pdf-sizer

A command-line tool that converts a folder of images into PDF files, automatically finding the best quality/scale combinations that hit a target file size.

The Problem

When embedding images into a PDF, the resulting file size is often unpredictable. JPEG or PNG files that sum to your desired size may produce a much larger PDF when packaged together – because most PDF libraries decode images and re-store them as raw pixels. This tool solves that by embedding JPEG data directly into the PDF stream (DCTDecode filter), making the file size predictable and controllable.

How It Works

Instead of producing a single PDF, the tool generates multiple variants – each with the same target file size but a different quality/scale tradeoff:

• High quality, low scale – fewer compression artifacts, sharper edges and text, but lower resolution
• Low quality, high scale – more compression artifacts, but higher resolution with more fine detail

For each JPEG quality level (100, 95, 90, ... down to 5), the tool runs a binary search over the scale factor to find the exact scale that produces a PDF within ±2 MB of your target. Only successful variants are saved to disk.

quality=100, scale=62%  →  variant_q100_s062.pdf  (19.8 MB)   ← sharper
quality= 90, scale=73%  →  variant_q090_s073.pdf  (19.5 MB)
quality= 80, scale=82%  →  variant_q080_s082.pdf  (19.9 MB)
quality= 70, scale=91%  →  variant_q070_s091.pdf  (19.7 MB)   ← more detail

You open all variants and pick whichever looks best for your specific images.

Technical Details

• Images are resized using the Lanczos3 filter (high-quality downscaling)
• JPEG bytes are embedded in the PDF with the DCTDecode filter – the original compressed data is stored as-is, not re-encoded
• PDF size ≈ sum of JPEG sizes + ~300 bytes overhead per page
• Built with pdf-writer and image

Installation

Prerequisites

• Rust (edition 2024, Rust 1.85+)

Build

git clone <repo>
cd pdf-converter
cargo build --release

The binary will be at target/release/pdf-sizer (or pdf-sizer.exe on Windows).

Usage

1. Create an images/ folder next to the binary:

   target/release/
   ├── pdf-sizer.exe
   └── images/
       ├── page_01.jpg
       ├── page_02.png
       └── ...
   

2. Run with your target size in megabytes:

   pdf-sizer 20

3. The tool prints progress and saves all variants in the current directory:

   Found images: 5
   Target size: 20 MB (±2 MB)
   
   Searching variants: finding scale for each quality level...
   
     quality=100, scale= 62% -> variant_q100_s062.pdf (19.8 MB)
     quality= 95, scale= 68% -> variant_q095_s068.pdf (20.3 MB)
     quality= 90, scale= 73% -> variant_q090_s073.pdf (19.5 MB)
     quality= 85, scale= 77% -> variant_q085_s077.pdf (20.1 MB)
     quality= 80, scale= 82% -> variant_q080_s082.pdf (19.9 MB)
     quality= 65: even at scale=100% size is 18.1 MB – below target, stopping
   
   --- Summary ---
   Variants created: 5
   

4. Open the variants and visually pick the one that looks best.

Supported Image Formats

jpg, jpeg, png, bmp, tiff, tif, webp

Images are sorted alphabetically, which determines the page order in the PDF.

Output Files

Output files are named variant_q{quality}_s{scale}.pdf and saved in the working directory (where you run the command from), not next to the binary.