Ah yep, Flagged for moderation. Despite them both being FAR more SFW than the 2 that passed. lol.

Yeah, I read not long ago that the auto-detection system is extremely sensitive to certain words, even if they have nothing to do with NSFW content. For example, if you pair Pan with Gohan using his Danbooru tag "son_gohan", the word "son" can trigger a flag and get your image banned instantly. The same thing happens with terms like "school" or anything loosely related to minors. It is frustrating, especially when the image is totally harmless. Honestly, I have seen much better moderation systems on smaller, cheaper platforms. :S

On top of that, it is surprisingly easy to bypass the first auto-detection layer just by doing a second pass and stripping out any prompts that might trigger the system. Ironically, having no metadata at all now seems riskier. I am basically forced to reprocess all my images, especially when I use an external upscaler that wipes metadata. I learned that the hard way after a week of no reactions, even on PG-13 images, I realized they were shadowbanned. Private mode revealed they were invisible to everyone but me, with no warnings or notices whatsoever.

CIVITAI can be really sneaky and shady sometimes. Honestly, there is a chance this LoRA might disappear too, this is already the third Pan LoRA I have used. That said, you do not need a LoRA to recreate her. Just using tags like pan \(dragon ball\), medium black hair, bangs, black eyes, etc., works fine. I just prefer to use a LoRA for Dragon Ball style because it makes it easier to capture that retro style I am going for without needing extra tokens.

@Hyokkuda I know what you mean. A checkpoint can usually get a character, but having a specific LORA makes it 1000 times better.