11 Nov 2025

Pivot to Text-to-3D Guessing

< ⌂ >

This is our next post in our dev blog. We hope you like it and stay tuned.

Description

New technology: Text-to-3D guessing loop
New game loop: prompt objects so teammates guess the secret concept

TL;DR

Tech changed from 2D sketch recognition to text-to-3D generation (server with OpenAI Shap-E)
Core loop: get concept -> prompt -> spawn object -> teammates guess
Pressure mechanic: each generation degrades quality and forces better planning

Supervisor Feedback and Response

Feedback was to make technology central and fun in the core loop, add pressure/cost, and provide a roadmap. Our response was a faster social loop around text-to-3D prompting instead of ambiguous time-pressured sketches.

Core Loop v2

The game assigns a concept (for example: cow).
The builder must not prompt the target word and uses related prompts (for example: milk, barn, steak).
The game generates 3D objects and Unity spawns them in scene.
Teammates guess the concept; correct guesses end the round and award points.

We can generate 3D models by text prompts using OpenAI Shap-E, and builders can prompt directly inside the game UI.

The generation server runs as a local service and returns .glb assets for Unity. This lets us tune generation settings while keeping integration fast for the game team.

Text-to-3D server and generation pipeline screenshot. — Object generation server and model pipeline.

Generation workflow from prompt to spawned object.

Weekly demo of the full text-to-3D guessing loop.

To keep rounds tense, each generation lowers model quality over time, directly supporting the tamed-growth idea.

< ⌂ >