# SuperStyle

- **Event:** [Nano Banana Hackathon](https://cerebralvalley.ai/e/nano-banana)
- **When:** Sat, Sep 6 at 9:00 AM – 9:00 PM (PDT)
- **Where:** 365 Toni Stone Crossing, San Francisco, CA
- **Team:** [Darius Emrani](https://cerebralvalley.ai/u/dariusemrani), [David Zhang](https://cerebralvalley.ai/u/dzhng), [Lily Zhang](https://cerebralvalley.ai/u/lilyzhng)
- **GitHub:** https://github.com/dzhng/nanobanana-hackathon
- **Demo video:** https://www.loom.com/share/f31e2d7b2ed1491a9b153c4a064b0c97?sid=f14170b1-51e5-48ea-b295-64e484fcf388
- **Gallery:** https://cerebralvalley.ai/e/nano-banana/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/nano-banana/hackathon/gallery/31

Our project is about changing your hairstyle with the nano banana model. We've all had the pain point where we want to try a new hairstyle, but don't want to actually take the risk to get it cut. Now AI can just help you find the perfect hairstyle, without the risk.

The entire image processing pipeline is actually quite in-depth. It takes about 60 seconds to generate a hairstyle because we're doing multiple pre-processing steps. First, it's relighting the reference image, and it's also pre-processing your own image by deleting your hair first. And then it's going to run a bunch of nano banana calls in parallel. And we actually have a VLM as a judge to find the best picture out of multiple parallel calls. 

We also built a data engine to curate the dataset for hairstyles by pulling images from Pinterest, hairstyles.com, where we pulled over 5,000 images and run VLM to do the data filtering and judge to make sure only high-quality images with good aesthetics are kept for the users.

---

Markdown version of https://cerebralvalley.ai/e/nano-banana/hackathon/gallery/31. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt