Zero123++: a Single Image to Consistent Multi-view Diffusion Base Model

By Ruoxi Shi, Hansheng Chen, Zhuoyang Zhang, Minghua Liu, Chao Xu, Xinyue Wei, Linghao Chen, Chong Zeng, Hao Su. arXiv.org, 2023

608 citations recorded by the source corpus.

We report Zero123++, an image-conditioned diffusion model for generating 3D-consistent multi-view images from a single input view. To take full advantage of pretrained 2D generative priors, we develop various conditioning and training schemes to minimize the effort of finetuning from off-the-shelf image diffusion models such as Stable Diffusion. Zero123++ excels in producing high-quality, consistent multi-view images from a single image, overcoming common issues like texture degradation and geometric misalignment. Furthermore, we showcase the feasibility of training a ControlNet on Zero123++ for enhanced control over the generation process. The code is available at https://github.com/SUDO-AI-3D/zero123plus.

Subfields: Computer Graphics and Computer-Aided Design, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/186c2b2386799fa599e2d77bc6945f7da66e5fab

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Graphics and Computer-Aided Design

In Computer Vision and Pattern Recognition

Citations in the corpus

Gen3C: 3D-Informed World-Consistent Video Generation with Precise Camera Control (2025, neutral): “While several methods have been proposed to address this challenge through fine-tuning with images, additional text prompts or camera parameters [16, 31, 47, 53, 59, 60], achieving precise control for subtle or complex camera movements or scene arrangement remains unsolved.”
One-2-3-45: Any Single Image to 3D Mesh in 45 Seconds without Per-Shape Optimization (2023, methodology): “In addition to the density ﬁeld representation, some methods such as SparseNeuS [40] and VolRecon [62] utilize SDF representations for geometry reconstruction.”
CAT3D: Create Anything in 3D with Multi-View Diffusion Models (2024, methodology): “Methods like MVDream [43], ImageDream [9], Zero123++ [44], ConsistNet [45], SyncDreamer [46] and ViewDiff [47] fine-tune text-to-image models to generate multiple views simultaneously.”
LGM: Large Multi-View Gaussian Model for High-Resolution 3D Content Creation (2024, methodology): “During inference, our method leverages existing image or text to multi-view diffusion models [27, 43, 44, 51] to produce multi-view images as inputs for our Gaussian fusion network.”
GaussianDreamer: Fast Generation from Text to 3D Gaussians by Bridging 2D and 3D Diffusion Models (2023, neutral): “Some single-image-to-3D meth-ods [13, 19, 30, 33–36, 38, 54, 60, 65, 66, 71, 78, 79, 82, 83] also employ similar ideas.”
CLAY: A Controllable Large-scale Generative Model for Creating High-quality 3D Assets (2024, neutral): “Latest developments [Blattmann et al. 2023; Li et al. 2023; Liu et al. 2024a; Long et al. 2024; Qiu et al. 2024; Shi et al. 2023, 2024] have focused on directly generating multi-view images with view consistency, by employing enhanced attention mechanisms.”
ImageDream: Image-Prompt Multi-view Diffusion for 3D Generation (2023, background): “Later works, including SyncDreamer [20], Consistent 1-to-3 [52], and Zero123plus [36], have sought to enhance multi-view consistency through joint diffusion processes, but the reconstruction of geometrically coherent 3D models remains a challenge.”
Hunyuan3D 2.0: Scaling Diffusion Models for High Resolution Textured 3D Assets Generation (2025, positive): “On the other hand, to regularize potential style bias introduced by the 3D rendering dataset, we abandon the shared-weight reference-net as in [88, 78] and freeze the weights of the original SD2.”
GRM: Large Gaussian Reconstruction Model for Efficient 3D Reconstruction and Generation (2024, methodology): “3.4, GRM can be used for single image-to-3D generation by combining it with image-to-multiview diffusion models, such as Zero123++ [79].”
GS-LRM: Large Reconstruction Model for 3D Gaussian Splatting (2024, methodology): “For the image-to-3D application, we use the image-conditioned multi-view diffusion model in Zero123++ [54], which generates 6 structured views at fixed viewpoints.”
Evaluating Real-World Robot Manipulation Policies in Simulation (2024, positive): “…segmentation mask; (4) “Unpro-ject” the real object’s RGB texture values onto the simulation object mesh; (5) Optionally, generate the remaining views of the object through a diffusion model (Zero123++ [60]), and refine the poses of novel views using a rendering loss with the existing object view.”
InstantMesh: Efficient 3D Mesh Generation from a Single Image with Sparse-view Large Reconstruction Models (2024, background): “To address the inconsistency among multiple generated views of Zero123, some works [24, 26, 42, 51] try to fine-tune 2D diffusion models to synthesize multiple views for the same object simultaneously.”