VToonify

Why we included this project

Most portrait toonifiers work on single images, so running them on video usually produces flicker and misaligned faces. VToonify gets around that by building on StyleGAN with a fully convolutional architecture that accepts non-aligned faces and variable frame sizes, meaning you don't have to crop and pre-align every shot before stylizing it. The result is high-resolution, anime-style video that keeps the subject's identity and motion consistent across frames, which makes it a practical fit for stylizing webcam or broadcast footage. Researchers and developers will find the official implementation, pretrained models, and Colab/Hugging Face demos a solid starting point, and the SIGGRAPH Asia paper documents the design choices behind the temporal consistency. One catch: the S-Lab license restricts use to non-commercial projects unless you contact the authors.