Every censored/mosaiced region in the videos are the results of downsampling. The simplest form of downsampling you can think of is you mark a region. This region is filled with a grid of blocks. Each block is filled with the average color value of the corresponding block in the original frame/picture. There are also variants, e.g. with additional blurring.
This knowledge seems to be used for creating the JavPlayer. The JavPlayer is using the neural network/AI called TecoGAN. TecoGAN is designed to upscale/upsample videos without the usual artifacts of standard methods. So in one step the JavPlayer downsamples the censored regions and upsamples them again with TecoGAN. This "destroys" the censored regions.
There is a problem.
TecoGAN was never designed to "decensor" or "remove" the mosaiced regions. It was designed to upscale/upsample videos. Nothing more and nothing less. Using this technique is like using a hammer for a screw. Sometimes it works, but mostly it fails and something is broken. That's why you see tons of deformed body parts in these videos. Sometimes this software even removes the pussy. LOL.
The downsampling to create the mosaiced regions in the videos is an operator that destroys information. Therefore you need something that actually can "reconstruct" lost informations. One example would be something like this:
https://news.developer.nvidia.com/n...e-reconstructs-photos-with-realistic-results/
You can compare it with TecoGAN:
https://ge.in.tum.de/publications/2019-tecogan-chu/
https://arxiv.org/pdf/1811.09393.pdf
You don't need to study computer science to see the difference. Looking at pictures / videos should be enough to understand what "reconstruction" and "upscaling" in the field of computer graphics / computer vision mean.
Using these neural networks/AIs also means you won't have a method to "reconstruct" the original uncensored movie. You really need a true uncensored leaked version of it. Such AIs can only "guess" how it looks like. A good example to understand this is the video on the nVidia page. You can actually see how the AI guesses the output. That also means the results of these neural networks/AIs will always be deepfakes. So the only hope here is to find a method to create "good" deepfakes, where you don't see the fake. The JavPlayer is far away from creating deepfakes which are actually "good".
Therefore, here is my advice. Don't waste any money for this shovelware and one of these deepfake videos.
The user GekkoE7 is currently posting lots of these deepfakes in the Uncensored Jav Downloads.
Example:
https://www.akiba-online.com/thread...wa-misa-love-toy-behind-closed-doors.1949601/
Look at 25:17. This is JAV 2.0 in its finest form. LOL.