Last built 2 years ago using go version go1.4.2 linux/amd64 and 72 seconds of time

File Size
Project-Readme for this version

picel Build Status Coverage Status GoDoc

picel is a light-weight, blazing fast REST-ful micro service for image processing with a lean API.

It does one thing and does it well: process images like a UNIX filter (see Basics of the UNIX Philosophy).


  1. Download the latest picel binary release for your platform from the releases page.
  2. Run it with no arguments or with something like --backend localhost:8080
  3. Use picel-js to encode your images.

The default port is 8123. Change it with --addr :8000 to listen on port 8000 (for example).

picel --help for more help.

tl;dr Docker

There’s a docker container you can get from the docker HUB registry as well.

You can get the henvic/picel container up and running quickly with

docker pull henvic/picel
docker run -d -p 8123:8123 picel

picel middleware between caching and storage

Use sane defaults to make your caching rules. Consider that images are large binary blobs that seldom change, are large in size, and hard to process. It is wise to consider carefully appropriate caching rules to avoid premature removal. See Real Time Resizing of Flickr Images though to see their new approach replacing ImageMagick with a (not open sourced so far) GPU library of their own.

Also, you want your proxy layer to have protection against abuse (enhance your calm to avoid trying to process too many suspicious requests). Refer to rfc6585#section-4 to know more. Modern HTTP servers such as nginx or HAProxy already have options to deal with such attacks.

picel is designed to be used in the wild, processing untrusted, user uploaded data (but it’s not been used in production so far and its performance - despite the light-weight on the very first sentence of this README - is not even being measured with metrics now).

Defaults, performance friendly, and more

By default, picel will try to use webp if the user doesn’t explicitly request another format and his client announces it accepts it (Chrome, for example).

Also, JPEG is the default image format for input.

The provided binaries are built without pprof support. You can compile yourself if you want it. The docker image provided has pprof support out of the box, but with a firewall rule to filter calls to it.

You can run picel (or any Go server for that matter) with pprof support enabled by default with no issues or performance penalties as long as you filter untrusted requests to /debug paths.

To build with pprof support use make build-with-pprof.


picel uses webp and ImageMagick. At startup it will warn if it doesn’t find the binaries for these processes. If you don’t have it (or are running old versions) use your operating system package manager system to install the newest versions.

libmagic is also used for discovering the mime type of the source files.


GET /<backend>/<id><params>.<output>

  • backend: image storage end-point (only when picel server is open / unrestricted)
  • id: part of the path before the last ‘.’
  • params: image manipulation parameters / raw
  • output: output delivery format

The id and output MUST be escaped by _ (underscore).

backend is the server backend, to be used if picel was not started with the --backend flag. It should be a host without a leading http:// or, in case of https://, with a s: prepended.

Available params are:

  1. raw
  2. crop {x, y, width, height}
  3. dimension {width, height}
  4. extension

Parameters MUST be given in this order or, otherwise, picel will not recognize them (this is by design on purpose, to avoid having multiple encoding implementations doing things differently / guarantee more cache hits when using a caching layer).

  • raw is a parameter without value and MUST NOT be used along others. It implies that picel SHOULD return the original file from the backend. This option might not be available.
  • crop MUST be given using the format <x>x<y>:<width>x<height> as in 0x0:100x200
  • width and height are pixel integers using the format <width>x<height>, when one is neglected the resizing is made proportional
  • extension is a string, when it’s the same as of the output it is discarded

All parameters are prefixed by a _ (underscore).

GET with request body

You can also make requests to the “/” end-point with a JSON-based request body with the following parameters:

  • backend (url string)
  • path (string)
  • raw (boolean)
  • crop (object wit x, y, width, height)
  • width (number)
  • height (number)
  • output (number)

The path parameter is required.


To help debugging you can use append the ?explain to a URL in order to get a JSON response that will tell you how a image was transformed (or failed to be).

Example: curl https://localhost:8123/ will return the requested image (if it exists).

curl https://localhost:8123/ will return an object like the following:

    "message": "Success. Image path parsed and decoded correctly",
    "path": "/",
    "transform": {
        "image": {
            "id": "foo",
            "extension": "jpg",
            "source": ""
        "path": "/foo_137x0:737x450_800x600_jpg.webp",
        "original": false,
        "width": 800,
        "height": 600,
        "crop": {
            "x": 137,
            "y": 0,
            "width": 737,
            "height": 450
        "output": "webp"
    "errors": null

Please notice that ?explain can only tell if a request is not bad and does NOT verify if processing works or even if an image exists on the backend server. If you just need to verify it process correctly you can judge by getting the Content-Length from a 200 OK’ed HEAD request.

Also notice that the file is not loaded to execute the explain so its mimetype is not returned.

For GET requests with body the path value will be calculated and given on the path key.

Encoding libraries

Currently you can encode urls using either the go package or using the auxiliary JavaScript encoding library that can be conveniently installed with npm or bower (package name is picel). If you need the encoder library available for another language let me know.

If you need to write URLs by hand take a look at the examples for the JS encoder and its source code.


In lieu of a formal style guide, take care to maintain the existing coding style. Add unit tests for any new or changed functionality. Check your code with go fmt, go vet, go test, go cover, and go lint.

test_assets branch

The test_assets branch exists to the sole purpose of serving as a branch for binary image files for the integration tests. It only contains binary files (and nothing else) and maybe rebased at any time since its history doesn’t matter. It’s used to checkout the test_assets directory whenever the tests are run.

Currently the images on the test_assets branch, besides AdditiveColor, are the following photos I’ve taken and published on my Flickr account: