On this page
1. Our principles
CineCLI is designed around five commitments that govern how we build features and how customers may use them:
- Lawful & rights-respecting. The platform must not be used to break the law or to infringe the rights of others, including intellectual-property and publicity rights.
- Safety by default. Protective controls are on for everyone and cannot be silently disabled. Customers may tighten — but not remove — core safety filters.
- Transparency. AI-generated media is labelled and carries verifiable provenance so audiences and platforms can identify it.
- Human accountability. Automated systems assist people; they do not replace human judgement for consequential decisions.
- Privacy & security. Customer content is protected, kept in-region where required, and never used to train foundation models.
2. Safety across the content lifecycle
We apply controls at every stage of generation rather than at a single checkpoint:
| Stage | What we do |
|---|---|
| Input | Prompts, scripts and reference uploads are screened before any model is called. Disallowed requests are blocked and logged. |
| Generation | Each foundation-model call (text, image, voice, video) runs through Amazon Bedrock Guardrails with our policy configuration. |
| Output | Generated media is re-screened. Assets that fail policy are quarantined and not delivered. |
| Export | Approved output is stamped with C2PA content credentials and an invisible watermark before download or publishing. |
| Post-delivery | Audit logs, takedown handling and abuse reporting allow issues to be traced and remediated. |
3. Content moderation & Amazon Bedrock Guardrails
CineCLI uses Amazon Bedrock Guardrails together with our own classification layer to filter content across categories including hate, harassment, violence, self-harm, sexual content, and dangerous or illegal activity. Guardrails are applied to both prompts and model responses. Our classification layer adds short-drama-specific checks, such as detecting attempts to depict identifiable real people or minors.
Filters are configured to fail safe: when the system is uncertain, it blocks or routes the item to human review rather than releasing it. Categories of prohibited content are detailed in our Acceptable Use Policy.
4. Likeness, consent & deepfakes
Synthetic video carries a heightened risk of misuse. CineCLI applies strict rules:
- No real-person likeness without consent. Generating the face, voice or distinctive likeness of an identifiable real person requires a verifiable consent record on file. Public figures are not an exception.
- No deceptive deepfakes. We prohibit content created to deceive viewers about real events, statements or endorsements, or to impersonate a person or organization.
- No election or public-safety disinformation. Content that fabricates statements by officials or institutions, or that could mislead about voting, emergencies or public health, is disallowed.
- Voice cloning controls. Cloned or custom voices require consent and are watermarked; impersonation of real individuals is blocked.
- Disclosure. Where a real person's likeness is used with consent (for example, an actor or a brand spokesperson), the output is labelled as AI-assisted.
5. Protection of minors
CineCLI does not permit the generation of sexual or exploitative content involving minors of any kind, real or synthetic. Detected attempts are blocked, logged, and may be reported to the appropriate authorities as required by law. The service is not directed to children, and accounts must meet the age requirements in our Terms of Service.
6. Provenance, watermarking & disclosure
Every export from CineCLI is designed to be identifiable as AI-generated:
- C2PA content credentials embed tamper-evident metadata describing how the asset was produced.
- Invisible watermarking is applied to generated imagery, video and synthetic audio to support downstream detection.
- On-screen and metadata labels can be enabled to disclose AI assistance to audiences and to satisfy platform requirements.
We do not provide tooling to strip provenance or watermarks from CineCLI output, and doing so is a violation of our terms.
7. Human oversight
People remain accountable for what is produced. CineCLI provides review queues, role-based approvals and the ability to require human sign-off before publishing. Our Trust & Safety team reviews escalations, investigates reports, and can suspend access for policy violations. Customers are responsible for reviewing AI output before distributing it.
8. Data & model governance
- Customer prompts and content are not used to train foundation models.
- Foundation models are accessed through Amazon Bedrock, which keeps requests within the selected AWS Region and encrypts data in transit and at rest.
- We pin model versions, evaluate changes before rollout, and maintain documentation of which model performs which task. See our Models page.
- Access to customer content is restricted, logged, and limited to legitimate operational and safety purposes. See our Security and Privacy pages.
9. Reporting & enforcement
If you believe content was generated in violation of this policy, or you are the subject of misused likeness, contact us at safety@cinecli.com. We investigate reports, can remove content and suspend accounts, and cooperate with lawful requests. Repeated or severe violations result in termination. Copyright matters are handled under our Copyright / DMCA process.
10. Contact
Questions about responsible use can be sent to safety@cinecli.com. This policy works together with our Acceptable Use Policy, Terms of Service and Privacy Policy.