AI-narrated version of this post using a synthetic voice. Great for accessibility or listening while busy.
I just shipped a 32-page color picture book to Amazon KDP. Total wall-clock build time: four evenings. Total cash outlay: under . The whole stack uses tools you probably already have access to.
This is a behind-the-scenes case study of the workflow – what worked, where the friction was, and the exact tools at each step. If you’ve been thinking about publishing on KDP with AI assistance but haven’t found a workflow that actually ships, this is one that does.
The book
Snouts & Snores Storytime #1: The Big Smell – a read-aloud picture book for ages 4-8 about two lazy dogs (Biscuit the pug and Hank the bulldog) who follow their noses three blocks to a bakery and end up locked inside it. 14 full-color illustrated spreads, ~340 words of story, .99 paperback / .99 Kindle, free on Kindle Unlimited.
It’s live on Amazon now. The series hub is at auburnai.ca/snouts-and-snores.
The four-tool stack
Nothing exotic. The four tools that produced the book:
- Claude Sonnet (Anthropic) – manuscript drafting + revision
- Microsoft Copilot Designer (DALL-E 3, free tier) – all 14 spread illustrations
- PowerShell 5.1 + Microsoft Edge headless – HTML → print-ready PDF rendering
- Amazon KDP – distribution, paperback + Kindle
The total marginal cost was the API tokens for Claude (under ), the time of the human writing the prompts (me), and the Amazon Ads campaign I’m running post-launch (/day × 30 days).
Day 1: Story drafting with Claude
Picture books for ages 4-8 have a very specific tempo – short declarative sentences, lots of repetition, big “sound” words (POP, SNIFF, CLICK), and a story that rewards the read-aloud parent who has to do it three nights in a row.
I gave Claude a tight prompt: 14 spreads, ~25 words per spread, central characters Biscuit + Hank, plot = bakery raid + accidentally locked in + rescue. Output a complete manuscript with explicit illustration notes per spread.
What I got back was 90% shippable on the first pass. The revisions I made were small – tightening the dog door scene, adjusting Hank’s signature “Mmf” placement, swapping a too-clever phrase for a simpler one a 4-year-old would actually hear.
Time on this step: ~90 minutes including revisions.
Day 2: Illustrations with Copilot Designer
I tried Stability AI first. It’s powerful but bad at character consistency – I’d ask for the same two dogs across 14 spreads and get a different pug each time. By spread 7 my “Hank” had gone from a bulldog to a pug to something with six legs to something with a glazed donut around its neck.
I switched to Microsoft Copilot Designer, which is free, uses DALL-E 3, and – this is the key thing – holds character consistency dramatically better. I anchored the prompt with a fixed character description: “Biscuit is a fawn pug with a black face. Hank is a tan English bulldog with a white chest and a wrinkly face. Both are drawn in a cheerful illustrated children’s book style, soft warm palette.”
Then for each spread I tagged on the scene description from the manuscript. Same two dogs every page.
I generated about 4 candidate variants per spread and picked the strongest. Total Copilot time: one long evening.
Day 3: HTML → PDF with PowerShell + Edge
For an 8.5″×8.5″ full-color paperback picture book, you need a print-ready PDF with the right page size and bleed margins. KDP’s templates are Word/InDesign-friendly; for an automated workflow you want to render directly.
My approach:
- PowerShell script that takes the manuscript + illustration paths and emits a single HTML file with one
<div class="spread">per page, sized with@pageCSS to the trim dimensions - Edge headless invoked with
--print-to-pdfagainst that HTML file - Result: a clean, page-perfect interior PDF, generated in seconds, reproducible from source
The cover wrap PDF is the same thing at a different page size (17.3251″×8.75″ for a 32-page color 8.5″×8.5″ book, math is from KDP’s wrap calculator).
This works for ANY KDP book format – chapter books, trivia books, journals, picture books – which is what makes it valuable as a workflow. You write the script once and reuse it.
Day 4: KDP previewer hell + the Kindle EPUB
This is the day nobody warns you about.
KDP’s previewer runs your PDF through its own validator and tells you if anything is “outside the margins.” For a picture book where the WHOLE POINT is full-bleed illustrations, this is a problem. My first interior PDF got rejected for every image page because Edge’s headless PDF renderer doesn’t write TrimBox or BleedBox metadata, so KDP can’t distinguish “intentional bleed” from “accident.”
The fix was to rebuild the interior as no-bleed with all illustrations inset 0.25″ from the trim edge (effectively, your illustrations sit inside a white border on each page) – then KDP is happy. The cover wrap is still rendered with bleed because KDP requires that for covers regardless.
The Kindle eBook is even more interesting. KDP accepts a PDF for Kindle eBook, but the format you actually want is a fixed-layout EPUB3 – it preserves spreads, looks crisp on Fire tablets, and Amazon classifies it as a Kindle Kids’ Book (not Print Replica).
I built a EPUB packer in raw PowerShell. The thing that almost killed it: .NET’s ZipArchive silently ignores CompressionLevel.NoCompression on the first entry, and EPUB spec REQUIRES the mimetype file to be the first entry, STORED uncompressed. So I had to write the ZIP container by hand – local file headers, central directory, end-of-central-directory record, CRC32 lookup table – to get a spec-compliant EPUB.
It works. The Kindle interior file is now a 22 MB fixed-layout EPUB that passes KDP’s Kindle Previewer cleanly.
Cost breakdown
- Manuscript (Claude API): ~.50
- Illustrations (Copilot Designer): (free tier)
- Rendering (PowerShell + Edge headless): (built-in to Windows)
- KDP upload: (no listing fee)
- Post-launch Amazon Ads: /day × 30 days = (optional)
The book pays out at .24/copy paperback (60% royalty on .99 minus print cost) and .75/copy Kindle (70% royalty on .99). At that math, ~70 paperback or ~85 Kindle sales pay for the ad campaign. Beyond that, every sale is pure margin.
What I would tell anyone trying this
- Character consistency is the bottleneck. Pick an image tool that holds character anchors. Copilot Designer beat Stability for this task because the DALL-E 3 model holds named character descriptions across prompts.
- Don’t full-bleed your interior on KDP. Use no-bleed with 0.25″ inset. The cover gets bleed, the interior doesn’t. This will save you the previewer-rejection day.
- For Kindle picture books, build a real fixed-layout EPUB, not a print PDF copy. The EPUB looks dramatically better on Kindle hardware and Amazon treats it as a Kids’ Book.
- Write the PowerShell pipeline once, then every future book uses the same render → preview → submit loop. The first book takes four days. The second takes a day. The fifth takes an afternoon.
The book itself
Snouts & Snores Storytime #1: The Big Smell on Amazon – .99 paperback / .99 Kindle / Free on Kindle Unlimited.
If you have a 4-7 year old with two parents at bedtime, two lazy dogs, a sense of humor, and a soft spot for donuts – this book was made for them. There’s also an illustrated chapter book version of the same story for ages 6-10 if you have an older sibling at home. Both books are linked from the series hub at auburnai.ca/snouts-and-snores.
I’m releasing one of each (Storytime #N + Book #N) per pair through the rest of the series. If you build a similar pipeline using this workflow, send me what you ship – I’d love to see it.
Related Auburn AI Products
Building content or automations around AI? Auburn AI has production-tested kits:
- 100 Claude Prompts for Canadian SMB Owners ($17)
- The n8n + Claude Blog Automation Stack ($47)
- Auburn AI Monitoring Stack ($37)
- Browse the full catalogue
- Patterns App Review 2026: Vertical AI Agent for M&A Deal Teams (and What It Says About Where AI Is Going) →
- Auburn AI Custom Agents 2026: When Off-the-Shelf Doesn’t Fit →
- Gemini CLI Agent Review 2026: Google’s Answer to Claude Code →
- OpenAI Swarm Review 2026: Lightweight Multi-Agent for OpenAI Shops →
- n8n AI Agent Node Review 2026: The Agent Layer in a Workflow Tool You Probably Already Run →
