GitHub - neilsonnn/image-blaster: An image-to-world skillset for Claude.

Image-blaster is a command-line tool that converts a single photograph into a complete 3D environment with models, textures, sound effects, and physics data in under five minutes. You feed it an image, Claude walks you through the generation process step-by-step, and it spits out GLB/OBJ 3D models for dynamic objects, Gaussian splat files for static environments, and MP3 audio tracks for ambient sound and object-specific effects. The whole thing runs locally after you clone the repo and supply API keys for World Labs and FAL. It's genuinely useful for anyone who needs 3D assets fast—game developers, architects, filmmakers—without needing to model anything from scratch. The tool chains together several specialized models to do the heavy lifting. World Labs' Marble 1.1 creates the explorable 3D space from your image, Hunyuan 3D generates the actual object models, and ElevenLabs handles sound generation. You get control over output quality through parameters like face count (defaults to 50,000 but can go up to 1.5 million), PBR material generation, and polygon reduction for low-poly versions. The image editing happens through either a default cleanup tool or GPT-image-2 as an alternative, so you can prep your source image or create clean plates before the main generation runs. The output files embed directly into whatever you're building—Unity, Unreal, Godot, Blender, Three.js, or standard web apps. The workflow assumes you want Claude to confirm each step rather than automating everything, which means you can catch problems and adjust before committing to a full generation. If you want to tweak the React viewer that comes with it, you can remove `/app` from the `.claudeignore` file and let Claude modify that too.