content rules

These rules apply to every prompt, every reference image, every generated asset, every model selection, and every tag. Mirrors the headline list in Terms §4, with the operational detail moderators apply day-to-day.

Real people — zero tolerance

Phantasm does not generate, host, or distribute depictions of real individuals — living or deceased, public or private, famous or obscure.

• Naming a real person in a prompt is grounds for refusal. Repeated attempts after a refusal escalate to suspension.
• Reference photos of a real person are not accepted for img2img, face swap, or LoRA training.
• Fictional characters whose canonical depiction is an actor's likeness (Iron Man → Robert Downey Jr., Wonder Woman → Gal Gadot, etc.) fall under this rule when the output resembles the actor.
• Public-domain historical figures are not exempt. Living politicians and elected officials are explicitly not exempt.
• An adult performer who appears in nude work professionally has not consented to AI generation; do not treat their public images as a consent signal.
• If the output coincidentally resembles a real person — same hair, eyes, jawline, but not a deliberate copy — it's fine. Coincidence is judged on intent and prompt, not pixel diff.

Reporting: anyone depicted without consent can email appeals@phantasm.app with the asset ID; we remove confirmed matches within 48 hours.

Minors — zero tolerance

Phantasm does not generate any sexualised, suggestive, or inappropriate depiction of any person under 18, real or fictional. The standard applies identically to fictional characters of any species, art style, or stated-but-implausible age.

• No photorealistic depiction of any minor, in any context, sexual or otherwise. The risk that a real child's likeness appears is too high.
• Tags / aesthetics including loli, lolicon, shota, shotacon, or any equivalent are prohibited.
• Black bars, blurring, side-eye framing, or other self-censoring does not redeem otherwise-prohibited content.
• Minor-and-adult interactions must be wholesome and contextually innocent — school, family, sport, classroom. If a moderator would not show the image to a child's parent, it doesn't ship.
• Stated age in prompt does not override visual cues. "She's 19" attached to a clearly underage character is not a defence.

Where confirmed content depicts a minor in a sexualised context, we file the required report with NCMEC, preserve evidence, and permanently terminate the account. Zero strikes. No appeal.

Sexual content

Adult sexual content between adults is permitted on tiers that allow it and subject to the limits below. Photorealistic adult sexual content may be additionally restricted by your platform's app-store policy.

• Consent. No depictions of rape, sexual violence, incest, or sexual activity under intoxication, drugging, hypnosis, mind control, or other impaired-consent framing.
• Bestiality. No sexual content involving animals.
• Excretions. No urine, faeces, vomit, menstrual fluid, smegma, or diaper content in any context.
• Weapons in sexual context. No weapons aimed at the subject of sexual content.
• Substances. No content depicting active drug use during sexual activity.
• Metadata. Sexually explicit generations must carry the AI-generation metadata block Phantasm attaches. Stripping or forging it before re-sharing is a violation.

Violence and gore

• No graphic depictions of human death.
• No brutal violence or graphic injuries presented for entertainment. Combat in a fantasy or action context is fine; bone, viscera, and dismemberment shot for shock value is not.
• No torture, no abuse of a non-consenting subject.

Self-harm and disordered eating

• No depictions or glorification of cutting, mutilation, suicide methods, or suicide-as-aesthetic.
• No anorexia, bulimia, or other disordered-eating content presented as desirable.

If you are struggling, please reach out: 113 (Netherlands), 988 (United States), or your local equivalent. The list at findahelpline.com is global.

Hate and extremism

• No content that attacks, dehumanises, or incites violence against people based on race, ethnicity, national origin, religion, disability, sex, gender, gender identity, sexual orientation, age, veteran status, or serious disease.
• No depictions of designated terrorist groups, no propaganda symbols, no glorification of mass violence.
• Critique, satire, and historical depiction of these subjects in news / educational context can be permissible; intent and framing matter.

Nudification and deepfake tooling

• No prompts, workflows, models, LoRAs, or reference uploads whose purpose is to remove clothing from a specific real individual.
• No prompts, workflows, models, or LoRAs trained on a specific real individual's likeness.
• Claimed consent of the depicted subject does not change the rule. We do not have a workable way to verify it.

Impersonation and fraud

• No labelling AI-generated content as authentic photography, document scans, news footage, or other non-synthetic media.
• No assuming another person's identity in profile, prompts, or generations.
• No bounties, solicitations, or commissions for any prohibited category.

Illegal activity

• No content that depicts or promotes the manufacture of illegal drugs, regulated weapons, explosives, or other unlawful goods.
• No content that violates sanctions regimes that apply to the operator (EU, U.S., UN).

How enforcement works

• Prompt-time classifier. Every prompt is screened by an LLM-based moderation pass before any GPU is dispatched. Hits return a refusal and a refund.
• Output-time classifier. Every generated image and video is screened before delivery. Hits are blocked and the credit refunded.
• Human review. Borderline classifier scores queue for a human reviewer. Reviewers see the prompt, the asset, and your moderation history — no more.
• Repeat-offender detection. Refusals are kept for 180 days. Patterns of attempts to circumvent the rules escalate to suspension regardless of whether any single attempt succeeded.
• Appeals. Within 30 days, email appeals@phantasm.app. Real-people / minors / CSAM decisions are not appealable.

Reporting

To report content you've seen on Phantasm, email appeals@phantasm.app with the asset ID (the gen_xxxxxxxx in the URL) and a brief description. We acknowledge within 48 hours and act on confirmed violations the same day.