IMAGE-BLASTER

Built by Neilson Koerner-Safrata from the World Labs design team, IMAGE-BLASTER is an open-source 3D generation toolkit for Claude Code that transforms a single image into a fully explorable environment complete with meshes, physics, lighting, and audio.

The Workflow

The project explores how agentic AI workflows can compress traditionally complex 3D pipelines into a lightweight prompt-driven process.

• Generate explorable 3D environments from a single image using World Labs' Marble
• Extract dynamic objects into editable meshes
• Automatically generate ambient audio and physics SFX

The system combines Marble + Claude skills + fal to generate 3DGS environments, meshes, interactive physics objects and SFX from one image.

How It Works

IMAGE-BLASTER uses Claude Code skills to orchestrate multiple generation systems together. Dynamic objects are identified and converted into standalone 3D meshes, while those same objects are removed from the source image to generate a clean static environment.

Marble generates the explorable Gaussian splat world, while additional models handle mesh generation, image cleanup, and sound creation. The resulting assets can then be viewed, posed, edited, or exported into realtime engines and DCC workflows.

Stack

• Marble 1.1 for world generation and Gaussian splat reconstruction
• Claude Code skills for orchestration and workflow control
• Hunyuan 3D for mesh generation
• Nano Banana + GPT Image for image editing and cleanup
• ElevenLabs SFX for ambient and object-based audio generation

Generated assets can be exported into Unity, Unreal, and Three.js

What’s Next

• Expanded realtime editing workflows
• More advanced object extraction and scene understanding
• Faster prompt-to-world pipelines for games, robotics, film, and spatial computing

IMAGE-BLASTER is fully open source and available on GitHub.