Bubbles | Devpost

Inspiration

Bubbles started from the desire to make 3D world‑building as easy as talking to a friend. Instead of complex tools or game engines, we wanted anyone with Snapchat Spectacles to create playful, personal 3D spaces using just their voice.

What it does

Bubbles turns Snapchat Spectacles into a voice-controlled 3D builder: you speak a prompt, an AI model generates a 3D asset, and you place it inside your own tiny world, your “Bubble”. Each Bubble is stored in Snapcloud (Supabase) and can be explored both in AR on Spectacles and on a companion website that shows the same worlds in 3D. Website: https://bubbles-tan.vercel.app

How we built it

We created a Spectacles lens that uses voice prompts and the AI Playground sample project to generate or fetch 3D models, which users can then place inside their own Bubble. Each Bubble, its models, and their positions are stored in Supabase, which also feeds a Next.js + Three.js website that displays an Apple Watch–style grid of Bubbles and a dedicated 3D viewer for each one.

Challenges we ran into

Working end-to-end in only 24 hours meant we had to scope very aggressively: connecting voice input, AI model generation, AR placement, and a shared backend is a lot of moving parts. Getting the data model for Bubbles and 3D assets stable enough so both Spectacles and the website could read it consistently was tricky, and integrating 3D rendering on the web while debugging on-device AR took careful time management.

Accomplishments that we're proud of

We’re proud that in just one day we connected Spectacles, AI-generated 3D content, and a live backend into a single experience that feels cohesive. Seeing a Bubble created in AR appear moments later on the website, complete with its 3D models and creator name, was a big milestone and showed that the “speak it, see it, share it” vision of Bubbles actually works.

What we learned

We learned how important a clean, simple data model is when you’re syncing AR experiences and web experiences on a tight deadline. We also deepened our understanding of building for Spectacles, handling voice input in a fast feedback loop, and balancing visual polish with reliability when you only have 24 hours.

What's next for Bubbles

Next, we want to improve the quality and variety of AI-generated 3D models, add more social features (following creators, liking or remixing Bubbles), and make exploration feel more like walking through a universe of tiny worlds. We also plan to harden the Supabase backend, refine the website’s 3D viewer, and turn this hackathon prototype into a platform where anyone can build and share living 3D spaces with just their voice.