Skip to content

Latest commit

 

History

History
102 lines (72 loc) · 4.22 KB

README.md

File metadata and controls

102 lines (72 loc) · 4.22 KB
Raw

Wav2Lip UHQ Improvement Script

This repository contains a script designed to enhance videos generated by the Wav2Lip tool.

🔥 New Update : Automatic1111 extension can be found here, https://github.com/numz/sd-wav2lip-uhq with big improvement !!

Illustration

Result video can be find here : https://www.youtube.com/watch?v=-3WLUxz6XKM

Description

This script provides an enhancement to the videos generated by the Wav2Lip tool. It improves the quality of the lip-sync videos by applying specific post-processing techniques with controlNet 1.1.

Prerequisites

  • Stable diffusion webui automatic1111 + ControlNet 1.1 extension
  • Python 3.6 or higher
  • FFmpeg
  1. You can install Stable Diffusion Webui by following the instructions on the Stable Diffusion Webui repository.
  2. You can install ControlNet 1.1 extension by following the instructions on the ControlNet 1.1 repository.
  3. Download ControlNet model control_v11f1e_sd15_tile at [ControlNet Models]https://huggingface.co/lllyasviel/ControlNet-v1-1/tree/main and install it in controlnet models folder in automatic1111
  4. FFmpeg : download it from the official FFmpeg site. Follow the instructions appropriate for your operating system.

Installation

  1. Clone this repository.
git clone https://github.com/numz/wav2lip_uhq.git
  1. go to the directory
cd wav2lip_uhq
  1. Create venv and activate it.
python3 -m venv venv
source venv/bin/activate
  1. Install the required Python libraries using the command :
pip install -r requirements.txt

Usage

  1. Launch Stable diffusion webui with "--api" flag.
  2. Choose your model in stable diffusion webui.
  3. Run using the following command:
python wav2lip_uhq.py -f <file> -i <input_file>

Here is a description of each argument:

  • -f or --file: Path to the video generated by Wav2Lip.
  • -i or --input_file: Path to the original video.
  • -p or --post_process: if set to false script only create images and mask for alternative process

Operation

This script operates in several stages to improve the quality of Wav2Lip-generated videos:

  1. Mask Creation: The script first creates a mask around the mouth in the video.

  2. Video Quality Enhancement: It takes the low-quality Wav2Lip video and overlays the low-quality mouth onto the high-quality original video.

  3. ControlNet Integration: The script then sends the original image with the low-quality mouth and the mouth mask to ControlNet. Using the automatic1111 API, it requests ControlNet to perform a render on the mouth, thereby enhancing the final quality of the video.

Payload

in the file "payloads/controlNet.json" you'll find the payload send to automatic1111 api. feel free to change it to your needs. following parameters could drastically change the result:

  • denoising_strength (0.2 - 1.0) default 1, high value can create flickering, low value can create blurry result
  • mask_blur (0 - 50) default 8
  • alwayson_scripts > controlnet > args > threshold_a (1 - 32) default 1
  • alwayson_scripts > controlnet > args > threshold_b (1 - 32) default 32
  • inpainting_fill (0 - 3) default 2, 0 = fill, 1 = original, 2 = latent noise, 3 = latent nothing
  • steps (1 - 100) default 30, number of steps for diffusion

alternative usage

if you set -p or --post_process to "False", the script will only create images and masks. you can then use those folders in automatic1111 webui in img2img Batch mode: Illustration It will give you more control over the result

Quality tips

  • use a high quality video as input
  • use a high quality model in stable diffusion webui like delibarate_v2
  • play with the payload parameters

Contributing

Contributions to this project are welcome. Please ensure any pull requests are accompanied by a detailed description of the changes made.

License

Specify the open-source license under which your project is published here.

Contact

Provide your contact details here for any questions or comments about the project.