Related issue: #2384

I am happy to work on a PR if this has a chance of being accepted.

The facecolor can be an RGBA value so setting

mpl.rcParams['figure.facecolor'] = (0, 0, 0, 0)

should do what you want?

Hi Thomas,

I tried this, but full RGBA values are not supported in matplotlibrc. When I set this

figure.facecolor  : (0, 0, 0, 0)

I get this error:

/Users/hdembins/Library/Python/2.7/lib/python/site-packages/matplotlib/__init__.py:1120: UserWarning: Bad val "(0, 0, 0, 0)" on line #26
    "figure.facecolor  : (0, 0, 0, 0)
"
    in file "/Users/hdembins/.matplotlib/matplotlibrc"
    Key figure.facecolor: (0, 0, 0, 0) does not look like a color arg
Color tuples must be length 3
  (val, error_details, msg))

So perhaps the more general solution is to allow full RGBA colors in matplotlibrc.

Update: It also turns out the following code does not work as expected:

import matplotlib.pyplot as plt
from matplotlib import rcParams

rcParams["figure.facecolor"] = (0, 0, 0, 0)

plt.figure()
plt.plot([], [])
plt.savefig("testme.png")

Opening testme.png shows a white background, not a transparent background for the figure patch. I think that is a bug. The following works:

import matplotlib.pyplot as plt

fig = plt.figure()
fig.patch.set_alpha(0)
plt.plot([], [])
plt.savefig("testme2.png")

See the result:

To me it looks like the figure is transparent but the axis is not. That is what I would expect when you set the figure transparent and the axes facecolor has its default value which is white.
I would expect

rcParams["axes.facecolor"] = (0, 0, 0, 0)

to do the trick

Hi Jens, yes, I had the same expectation from my naive point of view.

And just to clarify: the attached picture (testme2.png) is the desired outcome that I want.

Hi Jens, sorry, I was not reading closely enough what you wrote.

rcParams["axes.facecolor"] = (0, 0, 0, 0)

does not work either, it produces just a white figure background and a white axes background. Also, I don't want the axes background to be transparent, only the figure background.

Ok I assumed that the attached image was the broken behaviour

The right rcparam to set is savefig.facecolor.
Right now you can work around the lack of support for 4-tuples using #00000000 (i.e. #rrggbbaa format, introduced in #6382); #9038 adds support for 4-tuples.

@anntzer Thanks for the related PR :). I take whatever works, but I think that setting savefig.facecolor to get the desired result is not intuitive.

Hmm, we change savefig.dpi to default to 'figure', I thought we had done the same with the rest of them.

There is also 'savefig.transparent' which I think does the same thing in a more direct way (which I had forgotten about apparently).

@tacaswell I read about savefig.transparent but it does not what I need. It makes the figure background and the axis background transparent. I only want the figure background transparent, but keep the axes background white, i.e. what is shown in the screenshot above.

@anntzer I can confirm that this code

import matplotlib.pyplot as plt
from matplotlib import rcParams

rcParams["savefig.facecolor"] = (0, 0, 0, 0)

fig = plt.figure()
plt.plot([], [])
plt.savefig("testme2.png")

produces exactly what I want. Thanks! 😄

Following your advice as I understood it, I tried to add this line to my matplotlibrc:

savefig.facecolor: #00000000

But I then get this error:

/Users/hdembins/Library/Python/2.7/lib/python/site-packages/matplotlib/__init__.py:1120: UserWarning: Bad val """ on line #31
    "savefig.facecolor: "#00000000""
    in file "/Users/hdembins/.matplotlib/matplotlibrc"
    Key savefig.facecolor: " does not look like a color arg
  (val, error_details, msg))

Okay, looking into the relevant code myself helps ;)
It works with the current matplotlib 2.0.2, if I put this line into matplotlibrc

savefig.facecolor: 00000000

Note that one has to omit the #-prefix, which is not intuitive. I think the color format in matplotlibrc should be identical to the one in Python.

Ok, nevermind, the # in matplotlibrc starts a comment, that's why the color cannot be prefixed like this. I guess it is not worth the effort to make an exception in the parser for this sort of thing.

This is fantastic. Thanks everyone!