I Built and Published a Children's Picture Book in 2 Days Using AI — Now Live on Amazon

Two months ago I'd never used KDP. Three days ago I shipped my first paperback: What Does the Moon Dream About? — a bedtime story for ages 3-7. Update 2026-05-13: the Amazon listing is currently offline pending KDP follow-up (see banner above). Two more books built on the same pipeline are en route.

The whole pipeline ran on free tools + ~6 hours of focused work + 2 KDP rejection cycles. Here's what it actually looked like.

The pipeline

Three Python scripts, three open-source models, zero external services:

1. Pollinations FLUX (free, no auth) — generated 24 illustrations at 2550×2550 (300 DPI for 8.5" print) with seed=42 for character consistency across pages
2. reportlab + Google Fonts (Caveat + Patrick Hand) — assembled the 24-page interior PDF with text overlays in soft cream panels
3. reportlab again — assembled the print-ready 17.31"×8.75" cover spread (back + spine + front, with bleed)

That's it. No designer, no illustrator, no manuscript editor. The rhyming verses I wrote myself (24 quatrains in one sitting), the character — a sleepy crescent moon with pink cheek-blush and a small star companion — was held consistent via prompt-engineering and shared seed.

What KDP rejected (twice)

Both rejections were mechanical, not content. KDP's automated prepress checker caught real issues:

Both fixes took ~30 minutes once diagnosed. The interior PDF binary went from 313 MB to 313 MB (same — images dominate, the dimension change is metadata).

Automated preflight (saved me a third rejection)

After the first rejection I wrote a 50-line PyMuPDF script that extracts every text bounding box from the PDF and verifies each is inside KDP's 0.375" safety zone. Same logic runs on the cover (per-side: back / spine / front). 10/10 text spans inside safety = ship.

import fitz
TRIM = 8.5 * 72
BLEED = 0.125 * 72
SAFETY = 0.375 * 72
PAGE = TRIM + 2 * BLEED
safe = fitz.Rect(BLEED + SAFETY, BLEED + SAFETY,
                 PAGE - BLEED - SAFETY, PAGE - BLEED - SAFETY)

doc = fitz.open("book01_interior.pdf")
for i, page in enumerate(doc):
    for block in page.get_text("dict")["blocks"]:
        for line in block.get("lines", []):
            for span in line.get("spans", []):
                bbox = fitz.Rect(span["bbox"])
                if not safe.contains(bbox):
                    print(f"page {i+1} '{span['text'][:30]}' overflows")

If I'd run this before the first submit, I'd have saved a 24-hour KDP review cycle.

The economics

86 copies is the threshold for hourly-rate breakeven. At a competitive ranking for "moon bedtime story for kids," that's 2-4 months of organic sales without paid acquisition. After break-even, every additional copy is pure margin (KDP prints on demand — no inventory cost).

Why this isn't a one-off

The full pipeline is now a 90-minute template. Same Pollinations FLUX + reportlab + KDP workflow → next 4 books in the Dreaming World series are queued (Sun / Stars / Ocean / Forest themes). Each one carries the moon character forward as a recurring guide.

The unit economics scale. A series of 8 books at 20 sales/month each clears ~$550/month in royalties, compounding for 5+ years. No staff, no support burden, no servers. KDP handles fulfillment globally.

Tools used (full list, all free)

• Pollinations.ai — FLUX-based image generation, no auth, no rate limits I hit
• reportlab — Python PDF assembly
• Caveat + Patrick Hand — handwritten Google Fonts
• PyMuPDF — preflight checker for KDP safety zones
• Amazon KDP — distribution + print-on-demand

What I'd do differently

Run the preflight checker as the FIRST step after rendering. Both KDP rejections cost me 24-48 hours each of review-queue time when 30 seconds of Python would have caught them.