Create Viral Face‑Swap Videos with PaddleGAN’s First‑Order Motion in 3 Easy Steps

Recently the "Ant yahei" face‑swap meme has gone viral on social platforms. This article introduces the free, open‑source PaddleGAN toolkit, which implements the First‑Order Motion algorithm for facial motion transfer, allowing anyone to create their own Ant‑ya‑hei videos.

First‑Order Motion algorithm

The algorithm learns keypoints in an image in a self‑supervised way and transfers the facial motions of a driving video to a source image, without requiring prior knowledge such as a skeleton.

Using PaddleGAN on AI Studio

Step 1: Download PaddleGAN and install dependencies.

!git clone https://gitee.com/PaddlePaddle/PaddleGAN
!pip install -r requirements.txt

Step 2: Run the motion‑transfer command, replacing the driving_video and source_image paths with your own files.

!export PYTHONPATH=$PYTHONPATH:/home/aistudio/work/PaddleGAN && python -u tools/first-order-demo.py --driving_video ~/fullbody.MP4 --source_image ~/秃头乔哥.png --relative --adapt_scale

Step 3: Add audio to the generated video.

videoclip_1 = VideoFileClip("/home/aistudio/fullbody.MP4")
videoclip_2 = VideoFileClip("./output/result.mp4")
audio_1 = videoclip_1.audio
videoclip_3 = videoclip_2.set_audio(audio_1)
videoclip_3.write_videofile("./output/export.mp4", audio_codec="aac")

Other capabilities of PaddleGAN

PaddleGAN also provides lip‑sync synthesis, video/photo restoration (colorization, super‑resolution, frame interpolation), face‑to‑anime conversion, and many more AI‑driven image and video tools, all freely available.

For more details, source code, and community discussion, visit the PaddleGAN GitHub repository.