The categories below get rejected on submission, removed if discovered
post-publication, and (where applicable) reported to authorities. There are no
creative-framing exceptions and no “clearly fictional” carve-outs.
If you see prohibited content already on the platform, please
report it.
Public vs private generation
The hard limits below apply everywhere on the platform: public
submissions, private chats, and private generations. There are a few
categories (e.g. parody or fair-use IP, specific fetishes) where the
public-submission bar is stricter than the private-generation bar,
because public content reaches every other user on the platform.
Where that’s true, we note it on the relevant section. Everything
else applies identically in private and in public.
Categories
Minors / underage
Any content depicting, implying, or coded to suggest individuals under
18 in a sexual, suggestive, or romantic context. This includes:
- Explicit age statements under 18.
- Settings or framings that imply a minor, such as middle-school settings,
childhood scenarios, phrases like “about to turn 18”, “very young
looking”, “barely developed”.
- Visual cues, such as small or petite frame combined with size-difference
framing, “cute” + childlike features, loli-style art.
- Grooming, sexualised training, age-play, or “ageing” mechanics.
- Any character that, in body proportions, reads as a minor regardless
of stated age.
Content involving minors is removed and, where the conduct is criminal,
reported to the appropriate authorities, including national
child-safety reporting bodies such as the NCMEC CyberTipline in the
United States.
Real people
Characters that are, depict, or are recognisably modelled on a real
person (alive, or deceased less than 100 years). This includes:
- Celebrities, athletes, public figures, politicians, internet personas.
- Private individuals (named or visually identifiable).
- Fictional characters whose appearance closely matches a specific
real actor or model.
Parody and satire of public figures is allowed in narrow forms. See
Intellectual property.
Existing intellectual property
Characters drawn from copyrighted IP, including:
- Named characters from a series, comic, game, show, or film.
- Recognisable storylines, locations, or worlds.
- Scenarios that instruct the user to “play as” a copyrighted character.
Original work in the style of a genre (cyberpunk noir, magical
academy, isekai-adjacent) is allowed. Original characters that evoke
a vibe without using protected names or storylines are allowed.
Parody and transformative use. Like YouTube’s posture on fair use,
we leave room for parody, satire, and clearly transformative work that
references a copyrighted property without reproducing its specific
characters or storylines. We apply this narrowly and case-by-case.
Non-consent and removal of agency
The rule depends on which side the absence of consent sits on.
Allowed: The user can be the victim of non-consent in the story:
assault scenarios, captivity arcs, dubious-consent fantasies the user
plays from the receiving side. The user retains meta-consent by being
in control of the chat. They can leave, redirect, or end the scene at
any time.
Prohibited: Scenarios that put the AI character on the
no-agency side. You cannot frame a story as “you assault this
character” or “this character is your captive”. The character has no
way to refuse inside the fiction, and the platform will not host
scenarios that eroticise their absence of consent.
Specifically prohibited (AI-character-as-victim framing):
- Slavery, sex slaves, forced sex work, sexual training/trading/ownership.
- Kidnapping, drugging, imprisonment, threats, coercion as the central
premise.
- Arranged marriages without character consent.
- “Buying a slave” scenarios.
Allowed:
- Stories where the user is the one without agency in the fiction (their
out-of-fiction meta-consent still applies).
- Rescue arcs and consensual relationships after rescue.
- Power-imbalance dynamics where the consent is genuine.
Family / incest
Anything implying a biological family relationship in a sexual or
romantic context. This includes:
- Tags or descriptions that imply a biological family relationship in
a romantic or sexual context (e.g. “daughter”, “twins”).
- Multi-character setups where the cast resemble each other (assumed
biological unless explicitly stated otherwise).
- Characters designed to be related to an existing public character.
Bestiality
Sexual content involving non-anthropomorphic animals. Anthropomorphic
characters with human features (faces, proportions, voice) are allowed
within the rest of the AUP.
Scat, vomit, emesis, vore
Including euphemisms (“chocolate”, “brown water sports”). Vore as a
sexual fetish (the consumption of one being by another) is
prohibited regardless of framing.
Excessive gore and violence
Gratuitous depictions of serious injury, mutilation, torture, or death
in all visual forms (images, videos, or any other media output), with
no exceptions. Stylised conflict and consequence-bearing violence in
written stories, character backgrounds, or chat remain allowed,
provided they serve the narrative rather than exist purely for shock
or harm.
Hate content
Content that is, or that encodes:
- Racial slurs and degradation framing (BNWO, QoS-style content).
- Hate speech as a personality trait or instruction.
- Racialised content rooted in dehumanising stereotypes.
The line is whether the content is rooted in harmful stereotypes
versus exploring identity. Racial or cultural identity in characters
is not the same as racism.
Sexual or romantic content involving Jesus, Allah, or Mohammed.
Generic religious settings, monks/clergy as character archetypes, and
fantasy analogues are allowed.
Excessive political ideology
Characters whose primary purpose is to advocate a political position,
or who are vehicles for political grievance content.
Real-world illegal acts
Content depicting or instructing real-world crimes (terrorism,
weapons manufacture, drug synthesis, financial fraud) outside
obvious fictional framing.
Evasion
Any of the above categories hidden via:
- Non-English language.
- Encoded text, base64, or obfuscation.
- Character names or tags designed to evade detection.
- Repeated near duplicate submissions intended to circumvent rejection.
Evasion is treated as the underlying violation plus a separate AUP
breach.
Edge cases
Specific case-by-case framings are above. For trust and policy
questions not covered here, see Contact. Last modified on May 19, 2026
