Completely out of my wheelhouse here, but the comment in the matplotlibrc seems to suggest that bbox_inches='tight' should work for at least some writers.

Also if you set bbox_inches to a specific BBox, would it not produced the same size image each frame?

After going down a far deeper rabbit hole than I expected (that involved trying and failin gto get a mpeg4 decoder installed on fedora) to see what ffmpeg -i ... does with input files that vary in size.

import matplotlib.pyplot as plt

fig, ax = plt.subplots()

for j in range(100):
    ax.set_title(f"frame {j}")
    fig.set_size_inches(j % 10 + 2, j % 10 + 2)
    fig.savefig(f"/tmp/mv/frame-{j:04d}.png")

ffmpeg -i 'frame-%04d.png' -vcodec av1 test.mp4

It seems to work (but I would call the output more artistic than informative). I think it is scaling the subsequent files to match the size of the first frame so the axes move / resize frame to frame.

We already do some checking in Aimation.save to protect the user from this sort of thing going wrong:

We talked about this on the call today and the consensus is that while it may "work" with the file based writers, it is likely not what a user wants, if the user wants to save varying size pngs they can still do so (and will have to call ffmpeg them selves), and because this is a lower-level API than Animation.save it makes sense to be very strict an raise (rather than just warning).

I also think there is a consistency arguement that if we prevent something with some writers (because it silently produces totally broken output) we should prevent the same thing in other writers (where it would silently produces silly output).

@rcomer I updated the comment in the rcparams.