White-box Cartoonization
CVPR 2020 photo cartoonizer using a clear, inspectable method
Upvotes
0
Pricing
Open Source
Category
Cartoon Generators
In the database since
Oct 2026
Overview
About White-box Cartoonization
White-box Cartoonization is the TensorFlow code for a CVPR 2020 paper by Xinrui Wang and Jinze Yu. It turns photos of scenery, food, rooms, and people into clean cartoon images. The method uses white-box cartoon features, which means the AI learns simple cartoon rules like smooth surfaces and clear edges. A Docker image lets you run it with no local setup. Inference also works on Linux, Windows, and Mac. Pretrained weights handle the cartoon look for you. A separate facial model exists for portraits.
Key Features
1. Works on Many Photo Types
It cartoonizes scenery, food, indoor scenes, and people. The README shows examples of cities, dishes, rooms, and portraits. One model handles all these cases.
2. White-Box Method
Most cartoon models are black boxes. This one learns three clear features: smooth surfaces, clear structure, and texture. You can inspect what the AI learned.
3. Docker One-Command Run
A Docker image runs cartoonization with no local installs. Mount your photo folder and an output folder as volumes. The container does the rest.
4. Facial Cartoonization Add-On
The authors released a separate model tuned for faces. It gives better portraits than the general model. Find it in the FacialCartoonization repo.
5. Full Training Code Included
Training scripts are included for a custom style. Prepare cartoon frames from films, then run the pretrain and train scripts. An NVIDIA GPU is needed.
Frequently Asked Questions (FAQ)
Why is it called white-box cartoonization?
Black-box models hide how they work. This method breaks cartoon style into three visible parts the researchers can check. That is the white-box idea: you can see inside the process.
Can I use the results in a commercial product?
No. The code uses a CC BY-NC-SA 4.0 license, which bans commercial use. You must keep the same license if you share or modify the code.
Does it work on portraits?
The main model handles people, but faces get a dedicated model called FacialCartoonization. It is a separate repo by the same authors. Use it for better cartoon portraits.
Conclusion
White-box Cartoonization stands out for its clear, inspectable method and tidy, charming results. It is a strong pick when you want to understand the process, not just the output. Use the Docker route for quick tests and the facial add-on for portraits. For learners and researchers alike, this repo teaches as much as it produces.
Keep exploring
Similar AI tools
Ranked by the job they do, their capabilities, and how closely they fit this workflow.