Toonify Yourself is a StyleGAN-based cartoonization project released by Justin Pinkney, first made public on September 1, 2020. The project aims to transform ordinary face photos into cartoon characters with Pixar, Disney, or DreamWorks animation styles. Its technical core is StyleGAN network blending and transfer learning.
Purpose: Used to convert real face photos into cartoon style while preserving the person's identity features as much as possible, so that the generated result can both be recognized as the original person and have cartoon-style exaggerated features (such as large eyes, smooth textures, etc.).
Main capabilities: The project is based on a StyleGAN2 pretrained face model and uses a cartoon model fine-tuned by Doron Adler for layer swapping. Different layers of StyleGAN affect different features: low-resolution layers control head pose and face shape, while high-resolution layers control lighting and texture. By swapping the high-resolution layers, a blended model with cartoon structure but photorealistic rendering can be obtained. Users can run the entire process through the Colab Notebook provided by the author: first find the latent vector of the input face, then feed it into the blended model to generate a cartoonized version. The project shows multiple examples, indicating that the generated cartoon characters are highly consistent with the identity of the original person.
How to use: The author provides a Colab Notebook link where users can run the cartoonization process. In addition, the author later built a website that allows users to complete cartoonization within seconds, and said the website was once very popular, but did not provide the website's specific URL or current status. Users need to follow the steps in the Notebook, including inputting a face photo, searching for the latent vector, and generating a cartoon image. The specific installation commands and code content are not provided in the materials.
Applicable scenarios: Suitable for personal entertainment, social avatar creation, artistic creation, and other scenarios, especially for users who want to see a cartoonized version of themselves. Since the project is based on a StyleGAN face model, it mainly targets face images, and may perform poorly on non-face images or in cases of extreme angles, occlusion, and so on.
Limitations: The project relies on the StyleGAN model and requires certain computing resources (such as a GPU) to run; the availability of the Colab Notebook may change over time; the website may be unavailable due to excessive traffic; the model training data consists of only about 300 animated character images, so it may have limited performance in certain styles; no specific pricing or commercial use information is provided。
