Type the "why?" your 3–6 year old just asked. About a minute later there's a calm, 60-second animated answer — with the real science for you underneath. No feed, no autoplay, no rabbit hole: one question, one video, then you two talk.
My 4-year-old asks about thirty "why" questions a day. I could answer maybe half. The other half ended with "let's look it up," which meant YouTube, which meant forty minutes later we were watching someone unbox plastic eggs.
So I built the thing I actually wanted: you type the question, and about a minute later there's one calm 60-second video made for a 3-6 year old, with the real science written out for the grown-up underneath.
The interesting part isn't the generation. Script, illustrations and narration are a few API calls each. That's the easy 10%. The other 90% is the checking, because AI content for kids fails in specific, documented ways:
- Safety facts. Generated kids' scripts will happily show whole grapes as a snack or a child crossing on a red light. Every script is screened with those failure modes as negative examples. - Honest metaphors. "A magic dust makes you sleepy" fails the check; "a clock inside your body" passes: one has to be unlearned later, the other grows with the child. - Pronunciation. The narration is transcribed back by a separate model and diffed word-for-word against the script; bad takes are re-recorded automatically.
These run in parallel with rendering, so they cost almost no wall-clock time.
What it deliberately doesn't do: no feed, no autoplay, no recommendations, no accounts for kids, and a daily cap on new videos. It ends after one video instead of pulling you into the next one.
It is AI-generated, and I'd rather say that plainly than bury it. Every file is labelled as such, and the checks are written out in full at wonderchirp.com/how-its-made. They are not perfect. When something slips past them, I get alerted and can pull the video.
Watching is free and there's no account. There's a Japanese version too, at wonderchirp.com/ja.
I'd love to hear from other parents, especially where the answers feel too simplified, or not simplified enough. Happy to go deep on the pipeline for anyone curious.
Hi PH 👋
My 4-year-old asks about thirty "why" questions a day. I could answer maybe half. The other half ended with "let's look it up," which meant YouTube, which meant forty minutes later we were watching someone unbox plastic eggs.
So I built the thing I actually wanted: you type the question, and about a minute later there's one calm 60-second video made for a 3-6 year old, with the real science written out for the grown-up underneath.
The interesting part isn't the generation. Script, illustrations and narration are a few API calls each. That's the easy 10%. The other 90% is the checking, because AI content for kids fails in specific, documented ways:
- Safety facts. Generated kids' scripts will happily show whole grapes as a snack or a child crossing on a red light. Every script is screened with those failure modes as negative examples.
- Honest metaphors. "A magic dust makes you sleepy" fails the check; "a clock inside your body" passes: one has to be unlearned later, the other grows with the child.
- Pronunciation. The narration is transcribed back by a separate model and diffed word-for-word against the script; bad takes are re-recorded automatically.
These run in parallel with rendering, so they cost almost no wall-clock time.
What it deliberately doesn't do: no feed, no autoplay, no recommendations, no accounts for kids, and a daily cap on new videos. It ends after one video instead of pulling you into the next one.
It is AI-generated, and I'd rather say that plainly than bury it. Every file is labelled as such, and the checks are written out in full at wonderchirp.com/how-its-made. They are not perfect. When something slips past them, I get alerted and can pull the video.
Watching is free and there's no account. There's a Japanese version too, at wonderchirp.com/ja.
I'd love to hear from other parents, especially where the answers feel too simplified, or not simplified enough. Happy to go deep on the pipeline for anyone curious.