I'm converting a video to GIF file with ffmpeg
:
ffmpeg -i input.flv -ss 00:00:00.000 -pix_fmt rgb24 -r 10 -s 320x240 -t 00:00:10.000 output.gif
It works great, but output gif file has a very low quality.
Any ideas how can I improve quality of converted gif?
Answer
ffmpeg
can output high quality GIF. Before you start it is always recommended to use a recent version: download or compile.
ffmpeg -ss 30 -t 3 -i input.mp4 -vf "fps=10,scale=320:-1:flags=lanczos,split[s0][s1];[s0]palettegen[p];[s1][p]paletteuse" -loop 0 output.gif
- This example will skip the first 30 seconds (
-ss 30
) of the input and create a 3 second output (-t 3
). - fps filter sets the frame rate. A rate of 10 frames per second is used in the example.
- scale filter will resize the output to 320 pixels wide and automatically determine the height while preserving the aspect ratio. The lanczos scaling algorithm is used in this example.
- palettegen and paletteuse filters will generate and use a custom palette generated from your input. These filters have many options, so refer to the links for a list of all available options and values. Also see the Advanced options section below.
- split filter will allow everything to be done in one command and avoids having to create a temporary PNG file of the palette.
- Control looping with
-loop
output option but the values are confusing. A value of0
is infinite looping,-1
is no looping, and1
will loop once meaning it will play twice. So a value of 10 will cause the GIF to play 11 times.
Advanced options
The palettegen and paletteuse filters have many additional options. The most important are:
stats_mode
(palettegen). You can force the filters to focus the palette on the general picture (full
which is the default), only the moving parts (diff
), or each individual frame (single
). For example, to generate a palette for each individual frame usepalettegen=stats_mode=single
&paletteuse=new=1
.dither
(paletteuse). Choose the dithering algorithm. There are three main types: deterministic (bayer
), error diffusion (all the others including the defaultsierra2_4a
), and none. Your GIF may look better using a particular dithering algorithm, or no dithering at all. If you want to trybayer
be sure to test thebayer_scale
option too.
See High quality GIF with FFmpeg for explanations, example images, and more detailed info for advanced usage.
Also see the palettegen and paletteuse documentation for all available options and values.
Another command-line method is to pipe from ffmpeg
to convert
(or magick
) from ImageMagick.
ffmpeg -i input.mp4 -vf "fps=10,scale=320:-1:flags=lanczos" -c:v pam -f image2pipe - | convert -delay 10 - -loop 0 -layers optimize output.gif
ffmpeg
options:
-vf "fps=10,scale=320:-1:flags=lanczos"
a filtergraph using the fps and scale filters. fps sets frame rate to 10, and scale sets the size to 320 pixels wide and height is automatically determined and uses a value that preserves the aspect ratio. The lanczos scaling algorithm is used in this example.-c:v pam
Chooses the pam image encoder. The example outputs the PAM (Portable AnyMap) image format which is a simple, lossless RGB format that supports transparency (alpha) and is supported byconvert
. It is faster to encode than PNG.-f image2pipe
chooses the image2pipe muxer because when outputting to a pipeffmpeg
needs to be told which muxer to use.
convert
options:
-delay
See Setting frame rate section below.-loop 0
makes infinite loop.-layers optimize
Will enable the general purpose GIF optimizer. See ImageMagick Animation Optimization for more details. It is not guaranteed that it will produce a smaller output, so it is worth trying without-layers optimize
and comparing results.
Setting frame rate
Set frame rate with a combination of the fps filter in ffmpeg
and -delay
in convert
. This can get complicated because convert
just gets a raw stream of images so no fps is preserved. Secondly, the -delay
value in convert
is in ticks (there are 100 ticks per second), not in frames per second. For example, with fps=12.5
= 100/12.5 = 8 = -delay 8
.
convert
rounds the -delay
value to a whole number, so 8.4 results in 8 and 8.5 results in 9. This effectively means that only some frame rates are supported when setting a uniform delay over all frames (a specific delay can be set per frame but that is beyond this answer).
-delay
appears to be ignored if used as an output option, so it has to be used before -
as shown in the example.
Lastly, browsers and image viewers may implement a minimum delay, so your -delay
may get ignored anyway.
Video courtesy of U.S. Fish & Wildlife Service National Conservation Training Center.
No comments:
Post a Comment