AI Judge changed title after evaluation, original title: "World Labs releases image-blaster for single-image 3D worlds"
World Labs developer neilson open-sourced image-blaster, a toolkit that turns a single input image into a fully meshed 3D scene with Gaussian splatting, object meshes, physics colliders, lighting and audio. The pipeline uses Claude Code agents plus tools from Marble and fal.ai to analyze images, assign properties and generate effects. Output includes editable scene.json files and supports interactive desktop viewing. Demos converted photos of a wooden cabin, bookstore and Italian alley into navigable environments within minutes.
Fei-Fei Li#10
Turn a single image into a fully meshed 3D world in minutes 👀
Built by a World Labs team member, image-blaster combines Marble + Claude skills + @fal to generate 3DGS environments, meshes, interactive physics objects and SFX from one image.
learn more + try it yourself ↓
Many users expressed excitement over the open-sourced 3D generation toolkit for Claude Code because it reduces friction when turning images into interactive worlds with physics, while a few criticized glitches or paid API elements.
No Digg Deeper questions have been answered for this story yet.
Generate an interactive 3D world ✨
I opened this project in Cursor and uploaded an image of a cozy bookstore.
The agent then generated the environment, chose objects to turn into 3D meshes, added colliders, and created sound effects 🤯
open-sourcing a 3D gen toolkit for Claude Code
input image → environment, meshes, physics, lighting, & audio
its open source at https://github.com/neilsonnn/image-blaster
What the F
This is the coolest thing I’ve seen today and it’s blowing my mind
@mayfer can't wait to use this :)
https://vision-banana.github.io/
the skills harness against an input image, and generate meshes with a combination of image generation and 3D generation
and finally, there is a companion app to view and pose all the objects that were created
@liminal_warmth https://github.com/neilsonnn/image-blaster
@ChrisMasterton @fal You can reconstruct real world locations! Check this out
every object you create as a mesh is then removed from the input image to create the static version of the world
This project uses @theworldlabs to generate the world from an input image.
It selects items from the world that should be 3D objects, removes them from the image (using NB or GPT Image), and renders them with Hunyuan 3D.
@ElevenLabs provides sound effects and ambient noise.
Generate an interactive 3D world ✨
I opened this project in Cursor and uploaded an image of a cozy bookstore.
The agent then generated the environment, chose objects to turn into 3D meshes, added colliders, and created sound effects 🤯
@theworldlabs @ElevenLabs Huge props to @neilsonks for making this, he has a better explanation of how it works 👇
And also thanks to the guys at @fal for hosting almost every model so you only need two API keys for this entire thing!
open-sourcing a 3D gen toolkit for Claude Code
input image → environment, meshes, physics, lighting, & audio
@RiederBernhard Marble 1.1 from @theworldlabs
@neilsonks looks awesome!!
last model needed for pipeline is wiping shadows in the generated textures so lighting is dynamic for everything equally. should be relatively easy for someone to fine tune an image model to achieve it, or maybe even just a prompt operating on textures
image-blaster outputs: • explorable 3D environments generated by Marble • dynamic object meshes • ambient + physics SFX
Works with Three.js, Unity, Unreal and more.
Github: http://github.com/neilsonnn/image-blaster
@davidpantera_ click the pencil icon and you can edit pos/rot/scale and duplicate or delete meshes from the scene
delete the scene.json file that creates if you want to reset it
@neilsonks Great job Neilson. What model do you use to generate the gaussian splats? Keep rocking!
@LudditeDesign yes, talk to @morietschel
@theworldlabs @fal Now I’m going to have to go into the real world and take photos.
@neilsonks 👀👏
my take on this: for most vfx work, you don’t actually need unreal engine or some heavy production setup anymore.
you can create a marble world, add physics directly in the browser, record the scene, then run video-to-video style transfer models / video editing model
no dedicated vfx artist required. the whole workflow stays inside the browser.
this really feels like the future of vfx.
