Skip to content
spatialhub
← All work

2026 · In development

Toykind

AI companions generated from a single photo

Overview

Toykind is a mixed-reality life sim. Snap a photo of anything and it becomes a fully AI-driven character: segmented, reconstructed in 3D, rigged, and textured automatically, then dropped into your scanned room.

Characters become companions with emotions, goals, and relationships with each other, and take part in a small in-world economy. A companion app for iOS and Android keeps the conversation going when you're not in the headset.

How it's built

  1. 01

    A photo runs through segmentation (SAM 3), 3D reconstruction (SAM 3D, Hunyuan3D, TRELLIS), auto-rigging (UniRig), and texturing, each as a containerized GPU job on Google Cloud with durable orchestration.

  2. 02

    An agent system gives every character memory, emotions, goals, and relationships that persist between sessions.

  3. 03

    The React Native companion app renders 3D models and Gaussian splats on device, with push notifications and TestFlight and Firebase distribution.

  4. 04

    A native visionOS client places characters in the user's scanned space.