As fated courage did a waifu HD pack, maybe it would be good to ask him about the checkered effects.
We both use completely different models for our backgrounds. I'm pretty sure he used waifu, which basically denoises and 'cleans' for lack of a better word, so tbh, I doubt he ran into the same issue as I have, or at least a lot less. Since my own model adds 'noise' and guesses on detail to add, my own backgrounds tend to show this 'checkered' effect more. I do know how to fix it (at least I think I do, it worked on one example I tested at least), and may even find a way to fix it in my scripts so I don't have to fix them all by hand (which will require rebuilding all the images again), but currently I'm focused on the language issues, once I have that worked out I will begin focusing on the backgrounds again. The unfortunate part is that I will (again) have to extract the data files for another language (this is a very manually intensive pain in the butt process), which is time consuming, but I'm just about finished that as I'm typing this. Once I have extracted another languages data files, I'm hoping that a comparison of the files will produce a smaller list of exactly which files are different so that I can easily 'pull' and modify just the files necessary for that language (each language basically has every file, so I need to compare them all, so either I do that, or rebuild all the screens again for each language, which clearly sucks). Ultimately, just for convenience, I may go that route, but I'd rather not, since it will require extracting the data for each language, and not just the necessary files, but each file, which basically takes a day for each language and it's far from an automatic process.
I know that RichterB already posted an attempt to identify the different files, but at closer inspection, I'm pretty positive the list he gave me is unfortunately not completely accurate (I have no idea why), I believe what he did was fine, and I will be doing it again in a very similar way. The main reason why I believe this however, is that in the list he gave me it did not include (from what I could tell) screens that had text that should of been flagged. For example, the computer screen in the classroom, those images weren't present, or I didn't see them when looking at the images. I didn't look at them all (and his list was producing language conflicts as expected), but when I eyeballed the file names, I'm pretty sure those screens in the classroom are missing, which leads me to conclude there are likely more missing as well. Either way, I'd still rather check and compare it myself to be sure.
There are a lot of 'skipped' steps that I've done for the 'beta' version that likely won't get skipped in the final version. For instance, I will likely eventually generate a database on each image that will allow me to manipulate the images in a much more advanced way in my scripts, which will easily remove most of the problems I have already run into (specifically the checkered issue and any light/effects issues), but it's very time consuming to go over that many files and try to determine what each file is for, if I'm lucky, I'll be able to identify what screens have these issues and save time by not having to do that step, but only time will tell. Eventually, I think I'll end up making the database any way. But am still hoping to be able to find some 'standard' in the filenames produced by Pupu to be able to solve these issues instead (and already have 90-99% of the time for the lighting/effect issues as far as I can tell, otherwise the beta wouldn't even exist).
To explain the 'checkered' issue however, I saved a lot of time on the backgrounds, by using an archive posted here on qhimm that had 'most' of the original backgrounds already in it. The backgrounds in the game aren't actually one image, but many images layered on top of eachother (I know you already know this, am just stating it for clarification for others), and you must identify which files are needed for the background and merge them to get a full image. It was very simply to put the files in this archive in with the appropriate backgrounds and only required me rebuilding like 120 backgrounds myself by hand, because of this however, I haven't fully identified which files are used in each background. Every time my GAN produces an image, it slaps the image required on top of it's background. This prevents a border effect that generates on each image by instead using the whole screen (ie: the 'border' effect lands on the edges of the screen, instead of around the alpha image, which would appear very obvious if I didn't). Because each image has this background layered below each image, the background gets 'rebuilt' for each image. The GAN never produces an exact copy however, so in some situations, it produces this checkered effect whenever using screens that are part of the background, since the background is now built of 'slightly' different versions for each image that is part of that background. In order to fix this, I need to pull the background image files from a single image, instead of re-creating each image every time which is fine for most images, if that makes sense. Explaining this would be much easier to understand with a video demonstrating it, but this will have to do.