ConsistNet: Enforcing 3D Consistency for Multi-view Images Diffusion

ConsistNet is a multi-view consistency plug-in block for latent diffusion models to improve 3D consistency.

Given one reference image, it can generate 16 multi-view consistent images in just 40 seconds on an A100 GPU.

Paper | Project Page

Plan

• Inference Code
• Pretrained Model
• Live Demo
• Training Code

Pretrained Model

Coming soon!

More Examples

Citation

If you find this project useful for your project, please cite:

@article{yang2023consistnet,
        title={ConsistNet: Enforcing 3D Consistency for Multi-view Images Diffusion},
        author={Yang, Jiayu and Cheng, Ziang and Duan, Yunfei and Ji, Pan and Li, Hongdong},
        journal={arXiv},
        year={2023}
    }