imgsqz

Tiny, dockerized image compression toolbox. Bring best-of-breed encoders/optimizers to any project with one container.

Ships a curated set of CLI tools (mozjpeg, jpegoptim, guetzli, optipng, pngcrush, advpng) and a lean runtime so you can batch-optimize images locally or in CI without installing native deps.

---

Why

• Consistent results anywhere → same binaries on macOS, Linux, CI.
• No native install pain → everything runs inside a container.
• Composable → mix lossless & lossy steps per format.

---

Acknowledgements

• The authors and maintainers of mozjpeg, jpegoptim, guetzli, optipng, pngcrush, and advancecomp.
• Inspired by the need for a portable, zero-install image optimization toolkit for local use and CI.

---

What's inside

Format	Primary tools	Typical use
JPEG	mozjpeg (cjpeg/jpegtran), jpegoptim, guetzli*	High-quality lossy encode; progressive; metadata stripping; lossless transforms
PNG	optipng, pngcrush, advpng	Lossless structural optimizations; brute-force parameter search; advanced recompression

• Guetzli is very slow and mostly of historical interest; prefer mozjpeg for production.

---

Quick start

Build the image

git clone https://github.com/flavioheleno/imgsqz
cd imgsqz
docker build -t imgsqz .

Run it against your images

Mount your working directory into /app:

# List available tools inside the container
docker run --rm imgsqz sh -lc 'cjpeg -version || true; jpegtran -version || true; jpegoptim --version; optipng -v | head -n1; pngcrush -version | head -n1; advpng --version | head -n1'

---

Common recipes

All commands assume the current host directory is mounted at /app.

JPEG: high-quality encode (mozjpeg)

# Re-encode a JPEG with mozjpeg (quality 80, progressive, trellis, DC scan)
docker run --rm -v "$PWD:/app" imgsqz \
  sh -lc 'cjpeg -quality 80 -progressive -tune-ssim -optimize -baseline -sample 1x1 -outfile /app/out.jpg /app/in.jpg'

JPEG: lossless size reduction (jpegtran → jpegoptim)

# Normalize & strip metadata without re-encoding pixels
docker run --rm -v "$PWD:/app" imgsqz \
  sh -lc 'jpegtran -copy none -optimize -progressive /app/in.jpg > /app/tmp.jpg && \
          jpegoptim --strip-all --preserve --totals /app/tmp.jpg && \
          mv /app/tmp.jpg /app/out.jpg'

PNG: exhaustive lossless optimization (optipng)

docker run --rm -v "$PWD:/app" imgsqz \
  sh -lc 'optipng -o7 -strip all -out /app/out.png /app/in.png'

PNG: alternative pass (pngcrush)

docker run --rm -v "$PWD:/app" imgsqz \
  sh -lc 'pngcrush -rem alla -brute -reduce /app/in.png /app/out.png'

PNG: advanced recompression (advpng)

docker run --rm -v "$PWD:/app" imgsqz \
  sh -lc 'cp /app/in.png /app/out.png && advpng -z4 /app/out.png'

Batch a whole folder (recursive, keep structure)

# JPEGs: re-encode to mozjpeg q=78 keeping tree
docker run --rm -v "$PWD:/app" imgsqz \
  sh -lc 'find /app/src -type f \( -iname "*.jpg" -o -iname "*.jpeg" \) -print0 | \
          xargs -0 -I{} sh -c '\'' \
            out="/app/dist/${0#/app/src/}"; mkdir -p "$(dirname "$out")"; \
            cjpeg -quality 78 -progressive -optimize -outfile "$out" "$0" \
          '\'' {}'

---

License

This project is licensed under the MIT License.