/roopiy

Primary LanguagePython

Roopiy

A video face swap tool using insightface

Acceleration

  1. Install CUDA Toolkit 11.8
  2. Download cuDNN for Cuda 11.x
  3. For windows go C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v{YourVersion}, copy the bin, include, lib folders from cuDNN to the CUDA folder, and add C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v{YourVersion}\bin to path

Installation

  1. clone this project
  2. pip install -e ..
  3. change the torch and torchvision in requirements.txt as you need, then pip install -r requirements.txt
  4. Setup a model path, e.g. temp (note: use absolute path!)
  5. download inswapper_128.onnx into your model path, from huggingface or here
  6. download buffalo_l.zip and unzip under temp/models/buffalo_l; you folder now looks like: temp/models/buffalo_l/*.onnx
  7. download GFPGANv1.4.pth into your model path
  8. download detection_Resnet50_Final.pth into your model path gfpgan/weights
  9. download parsing_parsenet.pth into your model path gfpgan/weights

Model Path

either use ROOPIY_MODEL_PATH={PathToYourModel} roopiy or roopiy --model-path={PathToYourModel} to set the model path

Basic Run

use roopiy to run. If it does not work, run python -m roopiy

check roopiy --help for construction