Print-ready PDF pipeline (PowerPoint export)

This project defines a deterministic, folder-based workflow to convert a PDF exported from PowerPoint into a print-ready PDF for inside pages only. Covers are explicitly out of scope.

The workflow is designed for Linux, fully scriptable, and suitable for execution by an AI agent. It is written specifically for Ubuntu LTS; the install.sh installer relies on APT and will not work on other systems.

The core principles are:

• Deterministic and reproducible output
• One folder per step, no in-place modification
• Filenames always derived from the original input filename
• Full audit trail with reports per step
• Vector pages stay vector whenever possible
• Rasterization and upscaling only where strictly necessary

Assumptions

• The PowerPoint source already uses the correct final page size.
• Minimum safe margin of 5 mm is already respected inside PowerPoint.
• The PDF in 00-input was exported directly from PowerPoint using high-quality settings.
• Inside pages are delivered as single pages, not spreads.
• Target effective resolution for raster content is ≥300 dpi at final size.

Non-goals

• Editing layout, text, or margins
• Fixing PowerPoint design mistakes
• Cover, spine, bleed, or binding calculations
• Reflowing or reconstructing vector content

Repository structure

00-input/
01-validate/
02-analyze-dpi/
03-extract-images/
04-upscale-images/
05-verify-images/
06-resize-images/
07-resize-smasks/
08-replace-images/
09-normalize-pdf/
10-pdf-x4/
11-output/
preflight.sh (stdout only)

Each step reads only from earlier steps and writes only to its own folder.

Running the full pipeline

./convert.sh 00-input/boek.pdf

Runs steps 1–11 in order and stops on the first failure. Then runs preflight on both the PDF/X-4 and PDF/X-1a outputs. All script output is streamed to the terminal.

Removing converted artifacts

./remove-converted.sh 00-input/boek.pdf

Removes all pipeline outputs for the specified input (e.g., folders like 03-extract-images/boek/ and files like 08-replace-images/boek.*), while keeping the original file in 00-input/.

Naming conventions

Input file:

00-input/boek.pdf

Derived names always preserve the base name boek.

Image-based outputs are grouped in a folder named after the document:

03-extract-images/boek/...
04-upscale-images/boek/...
05-verify-images/boek/...
06-resize-images/boek/...
07-resize-smasks/boek/...

Step-by-step workflow

00-input

Purpose Starting point. Contains the PDF exported from PowerPoint.

Rules

• Treat as read-only.
• No renaming after ingestion.

Contents

00-input/boek.pdf

01-validate

Purpose Sanity check and baseline metadata extraction. No mutation.

Checks

• Page count
• Page size consistency
• Encryption
• Basic color space detection
• File hash

Outputs

01-validate/boek.validated.txt

Fail if

• Page sizes differ
• PDF is encrypted or unreadable
• Page count is zero

02-analyze-dpi

Purpose Identify embedded raster images below the target DPI.

Definition

Effective DPI = pixel resolution of a raster image relative to the physical size it is placed at on the page.

Outputs

02-analyze-dpi/boek.dpi.csv
02-analyze-dpi/boek.lowdpi.images.csv

boek.lowdpi.images.csv contains one image per line with page number, object id, and DPI.

Fail if

• DPI analysis cannot be computed

Control flow decision

If no low-DPI images are found in step 02:

• Steps 03, 04, 05, 06, and 07 are skipped.
• The pipeline continues directly with step 09 (normalize PDF) using the original PDF.

In other words:

02-analyze-dpi
   ├─ low-DPI images found → 03 → 04 → 05 → 06 → 07 → 08 → 09 → 10 → 11
   └─ no low-DPI images    → 09 → 10 → 11

After step 11, run `preflight.sh` on the output PDF you want to validate.

This ensures:

• No unnecessary rasterization
• Output remains fully vector where possible

03-extract-images (conditional)

Purpose Extract embedded raster images from pages that contain low-DPI content.

Inputs

• Images listed in boek.lowdpi.images.csv

Outputs

03-extract-images/boek/obj-<object>-<id>.png
03-extract-images/boek.images.csv

Rules

• Lossless output (PNG or TIFF)
• Preserve original pixel dimensions

04-upscale-images (conditional)

Purpose Upscale only the extracted images so effective resolution meets or exceeds target DPI.

Strategy

• Compute required scale factor per image
• Clamp to reasonable bounds (e.g. max x4)
• Do not blindly upscale everything

Outputs

04-upscale-images/boek/obj-<object>-<id>.up.png

Fail if

• Upscaler fails
• Output dimensions do not match expected scale

05-verify-images (conditional)

Purpose Verify that upscaled images meet the target DPI. Copies any that still fall short.

Outputs

05-verify-images/boek.verify.csv
05-verify-images/boek/obj-<object>-<id>.up.png

06-resize-images (conditional)

Purpose Non-AI resize for any images that still miss target DPI after AI upscaling.

Outputs

06-resize-images/boek/obj-<object>-<id>.up.png

07-resize-smasks (conditional)

Purpose Resize soft masks (SMask) to match the resized image dimensions.

Outputs

07-resize-smasks/boek/obj-<object>-<id>.up.png

08-replace-images (conditional)

Purpose Replace the original low-DPI image objects in the PDF with the upscaled versions, preserving vector content. Replacement images are converted to CMYK to avoid color conversion during normalization.

Outputs

08-replace-images/boek.pdf
08-replace-images/boek.replace.txt

09-normalize-pdf

Purpose Prepare final print-deliverable PDF.

For PDF/X-4, normalization embeds the required output intent (ICC profile) without rasterizing or resampling content.

Typical actions

• Flatten transparency
• Convert to CMYK
• Export to required PDF standard (PDF/X-4)
• Disable resampling

Outputs

09-normalize-pdf/boek.pdf
09-normalize-pdf/boek.normalize.txt

10-pdf-x4

Purpose Set print trim and bleed boxes. Uses a 3 mm trim inset by default.

Outputs

10-pdf-x4/boek.pdf
10-pdf-x4/boek.trim.txt

11-output

Purpose Convert PDF/X-4 output to PDF/X-1a.

Outputs

11-output/boek.pdf

preflight.sh

Purpose Final verification.

Checks

• Page count and sizes
• Color space
• Remaining RGB objects
• Remaining low-DPI issues
• File integrity

Outputs

Prints to stdout only.

Fail if

• Page sizes differ
• Low-DPI issues remain

Use preflight directly on a concrete file:

./preflight.sh 10-pdf-x4/boek.pdf
./preflight.sh 11-output/boek.pdf

The default final deliverable is 10-pdf-x4/boek.pdf, with optional PDF/X-1a output in 11-output/.

Configuration

All tunable values must live in a single config file:

TARGET_DPI=300
RASTERIZE_DPI=400
MAX_UPSCALE=4.0
UPSCALER_MODEL=RealESRGAN_x4plus
IMAGE_FORMAT=png
COLOR_PROFILE=/usr/share/color/icc/colord/FOGRA39L_coated.icc

Each report must log the effective configuration used.

Print Specs Summary (New Energy)

These requirements are summarized from New Energy’s Dutch print delivery specifications and are provided for convenience (inside pages only; covers are out of scope here).

• Images should be ≥300 dpi; below 240 dpi risks visible quality loss. Avoid web‑sourced images due to low quality/rights.
• Add 3 mm bleed for inside pages; bleed artwork must extend into the bleed area.
• Minimum line thickness: 0.1 mm (or 0.4 mm for foil finishes).
• Export as PDF/X-4 for print delivery in this pipeline (transparency preserved).
• Use CMYK only (no RGB); total ink coverage should not exceed 280%.
• Deep black (typically for covers): C50 M40 Y40 K100. Text/line art in body should be K100 only.
• Include trim marks on export; keep offsets outside the bleed.
• Deliver cPDF (certified PDF) and export using PDF/X‑4 presets. Profile used by default: /usr/share/color/icc/colord/FOGRA39L_coated.icc (Coated FOGRA39 / ISO 12647-2:2004).

Agent instructions

An AI agent working on this repository must:

• Never modify files in place
• Always write outputs to the next step folder
• Fail fast on invariant violations
• Skip steps cleanly when no-op conditions apply
• Produce human-readable reports at every step

Here is the clean, final list of required tools, aligned with the workflow as described. No extras, no overlap.

Required tools (core)

These are needed to run the pipeline end to end.

PDF inspection and manipulation

• poppler-utils Used for:

  • pdfinfo, page size and page count
  • pdfimages, extracting embedded raster images
  • pdftoppm, rasterizing pages

• ghostscript Used for:

  • transparency flattening
  • CMYK conversion
  • PDF/X export
  • final print normalization

• qpdf Used for:

  • safe PDF object inspection and replacement
  • sanity checks
  • metadata inspection

Image inspection and processing

• ImageMagick Used for:

  • verifying image dimensions
  • confirming DPI after rasterization and upscaling
  • format conversion (PNG, TIFF)

• exiftool Used for:

  • inspecting image metadata
  • validating DPI and color info in images

AI upscaling

• Real-ESRGAN (external, not via apt) Used for:

  • upscaling extracted raster images that fall below target DPI
  • GPU-accelerated where available

Recommended model:

• RealESRGAN_x4plus

Scripting and orchestration

• bash Primary orchestration language.

• coreutils (sed, awk, grep, cut, sort, uniq, wc) Used for:

  • parsing reports
  • page list generation
  • control flow decisions

• jq (optional but recommended) Used if DPI reports are emitted as JSON instead of CSV.