cd deepfake_models/fsgan/
--ec mp4v
decide specific encoder, because you may counter errors when using the defaut encoder avc1
--gpus 0 3 decide specific GPU
python swap.py ~/videos/taylorSwift.mp4 -t ~/videos/jenniferLawrence.mp4 -o ~/videos/converted/ --finetune --finetune_save --seg_remove_mouth --gpus 0 -ec mp4v
currently, there is no interfaces for image2image face swaping. So please swith master(v2)
version into v1
version by fellow the instructions below:
git clone https://github.com/YuvalNirkin/fsgan.git
cd fsgan
git tag # to see all available versions.
git checkout v1
then download the weights for v1
version.
python download_fsgan_models.py
Images to Images Face Swapping
python face_swap_images2images.py <source images directory> -t <target images directory> -o <output directory>
!import the image version #must be .jpg
If you use .png
version, you'll get the error RuntimeError: No faces were detected in the source image directory
CUDA_VISIBLE_DEVICES=0 python face_swap_images2images.py ~/image/src/ -t ~/image/target/ -o ~/image/converted/