AI disclosure

• Images. People, places, textures and atmosphere come from an image model, produced at native 2K.
• Motion, in a small number of shots. Two to four beats in an episode become true generated motion, each one starting from a still that was already approved.
• Voice. The narration is synthetic, from a voice you chose or a voice you cloned with consent.
• Text. Topics, scripts, titles and descriptions are written by a language model inside a locked contract, then checked.

• Anything that has to be accurate. Numbers, charts, timelines, maps, quotations and citations are drawn in code, deterministically. They are exact and identical every time, which no image model can promise.
• The camera pass and the audio pass. Grain, lens character, white balance, room tone and loudness are deterministic processing, not a model guessing.
• Sources. A source is found and checked, never invented. Where none exists, the claim is flagged and it holds up production.

The pipeline stops and waits for a person at 4 points: the plan and its cost, anything the proof gate cannot resolve on its own, the packaging choice, and publish. Publishing is never automatic. How it works sets those out in full.

When you publish, the AI-content disclosure is set on every platform that provides one. It is one of the 17 delivery checks, and it is a blocker — it stops delivery rather than shipping quietly, because a policy strike lands on your account, not on ours.

• Another channel’s identity. Signature phrases, voice fingerprint, palette, mascot, title format, thumbnail template and sting are blocked in the system and never reach the writer.
• A real person impersonated without consent, including their voice.
• A fabricated quotation, document or statistic. If it cannot be sourced, it does not get said.

• Avatar realism holds for about 12–15 seconds. Shots are capped at 15 seconds in code and long-form avatar video is not sold at any price.
• Arabic lip-sync is measurably weaker than English across every provider available today. It is scored on its own row and held to its own threshold rather than averaged with English.
• A product held in a hand is the hardest shot in this medium and is planned around rather than promised.
• Text inside a generated image is unreliable everywhere, which is why anything that must read correctly is drawn in code instead.

Music licensing is unresolved. We make no claim about music rights on any platform and we do not supply cleared tracks. If you add music, that decision and its rights are yours.