Skip to main content

Zero123-XL

Using Objaverse-XL, we train Zero123-XL, a foundation model for 3D.

The model works by taking in an input image and a camera transformation and predicts the output image after the transformation. The model is trained on 10M+ 3D objects from Objaverse-XL.

Below are some examples of the model in action:

zero123-xl

Weights​

The weights for the model can be downloaded with:

wget https://zero123.cs.columbia.edu/assets/zero123-xl.ckpt

These weights can be directly plugged into the original Zero123 model available on GitHub!

Image-to-3D​

Using Zero123-XL, we can perform single image to 3D. The implementation is available in threestudio, a popular unified framework for 3D content generation!