cvlab-columbia/zero123

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

3058 stars · 221 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Zero-1-to-3 is a research codebase and pretrained diffusion model from Columbia CVLab that changes the camera viewpoint of an object from a single RGB image, enabling zero-shot novel view synthesis. It can also be used for single-view 3D reconstruction by training a NeRF with its viewpoint-conditioned diffusion model.

Use cases

• generate new views of an object from a single photo
• reconstruct a 3D model from one image
• synthesize novel camera angles of objects zero-shot
• run viewpoint-conditioned stable diffusion inference
• train a NeRF using diffusion priors
• use zero123 weights in threestudio or stable-dreamfusion

When to choose

• you need zero-shot novel view synthesis from a single image
• you want a pretrained checkpoint for image-to-3D research
• you have a GPU with ~22GB VRAM (RTX 3090/4090)
• you want to integrate viewpoint diffusion into NeRF or SDS pipelines

When to avoid

• you need production-grade, actively maintained software
• you lack a high-VRAM GPU
• you need fast real-time inference
• you need a polished end-user application rather than research code

Facets

library · maturity maintenance

machine-learning image-processing stable-diffusion computer-vision computer-vision deep-learning artificial-intelligence graphics python image-to-3d novel-view-synthesis single-view-reconstruction diffusion-models zero-shot nerf iccv-2023 research-code gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="cvlab-columbia/zero123")