Toonify Yourself was made by Justin Pinkney and Doron Adler for fun and amusement using deep learning and Generative Adversarial Networks (StyleGAN for image pairs and Pix2PixHD for training).
Amazing StyleGAN project. Fun for all the family.
*the website does not store image data*
How does this work?
This toonification system is made using deep learning. It's based on distillation of a blended StyleGAN models into a pix2pixHD image to image translation network.
How did you come up with this idea?
This all started from some earlier experiments Doron shared on Twitter that got a lot of interest. For a description of those early experiments see this blog post on self toonification, the idea is the same, but the method is different (and not really suitable for hosting as a webapp).
Going even futher back in time, this was all based on some new methods of mixing StyleGAN models I came up with trying to make realistic Ukiyo-e portraits
How are you paying for this? Wasn't it costing too much?
We are now generously supported by DeepAI who are running the neural network backend for us. Although DeepAI are now doing most of the heavy lifting, there is still a bit of cost in running the site, and we're very grateful for those who have supported us on Ko-fi.
How do I get good results?
The algorithm works best with high resolution images without much noise. Looking straight on to the camera also seem to work best. Something like a corporate headshot tends to work well.
Do you store my photo?
We don't store any of the images uploaded or generated. We send the image to DeepAI's servers to run the network and then show you the results. No imagery is stored on our systems.
My face wasn't found!
We use the open source dlib face detector to find faces, it's designed to pick up frontal faces but isn't perfect.
Where did my glasses go?
Not many characters from animated films wear glasses so the model seems to have learnt to mostly remove them.
Can I use this model for my own project?
DeepAI provide an API that you can use to integrate this into your own project. It's what we're using to run this site!
Original tweet by Doron:
https://twitter.com/Norod78/stat...
See additional coverage here:
https://gizmodo.com/i-have-seen-...
A technical deep dive:
MAKING OFF:
https://www.justinpinkney.com/ma...
Report
Good work and idea, like :]
Result for me is not so good as I think. Will return in few month to try again)
@smredkey Well it's been a few weeks and there's a prototype of an upgraded high resolution version you can try out at: https://toonify.photos
Report
Really cool, I understand that is a WIP and still some stylizations are a bit clunky, but there is a load of potential, myself as an Illustrator I am wondering if there is a way to improve the parameters of stylization when / how a face feature could look better, I am interested to see how far this can go. let's chat!
Like this idea a lot as I'm building a business based on personas and some feedback has suggested that photos of people are a bit too precise. This looks like a great balance - real enough so not a cartoon
Fun! And much faster than other avatar products I've tried. I wish it exaggerated some key features a little more and others less (ex. DiCaprio's eyes are a little hooded and squinty in the example photo but his toon version has big round eyes and giant eyebrows).
Cannabiscope
Toonify
Cannabiscope
VYVE
Cannabiscope
Product Hunt
Product Hunt
Cannabiscope