html-to-press-pdf

Use when converting a Claude-designed (or any HTML/CSS) magazine, zine, booklet, or multi-page document into a crystal-clean, print-ready PDF for a print shop like Mixam — especially when Chrome's pri

SKILL.md

GitHub

HTML → Press-Ready PDF

Overview

Claude-designed print HTML lays each page out as a fixed-px "artboard" (e.g. 670×870) scaled up to the trim+bleed page with transform: scale(...) on a wrapper div (.pp, .sheet, etc.). Chrome's --print-to-pdf and even puppeteer page.pdf() mis-paint scaled content during pagination — the print compositor fits each page using the pre-transform layout box, so the artwork squashes into the top-left corner and the rest of the page is dead whitespace. This looks like a "margin" bug but is actually a scale bug.

Core principle: Don't print transformed HTML to PDF. Screenshot each rendered page element (post-transform pixels are correct) at high DPI, then assemble lossless into a PDF with exact page boxes. Photos in these designs are raster anyway, so 384 DPI raster is print-indistinguishable from vector and sidesteps the bug entirely.

When to use

  • Output has white margins / content shrunk to a corner after Chrome print-to-PDF
  • HTML uses transform: scale() on page wrappers (grep the source for transform:scale)
  • A print shop (Mixam, Mixbook, Lulu, local) needs full-bleed PDF at exact trim
  • Source is a hosted Claude claudeusercontent.com design URL with relative ../images/ assets

Don't use for vector-only / text documents with no transform tricks — plain page.pdf({preferCSSPageSize:true}) is fine there and keeps text vector.

The pipeline

# 0. Find Chrome + read the design's @page size and scale mechanism
grep -n '@page\|transform:scale\|class="sheet"\|<img' design.html

# 1. Screenshot every page element at high DPI (loads live URL so images/fonts resolve)
cd <skill>/scripts && npm i puppeteer-core@23   # first run only
node shoot.js "<LIVE_HTML_URL>" /tmp/sheets 384 ".sheet"
#                                          ^DPI  ^page-element selector
#   384 default, 480 for extra crispness. Match source photo res; 300 is the floor.

# 2. Assemble lossless PDF with exact trim+bleed page boxes
python3 scripts/assemble_pdf.py /tmp/sheets /tmp/out.pdf 384
#   Pass the SAME DPI as step 1, or page boxes mis-size.

shoot.js waits for document.fonts.ready + all images decoded, hides .toolbar/.hint, and reports broken images. assemble_pdf.py uses img2pdf (no recompression) with get_fixed_dpi_layout_fun so page box = pixels / dpi inches exactly.

Verify before delivering (always)

python3 scripts/verify_pdf.py /tmp/out.pdf

Confirms: page count, every page identical exact in/pt size, and content bleeds 0→~100% (no whitespace). If a page covers only ~66%, the transform bug is back — you printed instead of screenshotting.

Print-shop settings (Mixam defaults)

  • Page size in HTML = trim + 0.125" bleed all sides. E.g. trim 8.375×10.875 → @page { size: 8.625in 11.125in; margin:0 }. Order at the trim size; bleed is already included.
  • No crop/registration marks — Mixam wants bleed only.
  • Page count: saddle-stitch ÷4, perfect-bound ÷2. Perfect-bound often wants the cover as a separate file.
  • Color: deliver sRGB; Mixam auto-converts to CMYK (safest for dark/moody covers — CMYK conversion muddies deep blacks). Only make a CMYK PDF/X if their preflight demands it (needs Ghostscript; expect color shift).
  • Size: 150–200 MB is fine to upload. To shrink, re-embed photos as quality-95 JPEG.

Common mistakes

Mistake Fix
Chrome --print-to-pdf → whitespace margins It's the transform bug. Screenshot pages instead.
Downloaded HTML, images 404 ../images/ are relative — render the live URL, don't save HTML locally.
Different DPI in step 1 vs 2 Page boxes mis-size (e.g. 82.8in). Pass the same DPI to both.
Pillow .save(...pdf) for assembly Pillow recompresses to JPEG (lossy). Use img2pdf (lossless).
Added extra bleed in print-shop order Bleed is already in the page size. Order at trim.
Assumed it worked Always run verify_pdf.py — check coverage hits ~100%, not 66%.