Hey! I'm concerned that the new AI deep pixelization tool is based on this code and model: https://github.com/WuZongWei6/Pixelization
If this is the case it would be a violation of the non-commercial license unless you have arranged a deal with the author.
Is this the case?
Hi Actually you are right. While I was dealing with the subject of Stable Diffusion, I saw this plugin and I liked it. I rewrote it entirely in C++ with libtorch, but frankly, I didn't pay much attention to the licensing part. I added it for my own use, but then for some reason I decided to put PixageFX. Thank you for your sensitivity and warning me. I will remove this module in the next version as soon as possible. By the way, as far as I understand, the 'data' part in the license is because the model is made with data that may be licensed while it is being trained. Because other than that, I don't use the codes directly. Maybe if I retrain with data that has no license problems like you did, it might not be a problem. I will discuss this with the author.
By the way, your aseprite plugin inspired me too. I think you created it with Pixel Art Diffusion. In the next version, I will add Pixel Art Diffusion module in PixageFX Studio with a model that has no license problems. It has already been something users have been demanding for a while. Thanks again.
I've actually been training and releasing models for about 10 months now, all trained on my own artwork and other licensed assets, nothing to do with Pixel Art Diffusion (though KaliYuga is a friend and inspiration). You might actually be interested in this downscaling algorithm I designed myself, its not quite as good at turning things into pixel art as Pixelization, but it is MIT licensed. https://github.com/Astropulse/K-Centroid-Aseprite
Thank you for detailed information.
Although artificial intelligence is very exciting, I am much more satisfied with the Pixelate algorithms that I have prepared completely at PixageFX. Because it gives undesirable results in artificial intelligence models. For example, for the Deeppixelate module to be successful, the input image must also be suitable for it. It also distorts colors and values such as gamma/contrast. I went through the link you gave and it's pretty clever. But all the downscale algorithms I've seen don't scrutinize the image much. In addition, the entrance image should have some preliminary preparation. Usually this part is skipped. the algorithm in the link seems to have just destroyed antialias. This is an algorithm that has already been available in PixageFX from the very beginning.
