AI-Generated Kids Content: Our Safety Checklist Before Anything Ships

The 14-point safety checklist

We run this on every piece of kids content before it ships. Roughly half the items are catchable by automated tooling; the rest require a human review.

Visual safety

1. No uncanny faces — the most common AI failure mode. Reject any frame where a character's face triggers a "wrongness" reaction.
2. No frightening transitions — sudden scene changes, jump cuts to dark imagery, or anything that could startle a young viewer.
3. No ambiguous violence — slapstick is fine; AI-generated implied violence is not (it often gets weird in subtle ways).
4. Color palette stays in safe range — overly saturated reds, deep blacks dominating frames, or strobing patterns are removed.
5. No real-looking children depicted — all kid characters must be clearly stylized.

Audio safety

1. No sudden loud sounds — even within child-friendly content. Compress dynamic range so a child watching at bedtime isn't startled.
2. No music in dissonant keys without clear comedic intent — minor-key drifts can read as scary.
3. Voiceover is age-appropriate vocabulary — automated grade-level checking helps but humans confirm.
4. No incidental adult-context phrases — AI scripts sometimes hallucinate idioms that aren't kid-friendly. Human review catches these.

Story safety

1. No moral ambiguity in resolution — every kids piece resolves with a clear, positive emotional outcome.
2. No unresolved peril — characters can be in trouble, but the trouble must be resolved within the same piece.
3. No "and they all lived... or did they?" endings — beloved by indie creators, hated by parents.

Compliance

1. COPPA-compliant data practices if the content lives in an app or site we control — no behavioral tracking on under-13 audiences.
2. YouTube made-for-kids designation is set correctly on upload — this is a one-checkbox decision with major policy consequences.

Why the human-in-the-loop matters here more than anywhere else

AI tools are getting better at catching unsafe content, but they're not built to catch mood failures. A perfectly compliant-but-unsettling video can pass every automated check and still be wrong for kids. A human watching the final cut catches what automated review misses.

Our standard: every kids piece is watched start-to-finish by a human reviewer who is asked one question — "would you be comfortable putting this on for a 4-year-old you care about?" If the answer is anything other than "yes, completely," the piece doesn't ship.

The standard we hold ourselves to

There is no version of "fast" that's worth shipping a piece of kids content we'd be uncomfortable with. The tools let us produce volume; the policy requires that none of that volume be sloppy.