Scripting Flickr with Python and REST

January 25, 2006

Uche Ogbuji

Flickr probably needs no introduction for readers of this column. It's a hugely popular social-network site owned by Yahoo, focusing on sharing of photographs. It embodies most of the the current web buzzwords, including tagging, web feeds, AJAX, and accessibility to scripts. Flickr provides a set of HTTP-based APIs for accessing features both as a publisher and as a viewer of pictures. You get to choose between XML-RPC, REST (simple XML over HTTP), or SOAP, and the available functions cover every corner of the core Flickr service. In this article I'll look at some Python libraries for integrating with Flickr (all code tested with Python 2.4.2).

One thing you'll see across examples is reference to a Flickr API key. Such a key is always required to access Flickr through the official APIs, and if you want to take advantage of any of these scripting capabilities, you'll need to apply for a key from Flickr's API Keys page. Flickr uses a fairly elaborate system of token authentication based on your application key, fully described in the Flickr Authentication API Desktop Applications How-To. If you want to know the nitty-gritty details of an application's handshake with Flickr, do read this resource. A few Flickr libraries will try to abstract you from a lot of that, but for purposes of this article I'll dodge the issue by sticking to actions that are allowed without special user authentication. Be sure to browse the Flickr API home page for documentation of technical details of the Flickr API, as well as links to implementations for just about every popular language out there except for C/C++ and JavaScript.


The pioneer of Python Flickr tools is Michele Campeotto, whose work has been the inspiration for many of the libraries I discuss today. He is firstly known for FlickrUploadr, a program that provides a simple GUI using the GTK toolkit for Linux. The user drags and drops image files onto a simple window area, from which the pictures can be uploaded to a Flickr account. FlickrUploadr is focused more on end users and is not really a reusable library, so I shall not spend any more time discussing it. More to this article's purpose is FlickrClient, Campeotto's library module for Flickr. I installed FlickrClient 0.2 by copying the two Python files in the package to my Python site-packages. One of those packages is xmltramp, Aaron Swartz's simple Pythonic tree API for XML. Listing 1 is a simple example that gets all the public favorite photos for the Flickr user named "uche". The variable FLICKR_API_KEY in this listing and others should be set separately to your own Flickr API key.

Listing 1: Interactive session using FlickrClient to get a list of favorite photo titles

>>> from FlickrClient import FlickrClient

>>> client = FlickrClient(FLICKR_API_KEY)

>>> #Find the user ID of the person named "uche"

>>> person = client.flickr_people_findByUsername(username='uche')

>>> import pprint

>>> pprint.pprint(person.__dict__)

{'_attrs': {u'id': u'21902936@N00', u'nsid': u'21902936@N00'},

 '_dNS': None,

 '_dir': [<username>...</username>],

 '_name': u'user',

 '_prefixes': {}}

>>> userid = person(u'id')

>>> print userid


>>> faves = client.flickr_favorites_getPublicList(user_id=userid)

>>> faves[0]

<photo isfamily="1" title="P1010024" isfriend="1" ispublic="1" server="30"

secret="c68a340791" owner="75062596@N00" id="63291069"></photo>

>>> for fave in faves:

...     print fave(u'title')






In Concert

Fela Anikulapo Kuti

Heaven's Light on Lake Malawi


Flickr documents API requests in a form such as flickr.people.findByUsername. In FlickrClient you replace the dots with underscores and call the resulting method name on the Flickr proxy object (client in listing 1). FlickrClient does some remote method dispatch magic to forward the request to Flickr. The actual Flickr API is not hard-coded into FlickrClient (which is a mere 40 lines of code). You almost always have to pass at least one named parameter to Flickr. You must pass these as keyword arguments (for example, username='uche') for the dispatch magic to work; you can get the parameter name from the Flickr API documentation. If you make any mistakes that confuse FlickrClient you can expect some pretty cryptic error messages, but luckily the API is simple enough that you get the hang of it very quickly. You get back xmltramp nodes representing the Flickr response XML, and it's up to you to access the actual data you need. As in listing 1, I use __dict__ and repr() heavily for introspection of the result values so I can find out where to get the data I need. Some calls return lists, which come back as a container element with multiple children, as is the case for flickr.favorites.getPublicList.

As I mentioned, I use only Flickr calls that do not need user authentication. An example of a method that does is flickr.favorites.getList which gets all favorite photos including private ones. If you want to write code with Flickr user permissions in order to use such functions you have to redirect users to a web page so that they can log in and give your application permission. You then query Flickr again for a resulting authentication token as a result. It's a bit of a tricky process, and later on I cover a FlickrClient-derived tool that tries to do this handshake for you. You might be able to borrow code from there even if you're using FlickrClient.


FlickrClient uses the REST API, requesting simple parameterized URLs from Flickr via HTTP GET (even in some cases where Flickr should be requiring HTTP POST, but does not bother), and returning the raw XML response bodies. This keeps it very simple and flexible. Eitan Isaacson preferred to use Python standard library XML-RPC instead, so he wrote flickrlib. I downloaded version 0.5 of the package. Installation is a matter of the standard distutils python install, although all it does is copy the single file to your Python library. Listing 2 is an attempt to replicate the requests in listing 1. The first thing you'll notice is that method names are called exactly as in the Flickr API spec, and there is no need to translate the dots to underscores. The variable FLICKR_API_SSECRET in this listing and others should be set separately to your own shared secret.

Listing 2: Interactive session using flickrlib to get a list of favorite photo titles

>>> import flickrlib

>>> client = flickrlib.FlickrAgent(FLICKR_API_KEY, FLICKR_API_SSECRET)

>>> #Find the user ID of the person named "uche"

>>> person ='uche')

>>> person

{u'username': [{u'text': u'Uche', u'type': u'username'}],

 u'text': u'', u'type': u'user', u'id': u'21902936@N00',

 u'nsid': u'21902936@N00'}

>>> userid = person[u'id']

>>> faves =


UnicodeEncodeError: 'ascii' codec can't encode character u'\xc1' in

position 270: ordinal not in range(128)

I get this exception because there is a non-ASCII character in one of my favorite photo titles. flickrlib doesn't seem to handle this properly yet. In listing 3, I tried a different query with a list-like response.

Listing 3: Interactive session using flickrlib to get information on a photo

>>> import flickrlib

>>> client = flickrlib.FlickrAgent(FLICKR_API_KEY, FLICKR_API_SSECRET)

>>> #Find the user ID of the person named "uche"

>>> person ='uche')

>>> userid = person[u'id']

>>> #Get the first 10 photos for the user

>>> photos =, per_page=10)

>>> photos.keys()

[u'perpage', u'text', u'page', u'photo', u'total', u'type', u'pages']

>>> photo = photos[u'photo'][0]

>>> photo[u'title']


>>> photoinfo =[u'id'],


>>> for tag in photoinfo[u'tags'][0][u'tag']: print tag[u'text']





flickrlib turns XML-RPC responses into a data structure of dictionaries and lists. The best way to figure out how to navigate to what you want is to look at the sample XML response given in the Flickr API docs. For example, from you know the raw response is along the lines of:

<photo id="2733" secret="123456" server="12"

  isfavorite="0" license="3" rotation="90" originalformat="png">



    <tag id="1234" author="12037949754@N01" raw="woo yay">wooyay</tag>

    <tag id="1235" author="12037949754@N01" raw="hoopla">hoopla</tag>




From this you can figure out that you would get the photo's original format using response[u'originalformat'], the list of tags from response[u'tags'][0] (the first child element named tags), the first tag's display text from response[u'tags'][0][u'tag'][0][u'text'], and so on. XML attributes generally are accessed as dictionary keys, and XML child elements as list items.

Do mind Etian's warning where he says,

The library currently lacks a custom error class and it is not thread safe, and when I say that I mean it will do crazy stuff if you don't put mutex locks around calls to this library.

To be fair, all the Python Flickr libraries I've seen leave a lot of such matters for the developer to manage.

I've mentioned how the above Flickr APIs return fairly low-level data structures or raw XML for the developer to pick apart. This has the advantage of simple library code at the expense of more complex client code. James Clarke preferred to do the work to simplify the Flickr results in creating I downloaded the single Python file (revision 24) and copied it to my Python library by hand. Pay attention to how I set the API key directly in the module object.

Listing 4: Interactive session using to get information on a photo

>>> import flickr


>>> user = flickr.people_findByUsername(u'uche')



>>> photos = flickr.photos_search(

>>> for tag in photos[0].tags: tag.text




>>> faves = flickr.favorites_getPublicList(

>>> for fave in faves: print fave.title






In Concert

Fela Anikulapo Kuti

Heaven's Light on Lake Malawi


You can see how much more simple and direct the API is. The trade-off is that it takes a lot of work within, and not all the Flickr API is covered yet. You can see the module documentation (help(flickr)) for information on what is and is not covered. You might also have noticed that the function names are completely specialized from Flickr method names, so you have to rely on documentation rather than Flickr's own API docs.

Beej's Python Flickr API

Beej Jorgensen put together a simple Python library supporting the Flickr API, called He borrowed some code and reinvented some odd wheels, bundling an xmltramp-like XML node implementation right into the main module. He says he did so as a learning exercise, and I expect another motivating factor was avoiding any third-party module dependencies. At any rate, it weighs in at under 200 lines of code, so I won't scrutinize its design decisions too closely for now. The more important issue is that I found it a bit clumsy to use, as far as I got, and I couldn't really get it to do anything. Listing 5 illustrates how far I got.

Listing 5: A sample interactive session attempting to use

>>> from flickrapi import FlickrAPI

>>> #initialize a proxy instance for the Flickr API and get a session token



>>> token = fapi.getToken(perms='write', browser='firefox')

rsp: error 108: Invalid frob

The last line, looking to grab a token from Flickr, launches a separate browser session for the end user to log into Flickr and grant the application permission to act as the user's agent. Unfortunately, no matter how I tweaked the requested permissions and the web browser used for the request, I kept getting the "Invalid frob" error.


Another end-user-focused project that can be a useful source of Python code is C. Mallory's uploadr, which uploads all pictures in a given directory that have not already been uploaded to Flickr.

It's good to have options, and there are certainly many if you're needing to access Flickr from Python code. The different libraries have different philosophies and you should be able to find one to fit your needs.

As an aside, it's interesting to look at photos tagged with "python" on Flickr. Because of the popularity of Flickr among programmer types I had guessed that there would be more screenshots from Python and photos from Python developer events than photos of snakes. Evidence from the first few pages for this tag (there is a nice photo sequence of a python swallowing a large dead rodent) gives the lie to that preconception. On the other hand, Google search results for "python" offer nothing but the programming language and Monty Python comedy troupe for pages and pages of results. It's an interesting contrast, whether or not you are a Semantic Web advocate.