pdf-email-optimizer 0.1.0

Shrink PDFs to email-safe sizes while preserving visual quality.

PDF Email Optimizer

Shrink PDFs to email-safe sizes while preserving visual quality.

PDF Email Optimizer is built for posters, brochures, reports, photo-heavy decks, and design-tool exports that need to fit under a target like 5-7 MB. It starts with structural cleanup, recompresses images only when needed, and reports when a requested size conflicts with visual quality.

Install

From a checkout:

python -m pip install -e ".[dev]"
pdf-email-optimizer --help

Once published to a package index:

pipx install pdf-email-optimizer
pdf-email-optimizer input.pdf output.pdf --target-mb 7 --profile quality

Also supported:

uvx pdf-email-optimizer input.pdf output.pdf --target 7mb
python -m pdf_email_optimizer input.pdf output.pdf --target-mb 7

Quick Start

# Ordinary email optimization
pdf-email-optimizer input.pdf output_email.pdf --target-mb 7

# Preserve photos, screenshots, maps, and other detail
pdf-email-optimizer input.pdf output_email.pdf --target 7mb --quality

# Land inside a 5-7 MB range when possible
pdf-email-optimizer input.pdf output_email.pdf --range 5-7mb --quality

# Produce a Markdown report beside the output
pdf-email-optimizer input.pdf output_email.pdf --target-mb 7 --report report.md

# Inspect without writing an optimized PDF
pdf-email-optimizer input.pdf --audit

The source PDF is never overwritten. Existing output files are rejected unless --force is supplied.

Profiles

Profile	Use When	Behavior
quality	Photos, screenshots, maps, product images, "do not degrade" requests	High JPEG floor, protects small images, runs render QA, does not use Ghostscript by default
balanced	General email delivery	Moderate recompression ladder and conservative structural cleanup
aggressive	Smallest file matters more than perfect fidelity	Lower quality floor, smaller long-edge caps, optional Ghostscript fallback

If quality mode cannot hit the requested size, the tool keeps the smallest quality-preserving output and emits a direct warning with next steps.

Output

Use --json for machine-readable summaries:

pdf-email-optimizer input.pdf output.pdf --target-mb 7 --json

The JSON contract is documented in docs/json-output.md and validated by schema/output-summary.schema.json. Important fields include input/output size, target status, strategy, page count, private payload removals, image statistics, render QA, quality status, and warnings.

Gallery

Original page (left) vs. email copy (right). All inputs are synthetic, CC0 fixtures generated by benchmarks/make_fixtures.py; regenerate the images with python benchmarks/make_gallery.py.

InDesign-style export — 2.35 MB → 0.18 MB (92% smaller, PSNR 57.8 dB)

Scanned document — 0.73 MB → 0.25 MB (66% smaller)

Repeated images — 0.81 MB → 0.14 MB (83% smaller, lossless dedupe)

Benchmarks

The benchmark harness runs against the bundled redistributable fixtures:

python benchmarks/make_fixtures.py        # (re)generate CC0 sample PDFs
python benchmarks/run_benchmarks.py --manifest benchmarks/benchmark_manifest.yaml --output benchmarks/results/latest.json

It writes JSON plus a Markdown table. Missing fixtures are marked as skipped so published results stay honest. The table below is generated output (benchmarks/results/latest.md); PSNR/RMS compare the optimized copy against the original render, and inf/0.0 denote a pixel-identical (lossless) result.

Case	Input	Target	Profile	Output	Reduction	Target Hit	Worst PSNR	Worst RMS	Strategy
photo_brochure	1.10 MB	0.6 MB	quality	1.10 MB	0.1%	No	inf	0.0	pikepdf-structural
indesign_export	2.35 MB	1 MB	balanced	0.18 MB	92.3%	Yes	57.822	0.327679	image-recompress
illustrator_export	0.01 MB	7 MB	balanced	0.01 MB	18.6%	Yes	inf	0.0	structural-cleanup
private_payload_export	0.16 MB	7 MB	quality	0.16 MB	0.1%	Yes	inf	0.0	structural-cleanup
screenshot_report	0.27 MB	0.2 MB	quality	0.09 MB	66.4%	Yes	inf	0.0	structural-cleanup
text_vector_document	0.00 MB	7 MB	balanced	0.00 MB	12.2%	Yes	inf	0.0	structural-cleanup
scanned_pdf	0.73 MB	0.4 MB	balanced	0.25 MB	66.6%	Yes	inf	0.0	structural-cleanup
mixed_transparency	1.75 MB	1 MB	quality	1.75 MB	-0.0%	No	inf	0.0	structural-cleanup
embedded_metadata	0.12 MB	7 MB	balanced	0.12 MB	0.1%	Yes	inf	0.0	structural-cleanup
repeated_images	0.81 MB	0.5 MB	balanced	0.14 MB	83.2%	Yes	inf	0.0	structural-cleanup
forms_annotations	0.01 MB	7 MB	quality	0.01 MB	3.9%	Yes	inf	0.0	structural-cleanup
encrypted_pdf	-	7.0 MB	balanced	failed	-	-	-	-	Encrypted PDFs must be unlocked before optimization.

The quality profile deliberately refuses to degrade photo_brochure and mixed_transparency below their targets, emitting a warning instead of shipping a blurry file.

See docs/benchmarking.md before adding fixtures.

Visual QA

Render and compare two PDFs:

pdf-email-render-compare original.pdf optimized.pdf --output-dir qa-renders

This reports page-level pixel differences and can write original, optimized, and amplified diff PNGs for review.

Agent Usage

The repo includes SKILL.md for agent runtimes that load local skills. The short version:

• Use quality when the user asks to preserve image fidelity.
• Use balanced for ordinary email optimization.
• Use aggressive only when visible quality loss is acceptable.
• Report size, target status, strategy, and warnings.
• Never overwrite the source PDF.

More examples are in docs/agent-usage.md.

Development

python -m pip install -e ".[dev]"
pytest
pytest --cov
ruff check .
python -m build

CI runs linting, tests, coverage, package build, and CLI smoke checks on Python 3.9-3.13.

Documentation

• Installation
• Examples
• Benchmarking
• Compatibility
• JSON output
• Agent usage
• Known limitations
• Troubleshooting

License

MIT