Making GIFs From Video Files With Python

[Turk]

By Zulko •Jan 23rd, 2014

Sometimes producing a good animated GIF requires a few advanced tweaks, for which scripting can help. So I added a GIF export feature to MoviePy, a Python package originally written for video editing.

For this demo we will make a few GIFs out of this trailer:

Converting a video excerpt into a GIF

In what follows we import MoviePy, we open the video file, we select the part between 1’22.65 (1 minute 22.65 seconds) and 1’23.2, reduce its size (to 30% of the original) and save it as a GIF:

1 from moviepy.editor import *
2
3 VideoFileClip("./frozen_trailer.mp4").\
4 subclip((1,22.65),(1,23.2)).\
5 resize(0.3).\
6 to_gif("use_your_head.gif")

Cropping the image

For my next GIF I will only keep the center of the screen. If you intend to use MoviePy, note that you can preview a clip with clip.preview(). During the preview clicking on a pixel will print its position, which is convenient for cropping with precision.

1 kris_sven = VideoFileClip("./frozen_trailer.mp4").\
2 subclip((1,13.4),(1,13.9)).\
3 resize(0.5).\
4 crop(x1=145,x2=400).\ # remove left-right borders
5 to_gif("kris_sven.gif")

Freezing a region

Many GIF makers like to freeze some parts of the GIF to reduce the file size and/or focus the attention on one part of the animation.

In the next GIF we freeze the left part of the clip. To do so we take a snapshot of the clip at t=0.2 seconds, we crop this snapshot to only keep the left half, then we make a composite clip which superimposes the cropped snapshot on the original clip:

1 anna_olaf = VideoFileClip("./frozen_trailer.mp4").\
2 subclip(87.9,88.1).\
3 speedx(0.5).\ # Play at half speed
4 resize(.4)
5
6 snapshot = anna_olaf.\
7 crop(x2= anna_olaf.w/2).\ # remove right half
8 to_ImageClip(0.2).\ # snapshot of the clip at t=0.2s
9 set_duration(anna_olaf.duration)
10
11 CompositeVideoClip([anna_olaf, snapshot]).\
12 to_gif('anna_olaf.gif', fps=15)

Freezing a more complicated region

This time we will apply a custom mask to the snapshot to specify where it will be transparent (and let the animated part appear) .

1 import moviepy.video.tools.drawing as dw
2
3 anna_kris = VideoFileClip("./frozen_trailer.mp4", audio=False).\
4 subclip((1,38.15),(1,38.5)).\
5 resize(.5)
6
7 # coordinates p1,p2 define the edges of the mask
8 mask = dw.color_split(anna_kris.size,
9 p1=(445, 20), p2=(345, 275),
10 grad_width=5) # blur the mask's edges
11
12 snapshot = anna_kris.to_ImageClip().\
13 set_duration(anna_kris.duration).\
14 set_mask(ImageClip(mask, ismask=True))
15
16 CompositeVideoClip([anna_kris,snapshot]).\
17 speedx(0.2).\
18 to_gif('anna_kris.gif', fps=15, fuzz=3) # fuzz= GIF compression
19

Time-symetrization

Surely you have noticed that in the previous GIFs, the end did not always look like the beginning. As a consequence, you could see a disruption every time the animation was restarted. A way to avoid this is to time-symetrize the clip, i.e. to make the clip play once forwards, then once backwards. This way the end of the clip really is the beginning of the clip. This creates a GIF that can loop fluidly, without a real beginning or end.

1 def time_symetrize(clip):
2 """ Returns the clip played forwards then backwards. In case
3 you are wondering, vfx (short for Video FX) is loaded by
4 >>> from moviepy.editor import * """
5 return concatenate([clip, clip.fx( vfx.time_mirror )])
6
7 VideoFileClip("./frozen_trailer.mp4", audio=False).\
8 subclip(36.5,36.9).\
9 resize(0.5).\
10 crop(x1=189, x2=433).\
11 fx( time_symetrize ).\
12 to_gif('sven.gif', fps=15, fuzz=2)

Ok, this might be a bad example of time symetrization,it makes the snow flakes go upwards in the second half of the animation.

Adding some text

In the next GIF there will be a text clip superimposed on the video clip.

1 olaf = VideoFileClip("./frozen_trailer.mp4", audio=False).\
2 subclip((1,21.6),(1,22.1)).\
3 resize(.5).\
4 speedx(0.5).\
6 fx( time_symetrize )
6
7 # Many options are available for the text (requires ImageMagick)
8 text = TextClip("In my nightmares\nI see rabbits.",
9 fontsize=30, color='white',
10 font='Amiri-Bold', interline=-25).\
11 set_pos((20,190)).\
12 set_duration(olaf.duration)
13
14 CompositeVideoClip( [olaf, text] ).\
15 to_gif('olaf.gif', fps=10, fuzz=2)

Making the gif loopable

The following GIF features a lot of snow falling. Therefore it cannot be made loopable using time-symetrization (or you will snow floating upwards !). So we will make this animation loopable by having the beginning of the animation appear progressively (fade in) just before the end of the clip. The montage here is a little complicated, I cannot explain it better than with this picture:

1 castle = VideoFileClip("./frozen_trailer.mp4", audio=False).\
2 subclip(22.8,23.2).\
3 speedx(0.2).\
4 resize(.4)
5
6 d = castle.duration
7 castle = castle.crossfadein(d/2)
8
9 CompositeVideoClip([castle,
10 castle.set_start(d/2),
11 castle.set_start(d)]).\
12 subclip(d/2, 3*d/2).\
13 to_gif('castle.gif', fps=5,fuzz=5)

Another example of a GIF made loopable

The next clip (from the movie Charade) was almost loopable: you can see Carry Grant smiling, then making a funny face, then coming back to normal. The problem is that at the end of the excerpt Cary is not exactly in the same position, and he is not smiling as he was at the beginning. To correct this, we take a snapshot of the first frame and we make it appear progressively at the end. This seems to do the trick.

1 carry = VideoFileClip("../videos/charade.mp4", audio=False).\
2 subclip((1,51,18.3),(1,51,20.6)).\
3 crop(x1=102, y1=2, x2=297, y2=202)
4
5 d = carry.duration
6 snapshot = carry.to_ImageClip().\
7 set_duration(d/6).\
8 crossfadein(d/6).\
9 set_start(5*d/6)
10
11 CompositeVideoClip([carry, snapshot]).\
12 to_gif('carry.gif', fps=carry.fps, fuzz=3)

Big finish: removing the background

Let us dive further into the scripting madness: we consider this video around 2’16:
http://www.youtube.com/watch?v=_wZjCDXDNLI
And we will remove the background to make this gif (with transparent background):

The main difficulty was to find what the background of the scene is. To do so, the script gathers a few images in which the little pigs are are different positions (so that every part part of the background is visible on at least several (actually most) of the slides, then it takes the pixel-per-pixel median of these pictures, which gives the background.

1 # Requires Scikit Images installed
2 import numpy as np
3 import skimage.morphology as skm
4 import skimage.filter as skf
5
6 from moviepy.editor import *
7
8 ### LOAD THE CLIP
9
10 pigsPolka = VideoFileClip("pigs_in_a_polka.mp4").\
11 subclip((2,16.85),(2,35)).\
12 resize(.5).\
13 crop(x1=140, y1=41, x2=454, y2=314)
14
15
16 ### COMPUTE THE BACKGROUND
17 # There is no single frame showing the background only (there
18 # is always a little pig in the screen) so we use the median of
19 # several carefully chosen frames to reconstitute the background.
20 # I must have spent half an hour to find the right set of frames.
21
22 times = (list(np.linspace(2.3,4.2,30))+
23 list(np.linspace(6.0,7.1,30))+
24 8*[6.2])
25
26 frames_bg = [pigsPolka.get_frame(t) for t in times]
27 background = np.percentile(np.array(frames_bg), 50,axis=0)
28
29
30 ### MASK GENERATION
31
32 def get_mask_frame(t):
33 """ Computes the mask for the frame at time t """
34
35 # THRESHOLD THE PIXEL-TO-PIXEL DIFFERENCE
36 # BETWEEN THE FRAME AND THE BACKGROUND
37 im = pigsPolka.get_frame(t)
38 mask = ((im-background)**2).sum(axis=2) > 1500
39
40 # REMOVE SMALL OBJECTS
41 mask = skm.remove_small_objects(mask)
42
43 # REMOVE SMALL HOLES (BY DILATIATION/EROSION)
44 selem=np.array([[1,1,1],[1,1,1],[1,1,1]])
45 for i in range(2):
46 mask = skm.binary_dilation(mask,selem)
47 for i in range(2):
48 mask = skm.binary_erosion(mask,selem)
49
50 # BLUR THE MASK A LITTLE
51 mask = skf.gaussian_filter(mask.astype(float),1.5)
52
53 return mask
54
55 mask = VideoClip(ismask=True).\
56 set_get_frame(get_mask_frame).\
57 set_duration(pigsPolka.duration)
58
59 ### LAST EFFECTS AND GIF GENERATION
60
61 pigsPolka.set_mask(mask).\
62 subclip(12.95,15.9).\
63 fx(vfx.blackwhite).\ # black & white effect !
64 to_gif('pigs_polka.gif', fps=10,
65 dispose=True, fuzz=10)

Source

[jackieo]

gifs are fascinating to me...is this all your work Turk? or are you just posting this ?

[Turk]

gifs are fascinating to me...is this all your work Turk? or are you just posting this ?

Jackieo, I am posting, please see the author at the top of the post and at the bottom of the source page.

Edited January 26, 2014 by Turk

[jackieo]

Just so I know who to direct questuons to, if I have one...thanks :)

[iih1]

interesting topic to be studied

[iih1]

@Turk

i'm really going to learn more about this could you tell me more how to start python

in brief notes, i have visit the link as given but feel bit confused about python.

Thx.

Edited March 21, 2014 by iih1