All articles

How I had my agent make my kids an illustrated audiobook

Illustrated Audiobook Studio is a Varosity skill that turns a book manuscript into a finished illustrated audiobook: a watercolour illustration for every chapter, one narrator reading the whole book, a music bed underneath, and a stitched video where the art changes as the story moves. You give it a story and a style; it hands back the finished film. My kids asked for a bedtime story the other night, so I pointed my agent at it and got back a real one — five chapters, illustrated, narrated, scored.

I'll tell you the actual story it made, because it's the clearest way to show what comes out the other end.

What I actually asked for

I wrote a short manuscript — "Jamison and the Hollow Oak." An eight-year-old boy finds a little green door in the old oak at the bottom of the garden, meets a palm-sized blue dragon named Ember, and has to wind the Great Clock to bring the lost seasons home. Five chapters. That's it. I didn't draw anything, record anything, or open an editor.

What came back was a five-minute video: a watercolour picture for each chapter that fades in as that chapter is read, one warm narrator the whole way through, and a soft lullaby running underneath the whole time. My kids watched it twice and asked for another one about a different dragon.

The part most people get wrong

The hard part of an illustrated book isn't generating a pretty picture. It's generating five pictures where the kid and the dragon look like the same kid and the same dragon every time. Most image tools drift — chapter one's boy has brown hair and a yellow sweater, chapter four's boy is a stranger. For a children's book that breaks the whole thing.

The way the skill handles it is the move worth stealing. Before it draws a single chapter, it generates a character reference sheet — Jamison and Ember, full body, clean background — on flux-1.1-pro. Then every chapter illustration is generated on nano-banana with that same reference image passed in. The reference is the anchor. Because it's identical on all five calls, the boy stays the boy and the dragon stays the dragon across the whole book. Same trick for the voice: one narrator voice on every chapter, never mixed.

How the agent actually runs it

Here's the real flow, start to finish:

1. Split the manuscript into chapters. 2. Generate a character reference sheet for the recurring characters (flux-1.1-pro), in the style you want — for ours, soft watercolour, warm palette. 3. For each chapter: derive a visual prompt from what actually happens in that chapter (not just the title), then generate the illustration on nano-banana with the reference image passed in so identity holds. 4. For each chapter: narrate the text in one consistent voice. We used a warm storyteller voice through openai-tts-1 at 0.9x speed — slowed down a touch because it's a bedtime book and pacing matters. 5. Generate a gentle instrumental bed (ElevenLabs music) and mix it under the narration low — quiet enough to disappear, present enough to feel like bedtime. 6. Stitch it: concatenate the chapter narration into one audiobook, then build a video where each chapter's art holds on screen for exactly as long as that chapter is read. The result is a synced illustrated film.

You own that loop. Varosity is the part that routes each step to the right model and hands back durable URLs — the image models, the narration, the music, all behind one key, billed in one place. You don't sign up for four services.

What to watch out for

A few things I learned the practical way:

  • Skip the reference image and your characters will drift. Generate the sheet first, then pass that same URL to every chapter. This is the difference between a book and five unrelated pictures.
  • Use one voice for the whole book. Mixing narrator voices mid-story is jarring in a way that's hard to name but easy to feel.
  • Derive each chapter's art prompt from the chapter's events, not its title. "Chapter 4: The Great Clock" tells the model nothing; "frost on giant frozen gears, a golden key too big for a dragon's claws" gives it the scene.
  • There's no seed, so the exact images aren't reproducible. Persist every URL the moment you get it.
  • Long chapters need splitting at sentence boundaries before narration, then re-joining — the skill handles this, but know it's happening.
  • It costs real credits — a handful of images and a few minutes of narration per book — but it's cents-to-low-dollars per finished audiobook, not a studio budget.

Running it yourself

It's in the Varosity skills library. Any MCP-connected agent can pull it and run it against your own manuscript — your story, your art style, the narrator voice you want. Hand it a manuscript and a style, and let the loop run.

I built this because I wanted to make my own kids something, and it turned out to be the kind of thing a small publisher or a parent or a teacher could use to turn any story into a finished, narrated, illustrated book in one pass. The story above is real. My kids are still asking for the sequel.