Mark LVIII

Video Demo

About this project

## Inspiration In Spider-Man: Far From Home, Peter Parker designs his new suit by seamlessly interacting with holographic displays and an AI assistant. This sparked a question for us: in today's era of rapid AI advancement, can we build a tool that empowers students and designers to quickly prototype, modify, and simulate objects using that same kind of AI-driven power? That vision led to the creation of Mark LVIII. ## What it does Mark LVIII bridges the Gemini CLI with Blender via MCP to create a rapid, hands-free 3D prototyping workshop. Its core features include: * **Voice-to-CAD:** Speak directly to the AI to conceptualize and generate complex 3D objects on the fly. * **Gesture Control:** Replace the traditional mouse and keyboard with real-time hand-tracking to grab, manipulate, and assemble parts naturally. * **Procedural Generation:** Say "Build a rocket," and Mark LVIII breaks it down into a modular Build Spec, executing the code in Blender while keeping every individual part fully editable. * **Live Modifications:** Instantly spawn primitives ("Add a red cube") through live Python execution inside the Blender environment. * **Smart Inspection:** Query the system for structural data—like transforms or bounding boxes—or command the camera to "frame this" specific part for a closer look. * **Reference-Driven Design:** Instruct the agent to scrape the web for reference images and automatically translate those visual concepts into a 3D workspace. Under the hood it's three processes: a **cloud-style agent server** (Gemini Live + Google ADK), a **local macOS client** (mic, webcam, cursor overlay), and **Blender** running a custom addon. They talk over two clean seams: a WebSocket between server and client, and a JSON-over-TCP socket between the client and Blender. ## How I built it Under the hood, the architecture consists of three core processes: a cloud-based agent server (Gemini Live + Google ADK), a local macOS client (handling the mi

Gallery