1) Why this exists

AI can produce lifelike audio, images, and video that people may mistake for reality. To keep experiences transparent and safe, Twinen requires clear disclosure whenever content is AI-generated or materially edited in a way that could mislead a reasonable person. This mirrors recent moves by large platforms to require labels on "realistic" altered/synthetic media.

Scope

Applies to all uploads and posts (video, image, audio/voice, text, live) on Twinen, including material made with AI Twin or third-party tools. It also applies to ads and sponsored content.

2) What requires disclosure (the rule of "realistic & material")

Disclose if any are true:

1. Looks or sounds real: A typical viewer could reasonably think the scene, person, or audio is real. (Industry norm is to require disclosure specifically for realistic synthetic/altered media.)
2. Material edits change meaning: You've added/removed people or objects, changed speech, altered actions/timing, or composited scenes.
3. Likeness used: You used a real person's face/voice (clone, swap, TTS) without them actually doing/saying it.
4. Sensitive context: elections/civics, public safety/health, finance—where deception could cause harm; adds stricter disclosure.

No disclosure needed for minor, non-misleading edits (color grading, noise removal, basic filters, obvious AR stickers/animation). TikTok/YouTube similarly focus on realistic/meaningful alterations, excluding trivial edits.

3) How to disclose (badges, captions, provenance)

A) Visible labels (mandatory in UI)

Twinen will show a top-level badge plus a context tag on the asset and in details:

• Primary badge (choose one):
  • AI-generated – made wholly or substantially by AI
  • AI-edited – materially altered by AI
• Context tag (add as needed): Parody/Satire, Dramatization, Composite, Face swap, Voice clone, Restoration/Upscale, Staged

B) Caption text (short-form disclosure)

Add a concise, plain-language line in the caption when context matters, e.g.:

• "This clip is a parody with an AI voice clone."
• "Fictional dramatization; city scene AI-generated."

C) Alt text & accessibility

If the label isn't announced in audio/descriptions, include it in alt text or audio description (e.g., "AI-generated image of…").

D) Provenance metadata (Content Credentials / C2PA)

Where feasible, Twinen writes or preserves Content Credentials (C2PA) so viewers and tools can verify how media was made. Do not remove this information. (C2PA/Content Credentials is the leading provenance standard.)

Tampering with or stripping provenance/watermarks at scale may lead to enforcement.

4) Likeness & consent (faces, voices, names)

• Private individuals: realistic face/voice clones require verifiable consent. Keep a release or in-app consent record; we may request it on review.
• Public figures: permitted for satire/commentary with clear labeling; no sexual content, no fabricated crimes or medical claims as fact, and no false endorsements without authorization.
• Minors: never in sexual or exploitative contexts (real or synthetic). (All major platforms treat sexualized minors as zero-tolerance.)

5) Ads, sponsors, creators

If content is paid or sponsored, you must disclose both the ad status and the synthetic status:

"Ad – AI-generated product demo"

(Platforms require ad/synthetic labels in political contexts; Meta and others have added election-related disclosures for AI. Twinen mirrors this for clarity.)

6) Decision guide (quick flow)

1. Could a typical viewer think it's real? → Yes → Label AI-generated/AI-edited.
2. Is a real face/voice cloned or swapped? → Yes → Add Face swap/Voice clone and confirm consent.
3. Is it parody/satire or dramatization? → Yes → Add the corresponding context tag.
4. Is it civic/health/safety/finance? → Yes → Strongly prefer on-screen bumpers + extra context.

7) Examples

Allowed with disclosure:

• A public-figure sketch with an AI voice clone, labeled AI-generated + Parody; caption clarifies it's fictional.
• A disaster dramatization used for education; labeled AI-generated + Dramatization and noted in the caption.
• A restored old photo with AI colorization; labeled AI-edited + Restoration.

Removed or restricted:

• A fake evacuation alert posted as real (public safety harm).
• NCII deepfake of any person.
• Sexualized depiction of minors, including stylized "youth-looking" AI. (Consistent with industry policies.)

8) Enforcement if disclosure is missing

• First time / low risk: Twinen may add a label and notify you.
• Repeat / medium risk: Visibility limits, interstitial warnings, or age-gates; potential feature limits.
• High risk / harm: Removal; strikes/suspension; bans for zero-tolerance categories (e.g., NCII, minors, fraud).

This ladder is consistent with major platforms that label, reduce reach, or remove harmful manipulated media.

9) Developer & API requirements

• Include synthetic flags and context tags in your upload payload:
  synthetic=true|false; method=generated|edited; tags=[…]; provenance=C2PA manifest
• Bots/agents must disclose automation in their profile (name/bio) and pass our safety checks (inauthentic coordination is prohibited on other platforms too).

10) Reporting & appeals

Only the emails this standard needs:

• Report undisclosed or deceptive synthetic media: abuse@twinen.com
• Urgent takedown for non-consensual intimate imagery (NCII): ncii@twinen.com
• Appeal a label or removal: moderation@twinen.com

(Other addresses—privacy, DMCA, legal—are in the Terms; we keep this doc to the minimal inboxes it needs.)

11) FAQs