The game contains several duplicated textures and assets. And worse yet, practically for all but Live2D scenes use several 4k pictures rather than 1 background and changing character assets, so like non-programmer made Ren'Py games tend to do. Granted, that's much faster to devlope, but... well, we can see how it affects performance. For example, these are all 4K images that are mostly the same, but with small differences between them:
View attachment 4942357
Thankfully, the images are compressed. But .jpg is still a raster format, so in practice that means Ren'Py is decompressing them to be rendered. And one 4K image without transparency would take just under
100MB 48MB of RAM (did math bad, fix'd). So that's probably at least part of the problem. Using Live2D properly would most likely fix these problems.
I would propose a proper fix, but I have no idea what Ren'Py is capable of and not interested in learning lol, but compressing files into something that the CPU has to decompress to render is only going to help download time and in case storage speed is the bottle neck, it could also help load times. But it's more CPU taxing and just as memory heavy as not compressing them.