Michelangelo: 3D generator — what it does and what you need to run it

Michelangelo is an open-source 3D generator created by the NeuralCarver team. The tool tackles the task of creating 3D shapes from a text description or an image, relying on an aligned latent representation that combines shape, image, and text.

What it does

• Generates a 3D shape from a text description (text-to-3D).
• Generates a 3D shape from an input image (image-to-3D).
• Uses an approach the authors call "alignment-before-generation" — aligning shape, image, and text representations before the generation stage.
• Based on transformer models.

What you need to run it

• The code is written in Python and is open source under the GPL-3.0 license.
• The repository is available at github.com/NeuralCarver/Michelangelo.
• The latest code change is dated April 10, 2024; the repository was created on June 29, 2023.
• The authors do not specify exact hardware requirements (GPU, VRAM, system RAM) or the list of dependencies in the fact sheet.

Who it suits

The tool will be useful for researchers and developers working with generative 3D shape models who want to study or refine the method of aligning latent representations. The open codebase allows adapting the solution to your own tasks. For those looking for a ready-made application with a one-click graphical interface, Michelangelo is not suitable — it is research code that requires setting up the environment and running it yourself.

The authors presented Michelangelo at NeurIPS 2023. Additional materials and a description of the method can be found on the project website: neuralcarver.github.io/michelangelo/.