How to Turn a Picture Book Into a Read-Aloud Video With AI
Disclosure: I am an independent affiliate of companies linked in this post and may be paid for referrals through links marked “paid link”. That never decides what I recommend; see how this site makes money.
The short answer: split the book into one line per page, generate the whole narration in one take with a pause after each line, cut the audio at those pauses, and hold each page on screen for as long as its line runs. Then add captions and watch the whole thing.
That is how I made a read-aloud of my picture book, Floria. The book has a cover and 14 spreads, so the video has 15 lines. The finished video runs 4 minutes 12 seconds.
Who this suits
This fits you if you already have a finished picture book, with pages you own and the right to publish them. It also fits short illustrated stories and any writing that breaks cleanly into one idea per picture.
Skip it if the voice is the point. If families will come to you for how you read, record it yourself. A generated voice also makes less sense if your book depends on songs or several distinct character voices. I used one voice for the whole story.
The workflow
1. Write one line per page
Take the text from each page and keep it exactly as printed, so a child following along sees the same words. Add a short spoken line for the cover. Mine gives the title and one sentence about the story.
If you have a page with no text, decide now what happens there. A silent page held for a few seconds is fine, but it needs a planned length.
2. Generate the voice in one take
I used ElevenLabs (paid link). I am an independent affiliate of ElevenLabs and may receive compensation for referrals.
I pasted the whole story in as one piece, with a two-second pause between lines. One take keeps the voice and pace steady from page to page. If you generate each page by itself, the delivery can shift between pages.
The take came out at about 4 minutes 25 seconds. The audio was generated on the paid Starter plan, because I wanted to publish it. More on that below.
3. Cut the audio at the pauses
The pauses are what make this quick. A tool can find each silence and cut there, so you never have to mark 15 lines by hand.
FFmpeg, a free command-line tool, has a filter for this:
ffmpeg -i full.mp3 -af silencedetect=noise=-35dB:d=1.8 -f null -
That lists every silence quieter than -35 dB and at least 1.8 seconds long. Each one marks the end of a line.
4. Hold each page for exactly as long as its line
Each page stays on screen for the length of its narration, plus one second before the next page. This is why the video is shorter than the raw take: the two-second pauses became one-second gaps.
I also added a very slow zoom, 5% across each page. Alternate pages zoom in and out. It keeps a still page from feeling frozen.
The last page holds for five seconds, so the story does not end abruptly.
5. Add captions
Captions let a child, or an adult reading along, follow the words. I split the 15 lines into 86 short captions, timed against each line's audio, and saved them as a subtitle file to upload with the video.
6. Watch it from start to finish
Check three things: each page matches its line, every name sounds right, and no line is cut off at either end. I respelled one name so the voice would say it correctly.
What happened when I made Floria
The mistake worth knowing: the first cut gave me 17 lines instead of 15.
The voice had paused too long in two places. To the tool, those looked just like the pauses I had asked for. The fix was to cut only at the longest pauses, which is why the command above uses 1.8 seconds. My script now also stops if the count of lines does not match the count of pages. Without that check, every page after the first bad cut would have been out of step with its audio.
I had Claude Code assemble the video with FFmpeg. Any video editor can do the same job by hand, but expect to place 15 pages and 86 captions yourself.
Limitations
One bad line can mean more work. With a single take, you cannot simply swap one word. You regenerate the take, or generate that line separately and accept that it may sound slightly different.
The cuts depend on the pauses. As my 17 lines showed, the voice does not always pause where you expect. Always count the lines.
Pictures need rights too. My illustrations came from OpenAI's image tools. If you use another illustrator's art, check that your agreement covers video.
A license only covers what its plan covers. ElevenLabs says free-plan audio has no commercial license and needs attribution, while paid plans include a commercial license, excluding Beta services. Its help center also says audio made during a subscription keeps that license after you cancel. Check the current terms yourself before you publish. They were checked on November 1, 2026.
YouTube has its own rules. It requires you to set each video's audience, and a read-aloud for young children is likely to count as made for kids. Features such as comments and personalized ads are limited on those videos, so do not plan on them. Check the AI use setting when you upload, and tell viewers the voice is AI in the description. I also keep affiliate links out of the children's video and its description.
The alternative that costs nothing
Read it yourself. A phone in a quiet room and a free editor will get you a warm, usable read-aloud, and the steps above still work: pause after each page, then cut at the pauses. I earn nothing from this option, and for many picture books it may be the better one.
Do not use the voices built into a Mac as a free shortcut. Apple's license does not allow publishing them this way, so check it before you try.