Honestly, I might have messed up. The training didn't capture the original's hazy feel, nor did it achieve the clarity of the AI-generated high-resolution images; it ended up falling short of both. It feels like I only preserved some of the characters' style, while the rest didn't turn out well. I'm still grateful to the netizens for providing a way to obtain the materials; I'll try to find uncensored, high-resolution original images next.