Zero123++: 3D generator — what it does and what you need to run it

Zero123++ is an open source 3D generator created by the SUDO-AI team. The tool tackles the task of producing consistent multi-view projections of an object from a single image. The authors themselves describe the project as "a base diffusion model for generating consistent multi-view projections from a single image."

What it does

Based on an uploaded picture, Zero123++ generates several views of the object from different angles. These projections are consistent with each other, which makes them usable as an intermediate step for further construction of 3D geometry. The repository tags list image-to-3d and text-to-3d directions, but the main description focuses on generation from an image. The authors do not disclose the details of the text-to-3d implementation.

What you need to run it

The code is written in Python and distributed under the Apache-2.0 license. The repository is available at github.com/SUDO-AI-3D/zero123plus, and the model weights are at huggingface.co/sudo-ai/zero123plus-v1.2. The diffusers library is required to run it. The full set of weight files takes up 5.2 GB, with the largest one being 3.2 GB. According to the developer, running the example requires about 5 GB of VRAM. Running on your own hardware is possible; a cloud service is not required.

Who it is for

The tool will be useful for researchers and developers who experiment with image-to-3D pipelines and need multi-view projection generation as an intermediate representation. The open source codebase allows integrating the model into your own projects. For those expecting a ready-made 3D mesh as output, this solution is not suitable — Zero123++ produces exactly a set of flat images from different angles, not a final 3D model. It is also worth noting that the last code change dates back to February 2024, and the repository has not been updated since.

Zero123++ occupies its niche as a component of a larger pipeline for creating 3D content. It does not replace a full-fledged 3D editor and does not provide a ready-made model, but it solves the specific task of generating consistent views, which can then be passed further along the pipeline.