AdvancedLivePortrait-WebUI

If you do not want to install the project locally, you can use the following hosted environments:

• Google Colab: Use the provided notebook for a cloud-based execution environment.
• Hugging Face Spaces: Access the hosted WebUI directly on Hugging Face.

You can containerize the application using Docker Compose.

1. Clone the repository:

   git clone https://github.com/jhj0517/AdvancedLivePortrait-WebUI.git

2. Build the image:

   docker compose -f docker/docker-compose.yaml build

3. Start the container:

   docker compose -f docker/docker-compose.yaml up

4. Access the WebUI at http://localhost:7860/.

Note: If you are not using an Nvidia GPU, you must update the docker/docker-compose.yaml file to match your environment.

To run the WebUI on your local machine, follow these steps:

1. Clone the repository:

   git clone https://github.com/jhj0517/AdvancedLivePortrait-WebUI.git

2. Install dependencies. If you have an Nvidia GPU, use requirements.txt. If you are running on a CPU, use requirements-cpu.txt:

   pip install -r requirements.txt

3. Launch the application:

   python app.py

For Windows users, the project provides PowerShell scripts that automate the creation and activation of a virtual environment (venv), dependency installation, and app execution.

Available scripts:

• Install.ps1
• Start-WebUI.ps1
• Update.ps1

To use them, right-click the script file and select Run with PowerShell.

Before installing, ensure your environment meets the following requirements:

• Python: Version 3.9 to 3.12.
• CUDA (Optional): If using an Nvidia GPU, CUDA 12.4 is recommended.
• FFmpeg (Optional): Required if you are using the Video Driven feature. You must add the FFmpeg/bin folder to your system PATH after installation.

If you do not have an NVIDIA GPU, you must perform two steps to ensure the Docker container runs correctly:

1. Update Docker Compose: Modify the deploy section in docker-compose.yaml to match your specific device driver (refer to Docker Compose deploy documentation).
2. Update Dependencies: You must edit the Dockerfile to change the requirement installation from requirements.txt to requirements-cpu.txt to avoid attempting to install CUDA-dependent packages.

Launch the WebUI by executing app.py. Use the --share flag to generate a public Gradio URL, which allows you to access the interface remotely (useful for notebook environments like Google Colab).

!python app.py --share

To set up the environment, clone the repository and install the required Python packages including lmdb, ultralytics, tyro, dill, and gradio-i18n.

!git clone https://github.com/jhj0517/AdvancedLivePortrait-WebUI.git
%cd AdvancedLivePortrait-WebUI
!pip install lmdb
!pip install ultralytics
!pip install tyro
!pip install dill
!pip install gradio-i18n

Before running the application, ensure that your environment has a compatible GPU available, as some models may not function correctly on a CPU runtime. You can verify your GPU setup using the nvidia-smi command.

!nvidia-smi

The App class is the main entrypoint for the Gradio-based WebUI. It manages the LivePortraitInferencer, internationalization (i18n), and the Gradio interface layout.

To use the App class, you should pass an object containing configuration arguments (similar to argparse results) to the constructor. The launch() method starts the Gradio server with settings derived from these arguments.

import argparse
from app import App

parser = argparse.ArgumentParser()
parser.add_argument('--model_dir', type=str, default='path/to/models')
parser.add_argument('--output_dir', type=str, default='path/to/outputs')
parser.add_argument('--server_port', type=int, default=7860)
# ... add other args as needed
args = parser.parse_args()

app = App(args=args)
app.launch()

The Video Driven mode uses a driving video to animate a reference image. The following parameters control the alignment and quality:

• First frame eyes alignment factor: Alignment strength for eyes (0 to 1).
• First frame mouth alignment factor: Alignment strength for mouth (0 to 1).
• Face Crop Factor: Controls the face cropping area (1.5 to 2.5).
• Enable Image Restoration: Checkbox to enable RealESRGAN.