About cvlab-columbia/zero123

The official release of Zero-1-to-3: Zero-shot One Image to 3D Object (Liu et al., 2023). See the paper page for context.

Zero-1-to-3 fine-tunes Stable Diffusion to perform viewpoint-conditioned novel view synthesis from a single image, then uses the resulting model as a diffusion prior for Score Distillation Sampling (SDS) lifting to 3D. The repo provides the Objaverse-trained checkpoint and inference scripts that downstream methods in the corpus — Magic123, DreamGaussian, SyncDreamer — depend on. Mirrored under MIT — README only.