Content moderation
Depicta operates within Greek and EU law. Every generation request passes through a content moderation system that evaluates prompts and images before anything is produced.
How it works
Section titled “How it works”An automated classifier reviews each request — both text prompts and uploaded images — and blocks content that would not be lawful to produce or that the Acceptable Use Policy does not allow. This happens before the image is generated, so blocked requests are not charged: a blocked request returns HTTP 451 content_rejected. If the image model itself returns no image, the API returns 451 model_content_rejected, also not charged.
Uploaded images used in edit, combine, reference and processing operations go through the same moderation process.
User responsibility
Section titled “User responsibility”Regardless of whether content moderation allows or blocks a request, you are responsible for the images you create and how you use them. An image that is lawful to generate may still be unlawful depending on its usage — for example, using a generated likeness of a real person in a misleading context. Content moderation cannot assess intent or downstream use.
Strictness modes
Section titled “Strictness modes”The moderation system has two modes:
| Mode | Description |
|---|---|
| Normal (default) | Brand-safe filter. Suitable for most use cases — professional, commercial, and personal projects. |
| Liberal | Wider latitude. For erotic and provocative art of fictional adults, and dark themes. Explicit sexual content is never allowed, and nothing sexual of real people or of anyone who could be a minor. Photorealistic images of real, identifiable people are not allowed in any mode; caricature and clearly artistic depictions are. |
Setting the mode
Section titled “Setting the mode”You can set the strictness mode in three ways:
- Account default — change your default mode in Settings. Applies to all requests unless overridden.
- Per request (CLI) —
depicta image "prompt" --strictness liberal - Per request (API) — include
"strictness": "liberal"in the request body.
Prohibited content
Section titled “Prohibited content”The following categories are actively detected and blocked by the automated classifier. For the complete policy — including additional prohibited uses that are enforced through account review rather than automated detection — see the Acceptable Use Policy.
Zero tolerance
Section titled “Zero tolerance”These result in immediate account termination and referral to law enforcement:
- Child sexual abuse material (CSAM) — Any sexualisation of minors, real or fictional, including drawn, animated and AI-generated depictions. Sexual content is refused wherever a person could be a minor.
- Non-consensual intimate imagery — Sexual, intimate or nude images of a real, identifiable person, in any style and whatever the consent claims.
- Terrorist content — Material that incites, promotes or glorifies terrorism, or gives instructions for terrorist acts.
Detected categories
Section titled “Detected categories”These result in content rejection and graduated enforcement:
This list follows the line the law draws. Normal mode blocks more than this, for brand safety; Liberal mode blocks what this list and the model providers’ rules require.
- Incitement to violence — Content that incites violence against specific people or groups, or threatens them.
- Hate speech — Content that incites violence or hatred against people because of their race, colour, religion, descent, national or ethnic origin, sexual orientation, gender identity or disability, and content that condones, denies or grossly trivialises genocide, crimes against humanity or war crimes. Historical, educational or fictional depictions that neither incite nor condone are allowed.
- Self-harm and suicide — Instructions for self-harm or suicide, or content that encourages a person to harm themselves. Awareness, prevention and artistic work on these themes is allowed.
- Harassment and threats — Threats, intimidation or stalking aimed at specific people, and degrading content about private individuals.
- Dangerous instructions — Actionable instructions for making chemical, biological, radiological or nuclear weapons, explosives or illegal drugs, or for committing terrorist acts.
- Identity fraud and forgery — Fake identity documents, forged official documents or currency, and content made to impersonate real people in order to deceive.
- Sexually explicit content — Depictions of sexual acts, or images centred on genitals in a sexual context, in every mode. Erotic or sensual art, artistic nudity and provocative images of fictional adults are allowed in Liberal mode, never of real, identifiable people and never where anyone could be a minor.
- Real people — Photorealistic or realistic depictions of identifiable living people without their consent, in every mode. Caricature, cartoon and clearly artistic depictions of public figures, including political satire, are allowed unless sexual.
Normal mode also blocks graphic violence and other content that is not brand-safe.
Human review
Section titled “Human review”To check our automated classifier’s decisions, we select some requests for human review: requests our classifier refused, requests it allowed with low confidence, some requests from new accounts, and requests from accounts our systems flag for unusual patterns. This review:
- Is conducted without your account details — the reviewer does not see your identity unless prohibited content is identified and escalation is required
- Includes the prompt, generated image, classifier decision, and any uploaded input images
- Sampled content is stored in a secure, access-controlled environment for up to 7 days, then permanently deleted
See Section 10.2 of the Terms of Service.