AI Generated Content Policy

This document describes how Tengen AI Corp generates, moderates, attributes, and discloses AI-generated content on the Yurivan platform. It supplements the Terms of Service, Privacy Policy, and Content Moderation Policy.

1. What "AI-Generated Content" Means

AI-generated content on Yurivan refers to text stories, illustrated images, and multi-panel illustrated manga produced by Tengen's generation pipeline (Yurivan Credits) at the request of a verified adult user.

It does not include human-uploaded illustrations, even if those illustrations were originally created by the uploader using AI tools elsewhere. Such uploads are governed by the standard user-upload moderation in the Content Moderation Policy.

2. Models In Use

The model used for any given generation is recorded in the audit log and visible in your generation history.

3. AI Disclosure

• Public catalog labeling. Any AI-generated content the user opts to publish is labeled with an "AI Generated" badge visible on the catalog card and detail page. The label cannot be removed.
• Underlying model attribution. Story metadata exposes the model family (e.g., "Magnum v4 72B"). Specific generation parameters (seeds, samplers, step counts) are retained internally but not user-visible.
• No human-authored misrepresentation. Users may not represent AI-generated content as human-authored where applicable law requires AI disclosure. This obligation rests with the user; Tengen's labeling supports user compliance, it does not replace it.

4. Safety Pipeline (Mandatory, Pre-Storage)

Every AI-generated image is processed in this exact order before being written to durable storage. Any failure short-circuits the pipeline and the image is discarded.

1. Prompt-level filter. Lexical filter for minor-coded keywords, age-numeric <18 in any field, and real-person-name match against the Hive Celebrity name database. On match: REJECT before any model is invoked.
2. Brief schema validation (text generation). Every character has explicit age ≥18, 30+ word physical description, populated outfit/setting/ composition. On failure: re-prompt brief extractor once, then REJECT.
3. Generation. Text via OpenRouter. Image via RunPod ComfyUI workflow with hardened negative prompt baked in.
4. Anatomy QC (image only). Gemini 3 Flash vision validator. Up to 3 regeneration attempts on fresh seeds.
5. CSAM detection (image only). Hive AI CSAM Detection API. Threshold: any non-zero confidence score = REJECT, image not written to storage. Rejection logged.
6. Apparent age classifier (image only). Hive AI Apparent Age API. Threshold: probability of subject being <25 must be ≤0.05 (deliberate 7-year safety margin above the 18+ legal floor). On failure: REJECT.
7. Real-person classifier (image only). Hive AI Celebrity Recognition. Any positive match: REJECT.
8. Storage. Pass = upload to R2 (private bucket if user-private, public bucket if user opts in to publish).

5. Account-Level Safeguards

• Three CSAM rejections by the same account in any 24-hour rolling window automatically suspends the account pending Trust & Safety review.
• Three apparent-age rejections in 24 hours suspends the account similarly.
• Three prompt-filter rejections in 24 hours triggers a warning email; ten in 7 days suspends the account.

6. What Yurivan Will Not Do

These constraints are architectural — enforced in pipeline code, not just policy.

• No user-uploaded reference images into the AI pipeline. No IPAdapter from user photos. No LoRA training on user-uploaded images. No img2img seeded from user uploads.
• No real-person likeness generation, including by description, even if the prompt does not name the person.
• No content depicting minors, regardless of stylistic framing or fictional setting.
• No model fine-tuning on user-uploaded content without explicit written consent (separate from limited model-improvement use of generation prompts/ outputs governed by the Privacy Policy).
• No deepfake or face-swap features.

7. Intellectual Property in AI Output

• Subject to the Terms of Service, you may use content you generate via Yurivan Credits for personal and commercial purposes.
• Copyrightability disclaimer. US Copyright Office guidance (most recently in Thaler v. Perlmutter and 2023-2025 USCO registration guidance) holds that pure AI-generated output without sufficient human authorship is not copyrightable. Tengen makes no representation that AI-generated output qualifies as a copyrightable work.
• Tengen's image and text models were trained by their respective creators (Anthracite, Nous Research, etc.) on datasets we do not control. Users alleging a specific output infringes copyright should follow the DMCA Policy.

8. Use of Your Generations to Improve the Service

By default, anonymized prompts and outputs may be used by Tengen for product improvement, including potential model fine-tuning. You may opt out at Account Settings → Privacy without losing access to any feature. User-identifying information is stripped before any internal use; we do not associate individual generations with named users in training data.

9. Reporting Concerns

• CSAM or apparent CSAM: report immediately to trust@yurivan.com. Tengen will investigate and report to NCMEC under 18 USC § 2258A within 24 hours of confirming the report.
• Real-person likeness without consent: contact trust@yurivan.com or legal@yurivan.com.
• Copyright infringement by AI output: follow the DMCA Policy.
• Other violations: trust@yurivan.com.

10. Changes to This Policy

We will post material changes to this Policy on this page and notify registered users by email at least 30 days before they take effect, except for changes required by law or to address an immediate safety risk, which take effect upon posting.

Contact

• Trust & Safety: trust@yurivan.com
• DMCA: dmca@yurivan.com
• Legal: legal@yurivan.com
• Privacy: privacy@yurivan.com
• Mail: Tengen AI Corp, 2224 Augusta Place, Santa Clara, CA 95051 USA