It's not a bug. You are using an external AI image generation model and they don't come with built-in censorship/moderation. Censorship (or even moderation) is a feature you have to implement into the system, you can't implement it into the AI image model. Usually, censorship is based off of words used in a prompt or maybe how the words are used in the prompt. If you didn't add a censorship or moderation feature into the system, then its going to produce CSAM/CP since there's no filter or guards. People have reported innocent prompt being rejected because the AI thinks the prompt is inappropriate when its not. Like:

• Sally, a curvaceous and busty woman, wears a bikini at the beach (Automatically rejected)

• A man is uses a gun fires to fire a bullet at a zombie head and blows its brain (Automatically rejected)

• A mother breastfeeding her newborn baby (Automatically rejected)

• A guy punches another guy's face, leaving it bloodied and bruised (automatically rejected)

These examples are either consensual/legal adults, non-extreme violence, a mother doing a natural thing for her baby, or fictional violence regarding fictional creatures. However these types of prompts still get rejected because the censorship thinks the words contained in the prompt or how the prompt is written in those words are inappropriate even though its not? That's poor censorship and moderation. Normal stuff like this shouldn't be censored, instead stuff like this should:

extreme pornographic/18+/sexual (terms describing extremely sexual actions, behavior, poses, positions, erm...bodily functions..., sexualized partial or full nudity, extremely skimpy or revealing outfits, body parts, effects, nonconsensual acts, actions, behaviors as well as sexual violence and abuse like SA, rp, etc) associated with children under 18 (newborn, infant, toddler, child, kid, preteen, teen, adolescent, <18 years/years old/aged/age or like aged 9 or age 9, little girl, little boy, young girl, young boy, you get the point) and even animals. That type of stuff. AI generated CSAM/CP is also illegal now although it wasn't before because Grok allowed people to edit users' posts containing photos without their consent and creeps and pedos were using it to generate horrible sexualized images of women and children (specifically young girls under 18).

Obviously, sexual violence and abuse is disgusting and immoral for any one of any age or gender...but if its fictionalized with fictional adult characters, that's up for you to censor or not...

I am not sure how feasible it would be for you to implement a censorship or moderation features at the moment...that's beyond something for a public community website like Perchance and a solo site runner like you. You need a lot of resources and money for that...as well as multiple moderators or staff helping you maintain that stuff...You probably won't be able to implement some type of censorship for children for quite some time or any other type of safety features...to be honest, a lot of this censorship and safety features is the creator's responsibility since they're the ones who made these and there wasn't any universal report/moderation/safety features to begin with. I would suggest taking your time coming in researching and coming up with a way censor harmful and illegal images of children. This may take a long while but it'll be worth it. I think the best bet for now is someone reporting these issues to you, whether there's something wrong with the AI image model, generators, text, etc. And if you can, the most simplest way for safety, is adding a basic report button. To flag down any illegal content someone or something is producing. The gallery, for now, is the best place to catch someone producing and posting AI gen CSAM/CP. I wish you the best of luck, man. This type of stuff ain't easy, even for larger websites ran by multiple people.