconsistent outputs: We often found that NanoBanana would fail to generate the same person, and would often generate entirely new people, or would forget to apply certain articles of clothing. We found that the optimal amount of clothes to apply at a time is 2, and that applying outfit collages works even better than separately uploaded images. Furthermore, we ensured that the temperature, top_p, top_k, and other parameters were set to ensure the model was as deterministic as possible.
Accomplishments that we're proud of
We were able to get the outputs to be extremely consistent, and it works on both guys and girls! The frontend is stunning, and the backend/AI outputs are reliable. And most importantly, while building this, we've realized how important a tool like this really is: ever since building it, we've been playing around with tons of custom outfits online to see if we'd look good in them!
What we learned
We learned techniques for optimizing the reliability of models like nano banana. Learned lots about weird frontend gimmicks with different sized images. Solidified our understanding about threading and concurrency to improve performance of processing images.
What's next for Tryon-ai
As we built this, we've realized how useful this sort of tech could be. As mentioned above, we've been playing around with this all day, testing different outfits on ourselves and figuring out what to buy! We're going to turn this from a prototype to a real concept (real deployed website and app) and let people actually use it! We believe that we are going to transform the nature of online fashion shopping!
Log in or sign up for Devpost to join the conversation.