The FOMM does not require a 3D model of the face. Instead, it learns to predict motion keypoints from a video. It maps the motion of a "driving video" onto a "source image."
You can often find it on platforms like Hugging Face or Paper with Code by searching for "First Order Motion Model checkpoints." How to Use It vox-adv-cpk.pth.tar google drive download
The file is a pre-trained neural network checkpoint used for image animation and deepfake creation. It is the essential "brain" behind popular tools like Avatarify and the First Order Motion Model (FOMM) , allowing a single source image to be animated using the motion from a driving video. What is vox-adv-cpk.pth.tar? The FOMM does not require a 3D model of the face
| Checkpoint | Dataset | Advantages | Download Source | | :--- | :--- | :--- | :--- | | | VoxCeleb | Faster, smaller, more stable | FOMM official drive | | taichi-cpk.pth.tar | Taichi HD | Best for full-body motion | FOMM official drive | | fashion-cpk.pth.tar | Fashion Video | Ideal for clothing animation | FOMM official drive | | best_net_G.pth (StyleGAN2) | FFHQ | Higher resolution (1024px) | NVIDIA / GitHub | It is the essential "brain" behind popular tools
If you cannot access this specific checkpoint, consider these alternatives: