img2threejs — AI Image-to-3D Model Converter for Three.js

Introduction

img2threejs is an open-source tool that converts reference images into procedural Three.js 3D models using AI coding agents. Rather than generating mesh data directly, it produces clean TypeScript/JavaScript code that recreates the object procedurally in Three.js. The output is animation-ready, quality-gated, and optimized for token-efficient generation, making it practical for both interactive web experiences and agent-driven 3D content creation.

What img2threejs Does

• Converts reference images into procedural Three.js scene code
• Generates clean, readable TypeScript that builds 3D objects from primitives
• Applies quality gates to validate geometry, lighting, and proportions
• Produces animation-ready models with configurable motion parameters
• Optimizes prompts for minimal token usage during generation

Architecture Overview

img2threejs works as a Python CLI that orchestrates an AI coding agent to interpret a reference image and produce Three.js code. The pipeline has three stages: image analysis extracts shape, color, and spatial information; code generation produces procedural Three.js geometry using primitives and materials; quality validation renders the scene headlessly and compares it against the reference for fidelity. Failed quality checks trigger automatic refinement loops.

Self-Hosting & Configuration

• Install via pip with Python 3.10+
• Configure your LLM provider API key for the code generation backend
• Run the generate command with an input image path
• Customize quality thresholds in the configuration file
• Export scenes as standalone HTML files or importable Three.js modules

Key Features

• Procedural code output instead of opaque mesh files enables easy customization
• Quality-gated pipeline validates output before finalizing
• Animation-ready models with configurable rotation, bounce, and morph targets
• Token-efficient prompting minimizes API costs per generation
• Standalone HTML export for instant browser preview

Comparison with Similar Tools

• Three.js Editor — manual 3D scene building; img2threejs automates from images
• Blender — full 3D suite requiring modeling skills; img2threejs uses AI for generation
• TripoSR — generates 3D meshes; img2threejs outputs editable procedural code
• Shap-E — OpenAI 3D generation; img2threejs produces Three.js-native output
• Meshy — SaaS 3D generation; img2threejs is open source and self-hosted

FAQ

Q: What LLM providers does it support? A: Any OpenAI-compatible API, including Anthropic via proxy, DeepSeek, and local models.

Q: Can I edit the generated Three.js code? A: Yes, the output is clean procedural code designed to be human-readable and modifiable.

Q: How accurate are the 3D models? A: Accuracy depends on image complexity. Simple objects achieve good fidelity; complex scenes may need manual refinement.

Q: Does it require a GPU? A: No GPU is needed for generation. The headless quality-check renderer uses software rendering.

Sources

• https://github.com/hoainho/img2threejs
• https://github.com/hoainho/img2threejs/blob/main/README.md