This project addresses the conversion of 2D CAD images into 3D CAD models (STEP files) using Vision-Language Models (VLMs). It targets engineers, designers, and researchers needing to digitize or reconstruct 3D designs from 2D inputs, offering an automated pathway to generate editable 3D CAD data. The primary benefit is streamlining the 3D modeling process from legacy or image-based 2D designs.

How It Works

The system employs a multi-agent architecture orchestrated by VLMs. An input 2D CAD image is processed by a CAD Code Generator AI Agent, which translates the visual information into CAD code. This code is then executed and debugged by a Python Execution and Debugging AI Agent, producing a STEP file. The generated STEP file is rendered into a 3D model image, which is then fed back to a CAD Code Refiner AI Agent for iterative improvement, ultimately feeding back into the debugging agent.

Quick Start & Requirements

• Installation: Clone the repository (git clone git@github.com:neka-nat/cad3dify.git) and install dependencies using Poetry (poetry install).
• Prerequisites: Requires an API key for the chosen LLM provider (e.g., OPENAI_API_KEY). Supports GPT-5, Claude 4.5 opus, Gemini 3 pro, or Llama 3.2 on Vertex AI.
• Running: Execute via CLI (cd scripts && python cli.py <2D CAD Image File>)