The generation process can be divided into three stages:
1. Input preparation
- Select a frontal portrait (JPEG/PNG) with a resolution of 512 x 512 or higher
- Place the image into the
input_image/file (paper) - intend
dataset.jsonCamera profile (see PanoHead cropping guide)
2. Executive reasoning
- Go to the code catalog:
cd diffportrait360_release/code - Run the script:
bash inference.sh - Enable ControlNet (optional): Modify the script to add the
--use_controlnetparameters
3. Results processing
- The output contains a sequence of multi-angle views (saved by default in the
output/) - Obtain .nerf model files that can be rendered with tools such as NeRFStudio
- Image sequences can be converted to video with tools such as FFmpeg
For customized data, it is necessary to ensure that the camera parameters are accurate, and it is recommended to use the test data provided by the project (including 1000+ real portraits) as a reference standard.
This answer comes from the articleDiffPortrait360: Generate 360-degree head views from a single portraitThe
May not be reproduced without permission:AI productivity tools " What is the exact procedure for generating a 360-degree view using DiffPortrait360?
English 
简体中文 
日本語 
Deutsch 
Português do Brasil