Mobile apps or web tools for creating cartoons there are many. Those that work from a photograph are usually not something that goes beyond the application of a filter or decorative effect. However, there are also options that offer more striking and original results.
One application that fits perfectly in the latter category is Toonify, a playful tool to caricature faces from an image, which works on the basis of a deep learning system and Generative Adverse Networks (StyleGAN).
What started as an experiment shared by one of its creators through Twitter, ended up being a web application itself, given the good feedback received.
Using a simple form, simply upload a photograph of a face, ideally with all its visible features, to obtain a caricature. Unlike other alternatives, such as those mentioned generically at the beginning, this system analyzes the parts of a face, so based on those data build the illustration it processes.
As a result, you can get a foreground cut of the processed image, with a version of it similar to the appearance of an animated movie character. The illustration obtained will respect certain elements, such as the color of skin and hair (depending on how it looks in the photograph) and will alter others, such as the accent of certain features, following a caricature technique and aesthetic.
For best results, it is recommended to load high resolution images, with the least possible noise. This reduces the possibility of the algorithm failing. Other common difficulties occur with the use of glasses, which is recommended to avoid when loading an image to this tool.
At the moment, Toonify Yourself is able to process only one face at a time. If a photograph with more than one visible face is loaded, the AI will select the first one it detects. If the algorithm is not able to identify a particular face, this may be due to limitations of the open source dlib face detector used by the application.
By the way, it should be noted that this mechanism works only with human faces. If you try to use a photograph of your pet or other object, you will get bad results.
If you access the web app from your mobile, in addition to the ability to upload images from your gallery, you can use the camera and capture one instantly.
The heart of Toonify Yourself is a deep learning system, based on the distillation of StyleGAN models, combined into a network of image-to-image translation pix2pixHD.
The processing of the images is in charge of DeepAI, who through their infrastructure are running the backend of the neural network that powers this service.
At the privacy level, Toonify Yourself ensures that the loaded images are not stored on their systems, being transmitted only to the DeepAI servers to submit them to the corresponding processing.
Justin Pinkey, co-creator of this tool, in his personal blog keeps a log about his works with artificial intelligence, having an important space dedicated to the technology behind the system that makes it work.
If you are dedicated to the development of applications and you are interested in adding the functions of Toonify Yourself in any of your creations, the source code of the tool was released by its authors in GitHub, for free consultation.
The tool is free to access and can be visited on toonify.justinpinkney.com.
