A lossless image optimiser script for displaying images on the Web. It's a Python command-line tool which is an implementation of Yahoo's smush.it service that can be run over whole directories of images.
Switch branches/tags
Nothing to show
Pull request Compare This branch is 5 commits ahead of thebeansgroup:master.
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.


smush.py - Optimise images for the web

Smush.it is a service from Yahoo that applies lossless optimisations to images for display on the web. Unfortunately they don't provide an API, and sometimes it's convenient to have a local command-line script.

smush.py uses the same tools as Smushit (read more at: http://oreilly.com/server-administration/excerpts/9780596522315/optimizing-images.html), but can be run locally.

The following lossless optimisations are performed:

  • GIFs - If they're animated GIFs, they are optimised with Gifsicle. If they aren't animated, they're converted to PNGs with ImageMagick, then optimised as PNGs as below.
  • PNGs - Quantised with pngnq, then crushed with pngcrush.
  • JPGs - Optionally remove ALL metadata (it may not be legal to remove copyright notices, so only use this on images you own the copyright to or that don't have copyright notices). If they're larger than 10kb, they're converted to progressive JPGs. Compression is optimised with jpegtran.

Note: If a GIF is converted to a PNG, it keeps the old .gif file extension in case the file name is in a database.


Install all of the following required software:

Compile, patch, and install pngnq (read on):

In the contrib directory is a file pngnq-x.x.diff. This must be applied against v1.0 and v1.1 of pngnq.c - the patch has been submitted back to the author, so it's possible that later versions will include the patched functionality (check for the existence of a -o option that allows an output file name to be specified). In that case, then it's not necessary to apply the patch.

Apply the diff using patch, e.g.:

wget http://sourceforge.net/projects/pngnq/files/pngnq/1.1/pngnq-1.1.tar.gz
tar -xzvf pngnq-1.1.tar.gz
cd pngnq-1.1/src
patch < ~/smush.py/contrib/pngnq-1.1.diff

Then configure and make as usual.


This script is intended to be run over a collection of existing images - GIFs (perhaps animated), PNGs and JPEGs. It will perform lossless optimisations and will OVERWRITE THE ORIGINAL file. It is intended to be able to be used to optimise images that have a reference stored in a database, hence the reason for not modifying input file names at all.

It can be run as follows:

python smush.py /path/to/file/or/directory(ies)

where /path/to/file/or/directory(ies) is a file or directory path, or list of space-separated paths to files or directories that will be optimised.

Or type:

python smush.py --help

for usage information.

It is safe to run this script multiple times on the same files since all operations are lossless. In fact, PNG images may be optimised further by repeatedly smushing them.

Unless the -q option is given, statistics will be displayed when smushing has finished - these stats are approximate. GIFGIF refers to animated GIFs.

Note: This software has only been tested on Centos 5 Linux.