I bought some picture books for my three-year-old on a Korean secondhand marketplace app. The seller added 19 English board books for free, as a bonus. I said thank you, took them home, and put them on a shelf. I had no idea how to use them, and she was too young for English anyway.

That was a few months ago. This post is about how those books went from a shelf to a tab in her media player, and about the one thing that made it possible: I let the AI look at the books before I decided anything.

The Need: A Free Box of Books Is Not a Plan

If you are a parent, you may have a version of this at home. A set of books, a gift, a hand-me-down. You know it’s “English something.” You don’t know:

  • Is it a story set, a song set, a phonics set?
  • Is there supposed to be audio? Where do you get it?
  • Is it right for a three-year-old, or should it wait?

Board books with no obvious CD or QR code are the worst case. The book alone tells you very little, and searching the web by guessing the title from a spine photo is slow.

The Proof: It Is a Tab Now

The set is now a tab in the same app she already uses for songs and stories. Each of the 19 books is a card with its real cover. Tap one and you get the song, the video (behind a password, more on that in the next post), the English lyrics with a Korean translation, and a photo of the book’s own translation page.

Nothing about it is fancy. It is the same kind of folder-based tab I already had for her English picture-book CDs. The new part was working out what the books were and what to put in each folder.

The Story: Why Now

Two things changed.

First, she started enjoying the English tab. That tab holds picture-book CDs, and I noticed she asked for those episodes on her own.

Second, she had been memorizing Korean songs from the other tabs, almost word for word. I think she now has a basic feel for Korean. She also likes English sounds. So I decided it was time to start exposing her to English properly, and not just in passing.

My wife and I talked it over and agreed on one rule from the start: no video for now. We want her to listen first. I’ll describe how I handled that in the next post.

The How: Photos First, Then Questions

Here is the exact order I used. Nothing here needs coding skill. It is mostly photographing and asking.

1. Show the AI the whole set and one full book

I took two kinds of photos:

  • One photo of all 19 books stacked, with the spines facing the camera, so every title is readable.
  • The front and back cover of one book, laid flat.

Then I asked what the set is. This is the prompt from my coding history, translated from Korean:

💬 Prompt that worked “I want to add a new tab to the kids player and include about 20 English books. These two images are the front and back covers of one of the books. Please research whether this book has audio and how it is organized, and tell me how to include it so my 3-year-old can enjoy it.”

A little later I sent the photo of the whole stack and added:

“This is a photo of all the books. Please check whether the audio exists on YouTube. I’ll download it myself.”

What came back was the useful part. The set turned out to be a collection of classic nursery-rhyme and children’s song books, and each book is built around one song. The answer was not “read the book aloud.” It was “songs are the core of this set.” That one fact decided the whole design: audio first, book second.

I wouldn’t have found that by reading the covers myself. They are mostly pictures with a short English title.

I then asked it to find the matching video for each of the 19 songs. Claude doesn’t download videos, so the plan was: it produces a plain text list of the 19 addresses, one per line, and my own downloader program handles the rest.

“Please make a text file with only the 19 addresses.”

My downloader accepts a bulk paste, so one paste handled all 19. These are files for my own use at home, not for sharing. If you do the same, check the rules for the content you use.

3. Hand over the files and the photos together

I pointed the AI to the folder with the 19 downloaded videos and to a second folder with photos of each book’s English and Korean pages. The ask was:

“Check that the downloaded files match the 19 page images. Extract the audio from the videos. Give this tab the same translation-view feature as the English tab. I’ll scan the covers later, so use those images for it.”

It matched files to books by title and order, and extracted an MP3 from each video.

One real trap: my photos and scans were in the reverse order of the book stack, so the AI needed a mapping rule (number k in the stack becomes 20 minus k in the photos) before covers and songs lined up. If your photos and your list are in different orders, say so up front.

4. Crop the covers yourself

I scanned the 19 front covers with my phone and cropped the white margins in a photo editor. I could have asked the AI to do it. I didn’t, because last time I did that the image-editing back and forth used a lot of tokens, and doing it by hand is quick for me. I think this is worth knowing: some jobs are cheaper for you than for the AI. I told it in one line that the margins were already fixed and it should not touch the images.

What It Cost

  • The 19 books: $0 (they were a bonus)
  • Hosting: $0, same Synology DS224+ and Cloudflare Tunnel as before
  • Audio and video files: about 310 MB total on the NAS
  • AI usage: my existing Claude subscription, no extra charge for this work. Lyrics and translation use pay-per-use APIs, which I cover in the next post.

What I Don’t Know Yet

It has been about two days. She likes the songs and she likes picking a cover. I don’t yet know how much English she actually picks up this way, and I’d be guessing if I said. I’ll report back when I have something real to say.

In the next post I cover the part that took the most care: cutting the ads out of the videos, locking the video mode, and making the lyrics show up in time with the singing without scrolling the screen.