Generate a 3D environment from a single image - here's how

More and more Claude Skills are becoming open-source GitHub repositories, leveraging AI to create new apps and possibilities. This includes the image-blaster by neilsonnn.

What's behind the "image-blaster"?

Strictly speaking, it's largely World Labs behind it.

World Labs specializes in creating realistic 3D worlds from images or text. Gaussian Splatting on steroids, so to speak - generating spatial depth from a single point.

But there's more to it: Strictly speaking, it's a combination of Claude Code + World Labs + fal.

fal identifies objects in your scene and generates 3D models from them, which you can then place in your world.

Additionally, colliders are created that you can import to enable interactions.

If you run the repository with Claude Code, you're just a few prompts away from the finished 3D environment.

Instructions

1. Download GitHub Repository: image-blaster
2. Open Claude Code in the image-blaster folder
3. Obtain API codes for World Labs and fal and send them in the chat
4. In the image-blaster folder, there's a subfolder called "Input" -> upload your image there (AI or Real)
5. Continue following Claude Code's instructions and allow permissions.
6. Done! After about 10 minutes, your world, including 3D models, will be fully generated. You will find it in the output folder.

Review

I tried this out with this industrial image – a rather challenging source image – but I wanted to see its true capabilities.

The quality isn't quite usable yet, but see for yourself:

‍

‍

So, the quality isn't perfect yet, and admittedly, the input image wasn't ideal either.

But we all know how rapidly things can change these days. What do you think, where is this headed?

‍

‍

‍

‍