A ffmpeg workflow manager for transcoding

## pytranscoder

Python wrapper for ffmpeg for batch, concurrent, or clustered transcoding using defined profiles and optional rules engine for automation.

#### News

Pytranscoder now supports better profiles. Existing ones are backward-compatible unless you try to use mixings. The mixin feature allows you to create pieces of profiles and combine them into another profile on the commandline at runtime. This negates the need to create many unique permutations of a profile just for simple variances.

#### What it is

This script is intended to help automate transcoding for people encoding lots of video. It is more than a wrapper - it is a workflow and job manager.

There are 2 modes: local and clustered. Local mode is the most common usage and is for running this script on the same machine where it is installed. Cluster mode turns pytranscoder into a remote encoding manager. In this mode it delegates and manages encode jobs running on multiple hosts. This requires more advanced configuration and is documented separately in Cluster.md

The remainder of this document focuses on using pytranscoder in local mode.

#### Features:

• Sequential or concurrent transcoding.
• Concurrent mode allows you to make maximum use of your nVidia CUDA-enabled graphics card or Intel accelerated video (QSV)
• Preserves all streams but allows for filtering by audio and subtitle language.
• Configurable transcoding profiles
• Configurable rules and criteria to auto-match a video file to a transcoding profile
• Mixins are profile fragments you can provide at runtime to dynmically change your selected profiles.
• Transcode from a list of files (queue) or all on the command line
• Cluster mode allows use of other machines See Cluster.md for details.
• On-the-fly compression monitoring and optional early job termination if not compressing as expected.
• Experimental support for HandBrakeCLI

#### Requirements

• Linux or MacOS, Windows 10. For Windows, WSL (Ubuntu) recommended.
• latest ffmpeg (3.4.3-2 or higher, lower versions may still work)
• nVidia graphics card with latest nVidia CUDA drivers (optional)
• Intel CPU with QSV enabled (optional)
• Python 3 (3.7 or higher)

#### Support

Video Tutorials: Part 1 - Linux Setup, Part 2 - Usage

#### Installation

There are a few possible ways to install a python app - one of these should work for you.

Linux The confusion is due to the fact that not all distributions or OS's install pip3 by default. Either way, pytranscoder is available in the pypi repo.

pip3 install --user pytranscoder-ffmpeg
# or...
python3 -m pip install --user pytranscoder-ffmpeg


Windows (WSL - Ubuntu) Windows Subsystem for Linux is the best option, but requires a couple of maintenance steps first if you don't have pip3:

sudo apt update
sudo install python3-pip

# now we can install
pip3 install --user pytranscoder-ffmpeg


At this point you have a choice - either install ffmpeg for Windows ffmpeg.exe or install in bash as an Ubuntu package. Either will work but there are caveats, or you could install both and not worry.

• ffmpeg.exe can be run in Windows command shell or from bash but requires special attention when configuring pytranscoder paths.
• ffmpeg apt package can only be run from bash but is a more natural Linux path mapping.

After installing you will find this document and others in $HOME/.local/shared/doc/pytranscoder (on Linux/MacOS) and in$HOME/AppData/Python/pythonversion/shared/doc/pytranscoder** (on Windows). Also available online

Whatever method above for installing works for you, just use the --upgrade option to update, ie: pip3 install --upgrade pytranscoder-ffmpeg

#### Operation - Profiles and Rules

A profile is a named group of ffmpeg commandline options to encode a specific way. You can define all the combinations you use regularly in transcode.yml for easy selection later. At least 1 profile definition is required.

A rule is a YAML syntax of predefined predicates to allow simple matching on source media details and relate to a specific profile. For example, if you transcode 720p differently than 1080p, and still different than 4k you can set up rules to match those 3 resolutions to a specific transcode profile. Easy - let the script do the work of selecting the right ffmpeg options.

But you aren't required to use rules. You can specify the profile name on the commandline each run using the -p option. Or you can define 1 rule that acts as a default (see examples). It's up to you. But using rules is a great way to automate a tedious manual workflow.

When changing or adding profiles and rules it is useful to test them out by running in --dry-run mode first, which will show you everything that would happen if running for real.

#### Configuration

There is a supplied sample transcode.yml config file, or you can download it here. This can be customized all you like, however be sure to preserve the YAML formatting. Either specify this file on the commandline with the -y option or copy it to your home directory as .transcode.yml (default)

There are 3 sections in the file:

#### config - Global configuration information

Sample

config:
default_queue_file:   '/path/to/default/list/of/files/if/none/given'
ffmpeg:               '/usr/bin/ffmpeg'       # path to ffmpeg for this config
ssh:                '/usr/bin/ssh'    # used only in cluster mode (optional)
queues:
qsv:                1                   # sequential encodes
cuda:               2                   # maximum of 2 encodes at a time
colorize:             yes

setting purpose
default_queue_file A queue file is just a text file listing out all the media you want to encode. It is not required, but useful when automating a workflow. You can always indicate a queue file on the command line. This just sets the default, if any.
ffmpeg Full path to ffmpeg on this host
ssh Full path to ssh on this host
queues If using concurrency, define your queues here. The queue name is whatever you want. Each name specifies a maximum number of concurrent encoding jobs. If none defined, a default sequential queue is used.
colorize optional, defaults to "no". If "yes" terminal output will have some color added

#### profiles - Transcoding profiles (ffmpeg options)

Profiles are used to provide ffmpeg with various options for encoding. One profile definition is required, but mostly likely you will define multiples for different encoding scenarios. The name of the profile can be provided on the command line to select the appropriate one for your needs. Alternatively, you can define rules (see below) to auto-match media with profiles for a less manual encoding workflow.

Sample:

profiles:

# some common, reusable settings to keep things tidy
common:
output_options_subtitles:
- "-c:s copy"
- "-f matroska"
output_options_audio:
- "-c:a copy"
output_options_video:
- "-crf 20"
extension: '.mkv'
threshold: 20
threshold_check: 30

#
# Sample Intel QSV transcode setup (note to customize options for your environment)
#
hevc_qsv:
include: common
output_options_video:         # mixin-enabled section - overrides common
- "-c:v hevc_qsv"
- "-preset medium"
- "-qp 21"
- "-b:v 7M"

#
# Sample nVidia transcode setup
#

hevc_cuda:                  # profile name
include: common
input_options:          # ffmpeg input options
- "-hwaccel cuvid"    # REQUIRED for CUDA
- "-c:v h264_cuvid"   # hardware decoding too
output_options_video:   # mixen-enabled - overrides common
- "-c:v hevc_nvenc"   # REQUIRED for CUDA
- "-profile:v main"
- "-preset medium"
output_options_audio:
- "-c:a copy"

queue: cuda		# manage this encode in the 'cuda' queue defined globally

# optionally you can filter out audio/subtitle tracks you don't need
audio:
exclude_languages:
- "chi"
- "spa"
- "fre"
- "ger"
default_language: eng

subtitle:
exclude_languages:
- "chi"
- "spa"
- "fre"
- "por"
- "ger"
- "jpn"
default_language: eng
#
# This is a mixin, a profile fragment that can be provided on the command line
# to alter a profile (ie. pytranscoder -p hevc_cuda -m mp3_hq file.mp4)

mp3_hq:
output_options_audio:
- "-c:a mp3lame"
- "-b:a 384k"

x264:                            # profile name
include: common
input_options:
output_options:
- "-c:v x264"

h264_cuda_anime:
include: common
input_options:
output_options:
- "-c:v h264_nvenc"
- "-tune animation"

setting purpose
input_options ffmpeg options related to the input (see ffmpeg docs)
output_options ffmpeg options related to the output (see ffmpeg docs). Use for generic options, not mixin-enabled
output_options_video ffmpeg options specific to video (see ffmpeg docs). This section is mixin-enabled
output_options_audio ffmpeg options specific to audio (see ffmpeg docs). This section is mixin-enabled
extension Filename extension to use for the encoded file
queue optional. Assign encodes for this profile to a specific queue (defined in config section)
threshold optional. If provided this number represents a minimum percentage compression savings for the encoded media. If it does not meet this threshold the transcoded file is discarded, source file remains as-is, and the source job marked as complete. This is useful if a particular file doesn't compress much and you would rather just keep the original.
threshold_check optional. If provided this is the percent done to start checking if the threshold is being met. Default is 100% (when media is finished). Use this to have threshold checks done earlier to stop a long-running transcode if not producing expected compression (threshold).
include optional. Include options from one or more previously defined profiles. (see section on includes).
audio Audio track handling options. Include a list of exclude_languages to automatically remove tracks. If any track being removed is a default, a new default will be set based on the default_language.
subtitle See audio above.

CPU Note: When transcoding from h264 on an Intel I5/I7 6th+ gen chip, ffmpeg will use detected extensions to basically perform hardware decoding for you. So if you configured hardware encoding you'll see low CPU use. On AMD there is no chip assistance on decoding. So even if hardware encoding, the decoding process will load down your CPU. To fix this simply enable hardware decoding as an input option.

#### rules - simple profile matching rules

Simple expressions to match video files with the appropriate profile. They are evaluated top-down so make sure your default is the last one. You don't need to use the rules system. You can either explicitly give the desired profile name on the commandline or just have a single rule for default. But if you encode certain media differently then having the rules system make it a little easier using various options depending on the media attributes. No specific criteria is required - use the ones applicable to your rule.

Rule evaluation is as follows: for each input media file, compare against each rule criteria. All criteria of a rule must match in order for the given profile to be selected. If any one fails, evaluation continues to the next rule. If there are no matches, the default rule is selected.

Samples:

  'for content I consider too big':  # comment and unique identifier for this rule
profile: hevc_hd_25fps    # profile to use if the criterial below match
criteria:
runtime:      '<180'    # less than 3 hours long
filesize_mb:  '>5000'   # ..and media file larger than 5 gigabytes
fps: '>25'              # ..and framerate > 25

profile: 'SKIP'     # special keyword SKIP, means anything that matches this rule won't get transcoded
criteria:
'vcodec': 'hevc'	# if media video is encoded with hevc already

'skip files that are not appropriate for hevc':
profile: 'SKIP'
criteria:
filesize_mb: '<600'       # video file is less than 600mb
runtime: '<40'          	# ..and total runtime < 40 minutes

'anime to h264 using tuning':
profile: h264_cuda_anime
criteria:
filesize_mb: '>2500'   # larger than 2.5g
vcodec: '!hevc'            # not encoded with hevc
path: '/media/anime/.*'  # in a anime folder (regex)

'half-hour videos':
profile: 'x264'             # use profile called "x264"
criteria:
filesize_mb: '>500'       # 400mb file size or greater
runtime: '<31'        	# 30 minutes or less runtime
vcodec: '!hevc'	       	# NOT hevc encoded video

'small enough already':       # skip if <2.5g size, between 720p and 1080p, and between 30 and 64 minutes long.
profile: SKIP             # transcoding these will probably cause a noticeable quality loss so skip.
criteria:
filesize_mb: '<2500'    # less than 2.5 gigabytes
res_height: '720-1081'  # 1080p, allowing for random oddball resolutions still in the HD range
runtime:  '35-65'       # between 35 and 65 minutes long

'default':                       # this will be the DEFAULT (no criteria implies a match)
profile: hevc_hd_preserved   # use profile named "hevc_hd_preserved"

setting purpose
profile The defined profile name (from above) to select if this rule criteria matches. If the profile name is SKIP then matched media will not be transcoded
runtime Total run time of media, in minutes. Determined by ffmpeg. Optionally can use < or > or a range
source_size Size, in megabytes, of the media file. Optionally an use < or > or a range
fps Frames per second. Determined by ffmpeg. Optionally can use < or > or a range
vcodec Video codec used on the source media. Determined by ffmpeg. Can use ! to indicate not condition (negative match)
res_height Video vertical resolution. Determined by ffmpeg. Optionally can use < or > or a range
res_width Video horizontal resolution. Determined by ffmpeg. Optionally can use < or > or a range

So, for example, using the sample rule 'for content I consider too big', if the video is less than 180 minutes long and the file size is larger than 5 gigabytes and frames-per-second is greater than 25 then use the hevc_hd_25fps profile to encode.

For those settings that allow operators, put the operator first (< or >) followed by the number. For those that allow a range provide the lower and upper range with a hyphen (-) between. No spaces are allowed in criteria.

#### Sonarr-aware

You can invoke pytranscoder from a Sonarr custom script connection to handle recording of downloads and upgrades to your queue file. The filename passed from Sonarr will be appended to your default_queue_file (see global configuration above). Media is not transcoded at this time, only recorded for future processing. Simply having Sonarr call pytranscoder is all you need to configure - pytranscoder will detect it was invoked from Sonarr and act accordingly. No parameters are required.

#### Process Flow

High-level steps the script takes to process your media.

• Determine list of input files to transcode
• If a profile is given (-p) make that the starting default
• If a list file is given, read list of media files from that file.
• If media files are given on the command line, add those to the list, observing any -p profile overrides along the way.
• If running concurrent, interactive transcoding stats and screen logging for ffmpeg will be disabled.
• If running in --dry-run mode:
• For each media file print "what-if" transcoding details
• Exit script execution
• For each media file do the transcoding:
• If file has no given profile assignment, use the rules system to find a match. If no match, skip
• When file has finished transcoding:
• If the selected profile has a threshold value, compare original and transcoded file size.
• If threshold met:
• If -k not given, remove original and replace with newly transcoded .tmp file.
• If -k given, keep the original and leave the transcoded .tmp version in place for inspection.
• If threshold not met, inform user and remove .tmp file leaving original intact.
• If -k not given, remove original and replace with newly transcoded .tmp file.
• If a list file (queue) was used, the completed media file will be removed from that list.
• Exit script execution

#### Running without Concurrency

If you cannot transcode concurrently, or just don't want to you can still get value from this script. Just avoid defining any queues, or define them with a value of 1 (see global config section).

#### Notes on Concurrency

Concurrency here means running multiple encoding jobs at the same time by taking advantage of hardware support. Normally an encode run will almost max out a CPU until it is finished. But with hardware-assisted encoding very little of the CPU is used and most work is offloaded to the hardware. This allows the CPU to handle multiple files and still have processing power left over for regular system activities.

In the global profile section of transcode.yml you define your queues. Use whatever names you like, and provide a maximum number of concurrent jobs for that queue. Use 1 for synchronous or a higher value to run multiples (but not more than your hardware can support - use trial and error to figure out, but 2 is a good number to stick with).

You can optionally designate a specific queue for each of your profiles. If none defined, the default queue is used, which is sequential. What queue assigned allows is finer management of concurrent transcoding jobs. Queues are not necessary if you only plan to transcode sequentially.

If you want maximum use of your machine consider this scenario:

You have an 8th generation Intel i5 6-core machine with nVidia graphics card that can handle 2 concurrent encodes. You define 2 queues:

     queues:
one: 1
two: 2


You have 2 profiles: "qsv" (configured to use Intel QSV) and "cuda" (configured for nVidia CUDA). You associate the "qsv" profile with queue one, and "cuda" with queue two. You start a job like:

pytranscoder -p qsv /media/tv/new/*.mp4 -p cuda /media/movies/new/*.mp4


This example will run 3 concurrent jobs - 1 running on the CPU using QSV and 2 running on the nVidia card! This is how a multi queue configuration can be used. But I hope you have good system fans.

You must also supply the appropriate options for your transcode profiles to use the supported hardware, otherwise you'll completely bog down your system (see the transcode.yml "hevc" and "qsv" samples). If you transcode with a profile not setup for hardware support, or the rules matcher selects a profile without hardware support, that file will transcode using CPU time. Therefore, when using concurrent hardware transcoding using rules it is best that all your rules map to only profiles with hardware support.

You can always force non-concurrent CPU-based transcodes from the command line, selecting sequential-only (-s) and bypassing profile rules.

You should always do a dry-run test before committing to a configuration change. This will show you how your media will be handled but won't actually do any work or change anything. It will help you see that your defined rules are matching as expected and that hosts can be connected to via ssh.

    pytranscoder --dry-run /volume1/media/any_video_file


#### Running

Note that if using a list file (queue) as input, when the process is done that file will contain only those video files that failed to encode, or it will be removed if all files were processed. So if you need to keep this file make a copy first.

The default behavior is to remove the original video file after encoding and replace it with the new version. If you want to keep the source be sure to use the -k parameter. The work file will be placed in the same folder as the source with the same name and a .tmp extension while being encoded.

option purpose
--from-file <file> Load list of files to process from <file>
-p <profile> Specify <profile> to use. Can be used multiple times on command line and applies to all subsequent files (see examples)
-y <config> Specify non-default transcode.yml file.
-s Force sequential mode (no concurrency event for concurrent queues)
-k Keep original media files, leave encoded .tmp file in same folder.
--dry-run Show what will happen without actually doing any work
-c <name> Cluster mode. See Cluster.md for details
-m name{,name...} Add named mixin(s) to the given profile (-p)
##### Examples:

To get help and version number:

   pytranscoder -h


To transcode 2 files using a specific profile:

    pytranscoder -p my_fave_x264 /tmp/video1.mp4 /tmp/video2.mp4


To use a profile but alter the audio track using a mixin:

    pytranscoder -p hevc_cuda -m mp3_hq videofile.mp4


To auto transcode a file using the rules system but keep the original:

    pytranscoder -k testvid.mp4


To transcode 2 files using different profiles:

    pytranscoder -p my_fave_x264 /tmp/video1.mp4 -p cuda_hevc  /tmp/video2.mp4


To auto transcode everything in a specific queue file, defaulting to rules to match profiles:

    pytranscoder --from-file /tmp/queue.txt


To do a test run without transcoding, to see which profiles will match and the ffmpeg commandline:

    pytranscoder --dry-run atestvideo.mp4


To use a specific transcode.yml file and auto transcode using rules:

    pytranscoder -y /home/me/etc/transcode.yml *.mp4


To transcode everything in a queue file, using a forced profile for all:

    pytranscoder -p cuda_hevc --from-file /tmp/queue.txt


Complex example to show off flexibiliity. Using custom config file test.yml, keep original media, transcode all mp4 files in /media1 using rules,, transcode all files in /media2 using hevc_cuda profile, and transcode all files listed in listoffiles.txt using qsv profile:

    pytranscoder -y /home/me/test.yml -k /media1/*.mp4 -p hevc_cuda /media2/* -p qsv --from-file /tmp/listofiles.txt


If configured for concurrency but want to auto transcode a bunch of files sequentially only:

    pytranscoder -s *.mp4


To run in cluster mode (see Cluster documentation):

    pytranscoder -c *.mp4


#### Using includes

This feature requires a deeper familiarity with the YAML format. Essentially, you can define a partial profile or a full one and later "include" it into another profile. This facilitates reuse of definitions and simpler profiles.

#
# Merge-style example
#
profiles:
# values universal to all my high-quality transcodes
hq:
output_options:  # options using hyphens and separate lines are "lists"
- "-crf 18"
- "-preset slow"
- "-c:a copy"
- "-c:s copy"
- "-f matroska"
threshold: 20
extension: ".mkv"

hevc_cuda:
include: hq     # pull in everything defined above in "hq"
output_options: # combine these options with those from "hq"
- "-c:v hevc_nvenc"
- "-profile:v main"
threshold: 18    # replace "hq" threshold value with 18


The above example is equivalent to:

  hevc_cuda:
output_options: -crf 18 -preset slow -c:a copy -c:s copy -f matroska -c:v hevc_nvenc -profile:v main
threshold : 18
extension: ".mkv"


The advantage is that now we have a base (parent) profile we can include into many others to avoid repetative profile definitions. And, if we decide to change our base threshold, for example, we only need to change it in the base (parent).

Note that the profiles "hq" and "hevc_cuda" were combined, and the value for threshold was overridden to 18. Lets refer to the first (base) profile as the parent, and the second as the child. So a child profile can include one or more parent profiles. All values in the child are retained. However, if input_options or output_options are lists instead of strings, the parent and child values will be combined. Here is the same example slightly reformatted:

#
# Replace-style example
#
profiles:
hq:
output_options: -crf 18 -preset slow -c:a copy -c:s copy -f matroska
threshold: 20
extension: ".mkv"

hevc_cuda:
include: hq
output_options: -c:v hevc_nvenc -profile:v main
threshold: 18


This will produce a bad profile. Now I need to mention a feature of YAML only used in the include examples - lists. YAML-formatted data can be very complex but pytranscoder requirements are meager and simple. But to support the include feature in both replace and merge modes I needed another way to express input and output options. Note the difference in the Merge and Replace examples is that Merge uses hyphens and a separate line for the output_options sections. In Replace, all the options are on a single line. The former is an expression of a "list of arguments". The latter is just a "string of arguments" When a parent and child both have input_options or output_options that are lists, the two are combined. If either is not a list (just a string), then the child wins and the parent version is ignored. With this new information we can now see why the Replace example produces a bad profile. It will look like this:

    hevc_cuda:
output_options: -c:v hevc_nvenc -profile:v main
threshold: 18
extension: ".mkv"


Since output_options is a simple string rather than list, pytranscoder doesn't know how to merge them so it doesn't try. The child values always wins. So this profile will produce undesirable results because the parent options weren't merged. Convert both back to lists and it will work again.

## Project details

Uploaded source
Uploaded py3