ffmpy3

ffmpy3 is a Python wrapper for FFmpeg, originally forked from the ffmpy project. It compiles FFmpeg command line from provided arguments and their respective options and executes it using Python’s subprocess .

ffmpy3 resembles the command line approach FFmpeg uses. It can read from an arbitrary number of input “files” (regular files, pipes, network streams, grabbing devices, etc.) and write into arbitrary number of output “files”. See FFmpeg documentation for further details about how FFmpeg command line options and arguments work.

ffmpy3 supports FFmpeg’s pipe protocol. This means that it is possible to pass input data to stdin and get output data from stdout.

At this moment ffmpy3 has wrappers for ffmpeg and ffprobe commands, but it should be possible to run other FFmpeg tools with it (e.g. ffserver).

Installation

pip install ffmpy3

Quickstart

>>> import ffmpy3
>>> ff = ffmpy3.FFmpeg(
...     inputs={'input.mp4': None},
...     outputs={'output.avi': None}
... )
>>> ff.run()

This takes input.mp4 file in the current directory as the input, changes the video container from MP4 to AVI without changing any other video parameters and creates a new output file output.avi in the current directory.

Documentation

ffmpy3

exception ffmpy3.FFExecutableNotFoundError

Raised when FFmpeg/FFprobe executable was not found.

exception ffmpy3.FFRuntimeError(cmd, exit_code, stdout=b'', stderr=b'')

Raised when FFmpeg/FFprobe command line execution returns a non-zero exit code.

cmd

The command used to launch the executable, with all command line options.

Type:str

exit_code

The resulting exit code from the executable.

Type:int

stdout

The contents of stdout (only if executed synchronously).

Type:bytes

stderr

The contents of stderr (only if executed synchronously).

Type:bytes

class ffmpy3.FFmpeg(executable='ffmpeg', global_options=None, inputs=None, outputs=None)

Wrapper for various FFmpeg related applications (ffmpeg, ffprobe).

Compiles FFmpeg command line from passed arguments (executable path, options, inputs and outputs).

inputs and outputs are dictionaries containing inputs/outputs as keys and their respective options as values.

One dictionary value (set of options) must be either a single space separated string, or a list or strings without spaces (i.e. each part of the option is a separate item of the list, the result of calling split() on the options string).

If the value is a list, it cannot be mixed, i.e. cannot contain items with spaces. An exception are complex FFmpeg command lines that contain quotes: the quoted part must be one string, even if it contains spaces (see Examples for more info).

Parameters:

• executable (str) – path to ffmpeg executable; by default the ffmpeg command will be searched for in the PATH, but can be overridden with an absolute path to ffmpeg executable
• global_options (iterable) – global options passed to ffmpeg executable (e.g. -y, -v etc.); can be specified either as a list/tuple/set of strings, or one space-separated string; by default no global options are passed
• inputs (dict) – a dictionary specifying one or more input arguments as keys with their corresponding options (either as a list of strings or a single space separated string) as values
• outputs (dict) – a dictionary specifying one or more output arguments as keys with their corresponding options (either as a list of strings or a single space separated string) as values

run(input_data=None, stdout=None, stderr=None)

Execute FFmpeg command line.

input_data can contain input for FFmpeg in case pipe protocol is used for input.

stdout and stderr specify where to redirect the stdout and stderr of the process. By default no redirection is done, which means all output goes to running shell (this mode should normally only be used for debugging purposes).

If FFmpeg pipe protocol is used for output, stdout must be redirected to a pipe by passing subprocess.PIPE as stdout argument.

Returns a 2-tuple containing stdout and stderr of the process. If there was no redirection or if the output was redirected to e.g. os.devnull, the value returned will be a tuple of two None values, otherwise it will contain the actual stdout and stderr data returned by ffmpeg process.

Parameters:

• input_data (bytes) – input data for FFmpeg to deal with (audio, video etc.) as bytes (e.g. the result of reading a file in binary mode)
• stdout – Where to redirect FFmpeg stdout to. Default is None, meaning no redirection.
• stderr – Where to redirect FFmpeg stderr to. Default is None, meaning no redirection.

Raises:

• FFExecutableNotFoundError – The executable path passed was not valid.
• FFRuntimeError – The process exited with an error.

Returns:

A 2-tuple containing stdout and stderr from the process.

Return type:

tuple

run_async(input_data=None, stdout=None, stderr=None)

Asynchronously execute FFmpeg command line.

input_data can contain input for FFmpeg in case pipe

stdout and stderr specify where to redirect the stdout and stderr of the process. By default no redirection is done, which means all output goes to running shell (this mode should normally only be used for debugging purposes).

If FFmpeg pipe protocol is used for output, stdout must be redirected to a pipe by passing subprocess.PIPE as stdout argument.

Note that the parent process is responsible for reading any output from stdout/stderr. This should be done even if the output will not be used since the process may otherwise deadlock. This can be done by awaiting on asyncio.subprocess.Process.communicate() on the returned asyncio.subprocess.Process or by manually reading from the streams as necessary.

Returns a reference to the child process created for use by the parent program.

Parameters:

• input_data (bytes) – input data for FFmpeg to deal with (audio, video etc.) as bytes (e.g. the result of reading a file in binary mode)
• stdout – Where to redirect FFmpeg stdout to. Default is None, meaning no redirection.
• stderr – Where to redirect FFmpeg stderr to. Default is None, meaning no redirection.

Raises:

FFExecutableNotFoundError – The executable path passed was not valid.

Returns:

The child process created.

Return type:

asyncio.subprocess.Process

wait()

Asynchronously wait for the process to complete execution.

Raises:FFRuntimeError – The process exited with an error. Returns:0 if the process finished successfully, or None if it has not been started Return type:int or None

class ffmpy3.FFprobe(executable='ffprobe', global_options='', inputs=None)

Wrapper for ffprobe.

Compiles FFprobe command line from passed arguments (executable path, options, inputs). FFprobe executable by default is taken from PATH but can be overridden with an absolute path.

Parameters:

• executable (str) – absolute path to ffprobe executable
• global_options (iterable) – global options passed to ffprobe executable; can be specified either as a list/tuple of strings or a space-separated string
• inputs (dict) – a dictionary specifying one or more inputs as keys with their corresponding options as values

Examples

Format convertion

The simplest example of usage is converting media from one format to another (in this case from MPEG transport stream to MP4) preserving all other attributes:

>>> from ffmpy3 import FFmpeg
... ff = FFmpeg(
...     inputs={'input.ts': None},
...     outputs={'output.mp4': None}
... )
>>> ff.cmd
'ffmpeg -i input.ts output.mp4'
>>> ff.run()

Transcoding

If at the same time we wanted to re-encode video and audio using different codecs we’d have to specify additional output options:

>>> ff = FFmpeg(
...     inputs={'input.ts': None},
...     outputs={'output.mp4': '-c:a mp2 -c:v mpeg2video'}
... )
>>> ff.cmd
'ffmpeg -i input.ts -c:a mp2 -c:v mpeg2video output.mp4'
>>> ff.run()

Demultiplexing

A more complex usage example would be demultiplexing an MPEG transport stream into separate elementary (audio and video) streams and save them in MP4 containers preserving the codecs (note how a list is used for options here):

>>> ff = FFmpeg(
...     inputs={'input.ts': None},
...     outputs={
...         'video.mp4': ['-map', '0:0', '-c:a', 'copy', '-f', 'mp4'],
...         'audio.mp4': ['-map', '0:1', '-c:a', 'copy', '-f', 'mp4']
...     }
... )
>>> ff.cmd
'ffmpeg -i input.ts -map 0:1 -c:a copy -f mp4 audio.mp4 -map 0:0 -c:a copy -f mp4 video.mp4'
>>> ff.run()

Warning

Note that it is not possible to mix the expression formats for options, i.e. it is not possible to have a list that contains strings with spaces (an exception to this is Complex command lines). For example, this command line will not work with FFmpeg:

>>> from subprocess import PIPE
>>> ff = FFmpeg(
...     inputs={'input.ts': None},
...     outputs={
...         'video.mp4': ['-map 0:0', '-c:a copy', '-f mp4'],
...         'audio.mp4': ['-map 0:1', '-c:a copy', '-f mp4']
...     }
... )
>>> ff.cmd
'ffmpeg -hide_banner -i input.ts "-map 0:1" "-c:a copy" "-f mp4" audio.mp4 "-map 0:0" "-c:a copy" "-f mp4" video.mp4'
>>>
>>> ff.run(stderr=PIPE)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Users/ay/projects/personal/ffmpy3/ffmpy3.py", line 104, in run
    raise FFRuntimeError(self.cmd, ff_command.returncode, out[0], out[1])
ffmpy3.FFRuntimeError: `ffmpeg -hide_banner -i input.ts "-map 0:1" "-c:a copy" "-f mp4" audio.mp4 "-map 0:0" "-c:a copy" "-f mp4" video.mp4` exited with status 1

STDOUT:


STDERR:
Unrecognized option 'map 0:1'.
Error splitting the argument list: Option not found

>>>

Notice how the actual FFmpeg command line contains unnecessary quotes.

Multiplexing

To multiplex video and audio back into an MPEG transport stream with re-encoding:

>>> ff = FFmpeg(
...     inputs={'video.mp4': None, 'audio.mp3': None},
...     outputs={'output.ts': '-c:v h264 -c:a ac3'}
... )
>>> ff.cmd
'ffmpeg -i audio.mp4 -i video.mp4 -c:v h264 -c:a ac3 output.ts'
>>> ff.run()

There are cases where the order of inputs and outputs must be preserved (e.g. when using FFmpeg -map option). In these cases the use of regular Python dictionary will not work because it does not preserve order. Instead, use OrderedDict. For example we want to multiplex one video and two audio streams into an MPEG transport streams re-encoding both audio streams using different codecs. Here we use an OrderedDict to preserve the order of inputs so they match the order of streams in output options:

>>> from collections import OrderedDict
>>> inputs = OrderedDict([('video.mp4', None), ('audio_1.mp3', None), ('audio_2.mp3', None)])
>>> outputs = {'output.ts', '-map 0 -c:v h264 -map 1 -c:a:0 ac3 -map 2 -c:a:1 mp2'}
>>> ff = FFmpeg(inputs=inputs, outputs=outputs)
>>> ff.cmd
'ffmpeg -i video.mp4 -i audio_1.mp3 -i audio_2.mp3 -map 0 -c:v h264 -map 1 -c:a:0 ac3 -map 2 -c:a:1 mp2 output.ts'
>>> ff.run()

Using pipe protocol

ffmpy3 can read input from STDIN and write output to STDOUT. This can be achieved by using FFmpeg pipe protocol. The following example reads data from a file containing raw video frames in RGB format and passes it to ffmpy3 on STDIN; ffmpy3 in its turn will encode raw frame data with H.264 and pack it in an MP4 container passing the output to STDOUT (note that you must redirect STDOUT of the process to a pipe by using subprocess.PIPE as stdout value, otherwise the output will get lost):

>>> import subprocess
>>> ff = FFmpeg(
...     inputs={'pipe:0': '-f rawvideo -pix_fmt rgb24 -s:v 640x480'},
...     outputs={'pipe:1': '-c:v h264 -f mp4'}
... )
>>> ff.cmd
'ffmpeg -f rawvideo -pix_fmt rgb24 -s:v 640x480 -i pipe:0 -c:v h264 -f mp4 pipe:1'
>>> stdout, stderr = ff.run(input_data=open('rawvideo', 'rb').read(), stdout=subprocess.PIPE)

Asynchronous execution

In certain cases, one may not wish to run FFmpeg and block on waiting for results or introduce multithreading into one’s application. In this case, asynchronous execution using asyncio is possible.

>>> ff = ffmpy3.FFmpeg(
...     inputs={'input.mp4': None},
...     outputs={'output.avi': None}
... )
>>> ff.run_async()
>>> await ff.wait()

Processing FFmpeg output without multithreading or blocking is also possible. The following code snippet replaces CR with LF from FFmpeg’s progress output and echoes it to STDERR while FFmpeg processes the input video.

>>> import asyncio
>>> import sys
>>> ff = ffmpy3.FFmpeg(
...     inputs={'input.mp4': None},
...     outputs={'output.avi': None},
... )
>>> _ffmpeg_process = await ff.run_async(stderr=asyncio.subprocess.PIPE)
>>> line_buf = bytearray()
>>> my_stderr =_ffmpeg_process.stderr
>>> while True:
>>>     in_buf = (await my_stderr.read(128)).replace(b'\r', b'\n')
>>>     if not in_buf:
>>>         break
>>>     line_buf.extend(in_buf)
>>>     while b'\n' in line_buf:
>>>         line, _, line_buf = line_buf.partition(b'\n')
>>>         print(str(line), file=sys.stderr)
>>> await ff.wait()

Complex command lines

FFmpeg command line options can get pretty complex, like when using filtering. Therefore, it is important to understand some of the rules for building command lines building with ffmpy3. If an option contains quotes, it must be specified as a separate item in the options list without the quotes. However, if a single string is used for options, the quotes of the quoted option must be preserved in the string:

>>> ff = FFmpeg(
...     inputs={'input.ts': None},
...     outputs={'output.ts': ['-vf', 'adif=0:-1:0, scale=iw/2:-1']}
... )
>>> ff.cmd
'ffmpeg -i input.ts -vf "adif=0:-1:0, scale=iw/2:-1" output.ts'
>>>
>>> ff = FFmpeg(
...     inputs={'input.ts': None},
...     outputs={'output.ts': '-vf "adif=0:-1:0, scale=iw/2:-1"'}
... )
>>> ff.cmd
'ffmpeg -i input.ts -vf "adif=0:-1:0, scale=iw/2:-1" output.ts'

An even more complex example is a command line that burns the timecode into video:

ffmpeg -i input.ts -vf "drawtext=fontfile=/Library/Fonts/Verdana.ttf: timecode='09\:57\:00\:00': r=25: x=(w-tw)/2: y=h-(2*lh): fontcolor=white: box=1: boxcolor=0x00000000@1" -an output.ts

In ffmpy3 it can be expressed in the following way:

>>> ff = FFmpeg(
...     inputs={'input.ts': None},
...     outputs={'output.ts': ['-vf', "drawtext=fontfile=/Library/Fonts/Verdana.ttf: timecode='09\:57\:00\:00': r=25: x=(w-tw)/2: y=h-(2*lh): fontcolor=white: box=1: boxcolor=0x00000000@1", '-an']}
... )
>>> ff.cmd
'ffmpeg -i input.ts -vf "drawtext=fontfile=/Library/Fonts/Verdana.ttf: timecode=\'09\:57\:00\:00\': r=25: x=(w-tw)/2: y=h-(2*lh): fontcolor=white: box=1: boxcolor=0x00000000@1" -an output.ts'

The same command line can be compiled by passing output options as a single string, while keeping the quotes:

>>> ff = FFmpeg(
...     inputs={'input.ts': None},
...     outputs={'output.ts': ["-vf \"drawtext=fontfile=/Library/Fonts/Verdana.ttf: timecode='09\:57\:00\:00': r=25: x=(w-tw)/2: y=h-(2*lh): fontcolor=white: box=1: boxcolor=0x00000000@1\" -an"}
... )
>>> ff.cmd
'ffmpeg -i input.ts -vf "drawtext=fontfile=/Library/Fonts/Verdana.ttf: timecode=\'09\:57\:00\:00\': r=25: x=(w-tw)/2: y=h-(2*lh): fontcolor=white: box=1: boxcolor=0x00000000@1" -an output.ts'