Small script and descriptions of steps to extract a GIF for a matplotlib animation widget embedded in a Jupyter notebook.

save_matplotlib_animation_from_notebook.py

# Matplotlib animations can be saved as a widget in a Jupyter notebook.
# It's unclear to how to recover it in Python once the notebook kernel has been restarted.
# But the image data is there in the notebook, embedded in a JS script in the output of the cell
# with the individual frames encoded as base64 png images.
# So here's the way I found to extract the images and make a GIF out of them.
# It's not all automatic and needs some manual action.
# Steps:
#   1/ Open the notebook in a web browser. Open the developer tools, and in the console, start typing "anim"
# then TAB to autocomplete. A variable with a unique name looking like "animae693007f74d44b0b917c61003e9490f" should appear.
# Cut this variable name, and copy it in the following command that you can then execute:
#    JSON.stringify(animae693007f74d44b0b917c61003e9490f.frames.map(function(elem) {return elem.src;}))
# Copy the output that appears.
#   2/ Open your favorite text editor, preferably not terminal-based; the output can be several Mo of text,
# and terminal usually don't react well to this. I used Sublime Text for this step.
# Paste the output, and save it under "frames.json"
#   3/ Create a "frames" directory next to it
#   3/ Fire an IPython terminal, pip install imageio if you don't have it, and execute the code below.
#   4/ You can now admire the created GIF and marvel at the wonders of technology.

import base64
import json
import imageio

with open("frames.json") as f:                                                  
    frames = json.load(f)

print("Found %d frames." % len(frames))

# Exporting individual frames as PNG images
for idx, frame in enumerate(frames):                                                   
    with open("%d.png" % (idx + 1), "wb") as f:              
        f.write(base64.b64decode(frame.split("base64,")[1]))

# Opening all frames and creating a GIF
# With the options there it will loop, and each frame will be displayed for 200ms
with imageio.get_writer('./my_gif.gif', mode='I', duration=0.2) as writer:
    filenames = ["./frames/%d.png" % i for i in range(1, 31)]
    # If you want to have the GIF looping back-and-forth, you can replace the for-loop below by the commented one
    # for filename in filenames + filenames[:-1:-1]:
    for filename in filenames:
        image = imageio.imread(filename)
        writer.append_data(image)