Practical overview
When it fits
Use the repository when you need research code and a documented sample-data path for generating portrait views from different viewpoints.
Who it is for
- Researchers exploring diffusion-based portrait view synthesis.
- Developers prototyping portrait-generation experiments or sample-data demonstrations.
Documented uses
- Generate novel portrait views with the supplied sample-data inference script.
- Build research prototypes for controllable portrait view synthesis.
Requirements & limits
The repository is public and declares an MIT license. The pretrained checkpoint is downloaded separately from Google Drive.
- The Conda environment is described for Python 3.8.5 and CUDA 11.7, with other compatible versions noted as possible.
- For in-the-wild input, the 3D-aware noise model is not provided; the README recommends obtaining 3D-aware noise from another pretrained 3D GAN method.
- The README identifies V100, A100, and A6000 GPUs as hardware on which the code was tested.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Prepare the sample-data inference environment
Set up the documented Conda environment, install the pretrained checkpoint, and run the repository's sample-data inference script.
Before you start
- A Linux or server environment with Python and Conda.
- The repository's diffportrait3D.yml environment file.
- The pretrained checkpoint downloaded from the linked Google Drive file.
- A CUDA-capable NVIDIA GPU; the README lists V100, A100, and A6000 as tested hardware.
Steps
Download the pretrained checkpoint and place it at checkpoints/model_state-540000-001.th within the repository.
Create and activate the environment with conda env create -f diffportrait3D.yml and conda activate diffportrait3D.
Run bash script/CVPR_Inference/inference_sample.sh.
Expected outcome
The documented sample-data inference workflow is invoked using the pretrained checkpoint.
Guide limitations
- The README lists Python 3.8.5 and CUDA 11.7 as the authors' environment and says other compatible versions may work.
- The README does not include the 3D-aware noise model needed for inference on in-the-wild data.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 17 October 2026
- Original website title
- FreedomGu/DiffPortrait3D · Official homepage title · retained in its source language