path: root/localvideowebencode

#!/bin/sh
# Copyright © 2010-2014, 2016 Jonas Smedegaard <dr@jones.dk>
# Description: Recode a video into web-optimized format(s)
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 3, or (at your option)
# any later version.
#
# This program is distributed in the hope that it will be useful, but
# WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
# General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program.  If not, see <http://www.gnu.org/licenses/>.
#
# Depends: ffmpeg melt mediainfo
# Recommends: moreutils vainfo vpx-tools vorbis-tools opus-tools
#
# TODO:
#  * improve workflow
#    + check initially and fail if all needed tools are not available
#    + offer to skip rendering again if an output file exist already
#    + maybe support extracting keyframes to aid in picking cover image
#      <https://video.stackexchange.com/questions/26066/dump-keyframes-of-a-livestream>
#      <https://video.stackexchange.com/questions/21970/speed-up-a-video-30x-using-only-key-frames-with-ffmpeg>
#    + offer to produce a full ladder at once
#      <https://developers.google.com/media/vp9/settings/vod>
#      <https://developer.apple.com/documentation/http_live_streaming/hls_authoring_specification_for_apple_devices>
#  * improve configurability
#    + support --width and --height,
#      resolving other dimension from (explicit or source) aspect ratio
#    + make choice of encoders configurable
#    + warn if inferior containers are used, e.g. non-WebM for VP8/VP9
#      <https://trac.ffmpeg.org/ticket/1171#comment:2>
#  * tune defaults
#    + adapt VP8/VP9 quantizer based on frame size
#      <https://developers.google.com/media/vp9/settings/vod#quality>
#    + maybe pass option real_time=-2 to melt
#      <https://www.reddit.com/r/kdenlive/comments/ka0aak/kdenlive_gpucpu_use_threads_mlt_and_ffmpeg_tips/>
#    + maybe reduce threads to "spend" total available cores only once
#      <https://blog.maxwellgale.com/2021/01/30/streaming-video-over-webrtc-using-ffmpeg/>
#    + apply application option when using opusenc
#    + use joint stereo for low-bitrate Opus speech
#    + maybe refine audio bandwidth algorithms
#      <http://trac.ffmpeg.org/wiki/Encode/HighQualityAudio>
#    + change VP8/VP9 bandwidth algorithm
#      <https://developers.google.com/media/vp9/settings/vod/>
#    + tune VP8 parameters:
#      <http://www.webmproject.org/docs/encoder-parameters/
#    + tune VP9 parameters
#      <https://sites.google.com/a/webmproject.org/wiki/ffmpeg/vp9-encoding-guide>
#    + maybe use High profile for H.264
#      <https://stackoverflow.com/questions/21120717/h-264-video-wont-play-on-ios#comment43963026_21121914>
#    + compute (or hardcode) html5 codec types, e.g. using mp4v2-utils
#      <http://stackoverflow.com/a/16365526>
#    + generalize GOP handling to use "Closed GOPs",
#      with default "normal" option of 200f@25fps or 240f@30fps (both 8s)
#      and "low-latency" option of 48f (1.92s at 25fps or 1.6s at 30fps)
#      <https://streaminglearningcenter.com/articles/open-and-closed-gops-all-you-need-to-know.html>
#      <https://stackoverflow.com/questions/30979714/how-to-change-keyframe-interval-in-ffmpeg/41735741#41735741>
#      <http://anton.lindstrom.io/gop-size-calculator/>
#      <https://docs.unified-streaming.com/best-practice/content-preparation.html#recommended-fragment-boundaries-are-aligned-across-all-tracks-audio-video-text>
#  * players
#    + disable flash player by default
#    + support HLS player, when HLS is supported
#      <https://github.com/video-dev/hls.js>
#  * normalize each infile separately when xml fed as infile keeps sync
#    maybe as workaround re-feed audio separately from xml, as done at
#    <http://bernaerts.dyndns.org/linux/74-ubuntu/214-ubuntu-stabilize-video-melt>
#  * support DASH
#    <https://sites.google.com/a/webmproject.org/wiki/adaptive-streaming/instructions-to-playback-adaptive-webm-using-dash>
#  * support watermark
#  * support superstable live streaming, e.g. (maybe faked) endless source
#    + lipsync may require single stream of equally chunked tracks
#      <https://sites.google.com/a/webmproject.org/wiki/adaptive-streaming/instructions-to-do-webm-live-streaming-via-dash>
#  * support adaptive streaming
#    + compute ladder from mathematical model
#      as in "Optimal design of encoding profiles for ABR streaming"
#      <http://www.reznik.org/publications.html>
#  * support low-latency streaming a.k.a. "LL", i.e. 3-5s latency
#    + LL-HLS
#      <https://stackoverflow.com/questions/56700705/how-to-enable-lhls-in-ffmpeg-4-1>
#    + CMAF, using either fmpeg or (when released) GPAC
#      <https://github.com/FFmpeg/FFmpeg/commit/cc929ce>
#      <https://github.com/gpac/gpac/commit/4529c60>
#  * support realtime streaming a.k.a. "ULL", i.e. less than 0.5s latency
#      <https://blog.maxwellgale.com/2021/01/30/streaming-video-over-webrtc-using-ffmpeg/>

set -e

PRG=$(basename "$0")

showhelp() {
    cat <<EOF
Usage: $PRG [OPTION...] [--] [ARG=VALUE...] INPUTFILE... [ARG=VALUE...]
Encode video file in multiple web-optimized formats, and provide sample
html favoring open formats with optional non-JavaScript Flash fallback.

  --video           Video style:
                                   [refbpp]
                      talkinghead    0.1
                      action         0.15
                    (default: none)
  -p, --profile     Video frame size and optional rate, delimited by @
                    except "p" labels: e.g. 848x480@25 480p25 wvga@25
  -s, --size        Set video frame size:
                    [modulus 16]
                         320x240 qvga
                         640x480 vga          480p  848x480 wvga
                                              576p 1024x576 wsvga
                        1024x768 xga       hd 720p 1280x720 wxga
                    [modulus 8]
                                              240p  424x240 wqvga
                         480x360 hvga     nhd 360p  640x360
                         800x600 svga
                    (default: use input size)
  -a, --aspect      Display Aspect Ratio in melt format: e.g. @16/9
                    (default: no aspect hinting)
  -r, --rate        Video framerate: e.g. 25 1001/30000
                    (default: use input framerate)
  --refbpp          Bitrate reference: Bits-per-pixel for 360p30 H.264
                    (default: 0.12)
  --formats         Container and media formats to use, comma-separated:
                                  [container]   [video]   [audio]
                      theora ogg     Ogg        Theora     Vorbis
                      vp8 webm       WebM       VP8        Vorbis
                      vp9            WebM       VP9        Opus
                      avc h264 mp4   MPEG-4     H.264/AVC  AAC-LC
                    (default: vp8,vp9,avc)
  --audio           Audio style:
                              [channels] [bitrate per channel]
                      music    max. 2         ~48-64k
                      speech        1         ~32-64k
                      silence       0
                    (default: channel count of first input, ~48-64k)
  --audioprefilter  Add audio filter before loudness
  --loudness        Add EBU R128 loudness filter
  --loudness-data   Add loudness filter, with precomputed results
                    (default: none - compute results on-the-fly)
  --filter          Add filter
  --stem            Stem of output filenames, optionally with path
                    (default: basename of first input)
  -t, --title       Title used in html fallback graphics
                    (default: stem)
  --still-frame     Frame used as still picture
                    (default: 0)
  --sample          Create only a 150 frames long sample
  --sample-start    Create sample, starting at a specific frame
                    (default: start at first frame)
  --sample-length   Create sample of specified length, in frames
                    (default: 150 frames i.e. approx. 5s)
  --compression     Prioritize quality, size or render time:
                              [quality] [bitrate] [speed] [method]
                      dirty                        fast      CQ
                      normal   medium    average   medium    CQ
                      hq       high      exact     slow    2xCQ¹
                      exact    medium    exact     slow    2xCQ¹
                      quality  medium    vary      medium     Q²
                    ¹ theora uses CBR, avc uses 1-pass
                    ² vp8 uses "soft-CQ" (CQ with very high bitrate)
                    (default: normal)
  --quality         Quality rate factor: auto [integer] none
                      suggested, avc:         19  23  27  31  35  39
                      suggested, vp8/vp9:     27  33  39  45  51  57
                    (default: auto - 23 for avc, 33 for vp8/vp9)
  --gpu             Use hardware acceleration: none auto vaapi qsv
                    (default: auto when compression=dirty, else none)
  --gpudevice       Device used for hardware acceleration
                    (default: /dev/dri/renderD128)
  -h, --help        This help text

Examples:
  $PRG -s qvga -t "Funny guy" intro.dv myvideo.dv
  $PRG -p 480p15 --filter "grain noise=20" myvideo.dv

Recommendations for best results:
  * Use square pixels, modulus 16, max. 480 width (qvga vga 480p 720p).
  * Try lower --refbpp while visually acceptable, or raise to max. 0.2.
  * Expand and compress noisy speech with ladspa.1075 and ladspa.1916.

More info:
<http://camendesign.com/code/video_for_everybody>
<http://www.streaminglearningcenter.com/articles/configuring-your-streaming-video-(for-newbies).html>
<http://en.wikipedia.org/wiki/HTML5_video>
<https://web.archive.org/web/20121101044854/http://www.penguinproducer.com/2012/01/ladspa-noise-removal/>
<http://theproaudiofiles.com/mixing-vocals-dynamics-compression-limiting>
<http://www.internetmarketingnotes.com/2010/07/free-embeddable-flash-video-flv-players-for-commercial-use/>
EOF
}

exit1() {
    response="${1:+Error: }${1:-Internal error!}"
    echo >&2 "$response"
    exit 1
}

# defaults
formats=vp8,vp9,avc
stillframe=0
samplestart=0
samplelength=150
compression=normal

# establish/verify rendering settings for reference codec
#
# 1) pick reference frame:encoding format: 360p30:h.264
# 2) pick reference codec (incl. normal settings)
# 3) render reference frame format with --compression=quality
#    * check that bpp is near "refbpp" of 0.12
#      (otherwise adapt settings or do research and update refbpp)
# 4) render misc. frame formats (esp. modulus 16) with --compression=quality
#    * check that bpp derives from refbpp by "rule of .75"
#      (otherwise adapt settings and/or do research and update rule)
# 5) render misc. frame formats with --compression=normal
#    * check that bpp derives from refbpp by "rule of .75"
# 6) optionally pick dirty/hq settings
#    * check that --compression=quality has filesize similar as normal

# establish/verify rendering settings for non-reference codec
#
# 1) pick normal settings
# 2) render reference frame format with --compression=quality
#    * check that this "codec-bpp" deviates reasonably from refbpp
# 3) render misc. frame formats (esp. modulus 16) with --compression=quality
#    * check that bpp derives from codec-bpp by "rule of .75"
#      (otherwise adapt settings and/or extend rule)
# 4) render misc. frame formats with --compression=normal
#    * check that bpp derives from codec-bpp by "rule of .75"
# 5) optionally pick dirty/hq settings
#    * check that --compression=quality has similar filesize as normal

# Theora seems to compress ~100% worse than H.264
# (reportedly 76% worse in 2010, and libtheora has not improved since)
# <http://compression.ru/video/codec_comparison/h264_2010/>
factor_theora=200/100
# VP8 is rumored to compress ~20% worse than H.264
factor_vp8=120/100
# VP9 compresses ~30% better than H.264 for practical use-cases
# <https://engineering.fb.com/2018/04/10/video-engineering/av1-beats-x264-and-libvpx-vp9-in-practical-use-case/>
factor_vp9=70/100
# Mimic ReplayGain level when loudness is enabled (EBU R128: -23LUFS)
# <https://auphonic.com/blog/2013/01/07/loudness-targets-mobile-audio-podcasts-radio-tv/>
lufs=-18

# parse cmdline options
TEMP="`getopt -s sh -o hp:s:a:r:b:t: -l help,profile:,size:,aspect:,rate:,video:,refbpp:,formats:,audio:,audioprefilter:,loudness,loudness-data:,filter:,stem:,title:,still-frame:,sample,sample-start:,sample-length:,compression:,quality:,gpu:,gpudevice: -n "$PRG" -- "$@"`" || exit1 "Internal getopt error."
eval set -- "$TEMP"
while true ; do
    case "$1" in
        -h|--help) showhelp; exit;;
        -p|--profile) profile="$2"; shift 2;;
        -s|--size) size="$2"; shift 2;;
        -a|--aspect) aspect="$2"; shift 2;;
        -r|--rate) framerate="$2"; shift 2;;
        --video) video="$2"; shift 2;;
        --refbpp) refbpp="$2"; shift 2;;
        --formats) formats="$2"; shift 2;;
        --audio) audio="$2"; shift 2;;
        --audioprefilter) audioprefilters="${audioprefilters:+$audioprefilters }-filter $2"; shift 2;;
        --loudness) loudness=yes; shift;;
        --loudness-data) loudness=yes; loudness_data="$2"; shift 2;;
        --filter) filters="${filters:+$filters }-filter $2"; shift 2;;
        --stem) stem="$2"; shift 2;;
        -t|--title) title="$2"; shift 2;;
        --still-frame) stillframe="$2"; shift 2;;
        --sample) sample=yes; shift;;
        --sample-start) sample=yes; samplestart="$2"; shift 2;;
        --sample-length) sample=yes; samplelength="$2"; shift 2;;
        --compression) compression="$2"; shift 2;;
        --quality) quality="$2"; shift 2;;
        --gpu) gpu="$2"; shift 2;;
        --gpudevice) gpudevice="$2"; shift 2;;
        --) shift; break;;
        *) exit1 "Internal error resolving options.";;
    esac
done

[ -n "$sample" ] || notsample=yes

# Resolve if system has many CPU cores
processors=$(nproc)
[ $processors -gt 2 ] || processors=

# sanitize infiles
infiles=$*
infile_first=$(perl -e 'print shift @ARGV' $infiles)
[ -e "$infile_first" ] || exit1 "Cannot read first input file \"$infile_first\"!"

# resolve stem and title (if not explicitly set)
stem=${stem:-$(basename "$infile_first" | perl -pe 's/\.[^.]*//')}
title=${title:-$stem}

#use_ffmpeg_theora=yes
#use_ffmpeg_vp8=yes
#use_ffmpeg_vp9=yes
#use_ffmpeg_avc=yes
#use_ffmpeg2theora=yes # TODO...
#use_vpxenc=yes
# Avoid discrete audio encoders if possible: may cause sync problems
#use_oggenc=yes
#use_opusenc=yes
# Avoid AAC: Lame arguably better than any free AAC encoder
# Argument for AAC: Safari possibly lacks mp3 support in video container
#use_mp3=yes
[ -n "$use_mp3" ] || use_aac=yes

# Avoid MP3 VBR but seems ABR is fine: Set to CBR for strict correctnes
#use_lame_cbr=yes
[ -n "$use_lame_cbr" ] || use_lame_abr=yes

# resolve quality/speed hints
qkey_theora=q
qkey_vp8=crf
qkey_vp9=crf
qkey_avc=crf
q_theora=6
q_vp8=33
cpu_vp8=2
q_vp9=33
cpu_vp9=2
q_avc=23
preset_avc=medium
profile_vp8=0
profile_vp9=0
profile_avc=main
case "$compression" in
    normal) :;;
    dirty)
    gpu=${gpu:-auto}
    cpu_vp8=4
    cpu_vp9=4
    preset_avc=veryfast
    ;;
    hq)
    multipass=yes
    no_q_theora=yes
    no_multipass_avc=yes
    cpu_vp8=0
    cpu_vp9=1
#   preset_avc=veryslow # faster than medium unneeded as of June 2020
    ;;
    exact)
    multipass=yes
    no_multipass_avc=yes
    ;;
    quality)
    no_bitrate=yes
    ;;
    *) exit1 "Unknown compression mode \"$compression\".";;
esac
case "$quality" in
    auto|'') :;;
    none) no_q=yes;;
    *) q=$quality;;
esac
gpudevice=${gpudevice:-/dev/dri/renderD128}
if [ auto = "$gpu" ]; then
    vainfo=$(vainfo --display drm --device "$gpudevice" 2>/dev/null | grep -Po 'Driver version:\s*\K.*')
    case "$vainfo" in
        '') gpu=;;
        *)
        for vformat in theora vp8 vp9 avc; do
            eval "no_gpu_decoder_$vformat=yes"
            eval "no_gpu_encoder_$vformat=yes"
        done
        case $(vainfo --display drm --device "$gpudevice" 2>/dev/null | grep -Po 'VAProfile\S+\s*:\s*VAEntrypointVLD') in
            *VAProfileVP8Version0_3*) no_gpu_decoder_vp8=;;
            *VAProfileVP9Profile0*) no_gpu_decoder_vp9=;;
            *VAProfileH264Main*) no_gpu_decoder_avc=;;
        esac
        case $(vainfo --display drm --device "$gpudevice" 2>/dev/null | grep -Po 'VAProfile\S+\s*:\s*VAEntrypointEncSlice') in
            *VAProfileVP8Version0_3*) no_gpu_encoder_vp8=;;
            *VAProfileVP9Profile0*) no_gpu_encoder_vp9=;;
            *VAProfileH264Main*) no_gpu_encoder_avc=;;
        esac
        if [ -n "$no_gpu_decoder_vp8$no_gpu_decoder_vp9$no_gpu_decoder_avc" ]; then
            case "$vainfo" in
# TODO              *"Intel iHD driver"*) gpu=qsv;;
                *) gpu=vaapi;;
            esac
        fi
    esac
fi
case "$gpu" in
    none|'') gpu=;;
    qsv)
    multipass=
    preset_vp8=medium
    preset_vp9=medium
    preset_avc=
    qkey_avc=qp
    ;;
    vaapi)
    multipass=
    preset_avc=
    qkey_avc=qp
    case "$compression" in
        dirty)
        preset_avc=
        ;;
    esac
    ;;
    *) exit1 "Unknown GPU type \"$gpu\".";;
esac
[ -z "$gpu" ] && no_gpu=yes || eval "gpu_$gpu=yes"

# parse/resolve size and framerate
case "$profile" in
    '') :;;
    *@*)
    while read s r foo; do
        profilesize="${size:-$s}"
        framerate="${framerate:-$r}"
    done << EOF
$(echo "$profile" | perl -F@ -anE 'say join " ", @F')
EOF
    ;;
    *p*)
    while read s r foo; do
        profilesize="${size:-${s}p}"
        framerate="${framerate:-$r}"
    done << EOF
$(echo "$profile" | perl -Fp -anE 'say join " ", @F')
EOF
    ;;
    *)
    profilesize="$profile"
    ;;
esac
size=${size:-$profilesize}
case "$size" in
    qvga) size=320x240;;
    hvga) size=480x360;;
    vga) size=640x480;;
    svga) size=800x600;;
    xga) size=1024x768;;
    240p|wqvga) size=424x240;;
    360p|nhd) size=640x360;;
    480p|wvga) size=848x480;;
    576p|wsvga) size=1024x576;;
    720p|wxga|hd) size=1280x720;;
esac
if [ -n "$size" ]; then
    while read w h foo; do
        width="${width:-$w}"
        height="${height:-$h}"
    done << EOF
$(echo "$size" | perl -Fx -anE 'say join " ", @F')
EOF
    if [ -z "$width" ] || [ -z "$height" ]; then
        exit1 "Failed to parse size \"$size\"."
    fi
fi
case "$framerate" in
    */*)
    while read n d foo; do
        framerate_num="${framerate_num:-$n}"
        framerate_den="${framerate_den:-$d}"
    done << EOF
$(echo "$framerate" | perl -F/ -anE 'say join " ", @F')
EOF
    ;;
    ?*)
    framerate_num="$framerate"
    framerate_den=1
    ;;
esac

# resolve input size and framerate (needed for computing bitrate)
while read w h r s foo; do
    width_in="${width_in:-$w}"
    height_in="${height_in:-$h}"
    framerate_in="${framerate_in:-$r}"
    scantype_in="${scantype_in:-$s}"
done << EOF
$(mediainfo --Inform="Video;%Width% %Height% %FrameRate% %ScanType%" \
    "$infile_first")
EOF
[ Progressive = "$scantype_in" ] || do_deinterlace=yes

case "$video" in
    talkinghead)
    refbpp="${refbpp:-0.1}"
    x264tune=film
    ;;
    action)
    refbpp="${refbpp:-0.15}"
    x264tune=film
    ;;
    '')
    refbpp="${refbpp:-0.12}"
    ;;
    *) exit1 "Unknown video style \"$video\".";;
esac

for format in $(echo "$formats" | sed -e 's/,/ /g'); do
    case $format in
        theora|ogg) ogv=yes;;
        avc|h264|mp4) mp4=yes;;
        vp8|webm) webm=yes;;
        vp9) webm_vp9=yes;;
        *) exit1 "Unknown format \"$format\".";;
    esac
done

_width="${width:-$width_in}"
_height="${height:-$height_in}"

if [ -n "$_width" ] && [ -n "$_height" ]; then
    _pixels="$((_width*_height))"
fi
_frames="${framerate:-$framerate_in}"

qmin_vp8=0
qmin_vp9=0
qmax_vp8=63
qmax_vp9=63
lag_vp8=16
lag_vp9=16
tokenparts_log2_vp8=0
tokenparts_log2_vp9=0
if [ -n "$_pixels" ] && [ $_pixels -ge $((1024*768)) ]; then
    qmin_vp8=11
    qmin_vp9=11
    qmax_vp8=51
    qmax_vp9=51
    tokenparts_log2_vp8=2
    tokenparts_log2_vp9=2
    if [ -n "$_frames" ] && [ $_frames -gt 40 ]; then
        lag_vp8=25
        lag_vp9=25
        profile_vp8=1
        profile_vp9=1
    fi
fi

# compute average bitrate from reference data and "power of .75" rule
if [ -n "$_pixels" ] && [ -n "$_frames" ]; then
    bitrate_avc=$(perl -E '$refsize=640*360;' \
        -E "say int( +(($_pixels/\$refsize)**0.75*\$refsize*$_frames*$refbpp) )")
    bitrate_theora=$(perl -E "say int( +($bitrate_avc*$factor_theora) )") #"
    bitrate_vp8=$(perl -E "say int( +($bitrate_avc*$factor_vp8) )") #"
    bitrate_vp9=$(perl -E "say int( +($bitrate_avc*$factor_vp9) )") #"
fi

# default per-codec-channel bitrates
quality_vorbis=3
bitrate_opus=48
quality_lame=6
bitrate_lame=64
bitrate_aac=64

case "$audio" in
    music)
    opusapp=audio
    ;;
    speech)
    mono=yes
    quality_vorbis=2
    bitrate_opus=32
    quality_lame=7
    bitrate_lame=48
    opusapp=voip
    ;;
    silence)
    silence=yes
    ;;
    '')
    :
    ;;
    *) exit1 "Unknown audio style \"$audio\".";;
esac
which ffprobe > /dev/null && ffprobe=ffprobe || ffprobe=avprobe
channels=$($ffprobe -v warning -show_streams "$infile_first" \
    | perl -ne 's/channels=// and print $_' || echo -1)
# adapt channel count and flags to reflect actual downmix/silence need
if [ -n "$stereo" ]; then
    if [ $channels -gt 2 ]; then
        channels=2
    else
        stereo=
    fi
elif [ -n "$mono" ]; then
    if [ $channels -gt 1 ]; then
        channels=1
    else
        mono=
    fi
elif [ -n "$silence" ]; then
    if [ $channels -gt 0 ]; then
        channels=0
    else
        silence=
    fi
fi
[ -z "$silence" ] || channels=
if [ -n "$stereo$mono" ]; then
    # melt cannot downmix with (stereo?) filters applied (bug#763911)
    if [ -n "$audioprefilters$compress$loudness$limit$filters" ]; then
        if [ -z "${webm:+$use_oggenc}${webm_vp9:+$use_opusenc}" ]; then
            downmix_ffmpeg=yes
        else
            downmix_oggenc=yes
            downmix_oggenc=yes
        fi
    else
        downmix_melt=yes
    fi
fi

encoder_vp8=libvpx
encoder_vp9=libvpx-vp9
encoder_avc=libx264
for vformat in theora vp8 vp9 avc; do
    [ -z "$no_bitrate" ] || eval "no_bitrate_$vformat=yes"
    eval "[ -n \"$no_bitrate\$no_bitrate_$vformat\" ] || do_bitrate_$vformat=yes"
    eval "[ -z \"$no_q\$no_q_$vformat\" ] || q_$vformat="
    [ -z "$q" ] || eval "q_$vformat=$q"
    eval "[ -n \"\$q_$vformat\" ] || no_q_$vformat=yes"
    [ -z "$gpu" ] || eval "[ -n \"\$no_gpu_encoder_$vformat\" ] || encoder_$vformat=${vformat}_$gpu"
    [ avc = "$vformat" ] || eval "[ -n \"$use_vpxenc\$use_ffmpeg_$vformat\" ] || use_melt_$vformat=yes"
    [ -z "$multipass" ] || eval "[ -n \"\$no_multipass_$vformat\" ] || multipass_$vformat=yes"
    eval "[ -n \"\$multipass_$vformat\" ] || singlepass_$vformat=yes"
done
if [ -n "$use_ffmpeg_theora$use_ffmpeg_vp8$use_ffmpeg_vp9$use_vpxenc" ]; then
    [ -n "$use_oggenc" ] || use_ffmpeg_vorbis=yes
    [ -n "$use_opusenc" ] || use_ffmpeg_opus=yes
    [ -z "$use_oggenc$use_opusenc" ] || use_wav=yes
    [ -z "$use_ffmpeg_vorbis$use_ffmpeg_opus" ] || use_matroska=yes
fi
if [ -n "$use_ffmpeg_avc" ]; then
    use_matroska=yes
else
    use_melt_avc=yes
fi
use_melt_any="${ogv:+$use_melt_theora}${webm:+$use_melt_vp8}${webm_vp9:+$use_melt_vp9}${mp4:+$use_melt_avc}"
use_ffmpeg_any="${ogv:+$use_ffmpeg_theora}${webm:+$use_ffmpeg_vp8}${webm_vp9:+$use_ffmpeg_vp9}${mp4:+$use_ffmpeg_avc}"
use_vpxenc_any="${webm:+$use_vpxenc}${webm_vp9:+$use_vpxenc}"
use_opusenc_any="${webm:+$use_oggenc}${webm_vp9:+$use_opusenc}"
use_oggenc_any="${ogv:+$use_oggenc}${webm:+$use_oggenc}${webm_vp9:+$use_opusenc}"
[ -z "$use_ffmpeg_any$use_vpxenc_any" ] || use_any_nomelt_video=yes

[ -z "$gpu" ] || echo >&2 "GPU platform enabled: $gpu"

# gpu options
_melt_gpu_init=${gpu:+ ${gpu}_device=$gpudevice}
_ffmpeg_gpu_vaapi_init="-${gpu}_device $gpudevice"
_ffmpeg_gpu_qsv_init="-init_hw_device qsv=hw:$gpudevice -filter_hw_device hw"
_ffmpeg_gpu_init="${gpu_vaapi:+ ${_ffmpeg_gpu_vaapi_init}}${gpu_qsv:+ ${_ffmpeg_gpu_qsv_init}}"
_ffmpeg_gpu_vaapi_filter="-vf 'format=nv12|$gpu,hwupload'"
_ffmpeg_gpu_qsv_filter="-vf 'hwupload=extra_hw_frames=64,format=qsv'"
_ffmpeg_gpu_filter="${gpu_vaapi:+ ${_ffmpeg_gpu_vaapi_filter}}${gpu_qsv:+ ${_ffmpeg_gpu_qsv_filter}}"

# generic options
which ffmpeg > /dev/null && ffmpeg=ffmpeg || ffmpeg=avconv
_melt_infiles="${notsample:+$infiles}${sample:+$infile_first in=$((samplestart)) out=$((samplestart + samplelength))}"

render() {
    engine=$1; shift
    chained=
    echo=
    unchained=yes
    case "$engine" in
        ffmpeg-chained) engine=ffmpeg; chained=yes;;
        vpxenc-chained) engine=vpxenc; chained=yes;;
        ffmpeg-chained-echo) engine=ffmpeg; chained=yes; echo=yes;;
        vpxenc-chained-echo) engine=vpxenc; chained=yes; echo=yes;;
    esac
    [ -z "$chained" ] || unchained=
    case "$engine" in
        melt)
        set -- -progress$_melt_gpu_init "$@"
        ;;
        ffmpeg)
        engine=$ffmpeg
        set -- -threads auto -y -hide_banner -v ${unchained:+info -stats}${chained:+warning}$_ffmpeg_gpu_init "$@"
        ;;
        vpxenc)
        set -- --quiet ${processors:+-t $((processors-1))} "$@"
        ;;
        oggenc)
        set -- "$@"
        ;;
        opusenc)
        set -- "$@"
        ;;
        *) exit1 "Unknown render engine \"$engine\".";;
    esac
    ${echo:+echo }"$engine" "$@"
}

# filter options
_melt_video="progressive=1${framerate:+ frame_rate_num="$framerate_num" frame_rate_den="$framerate_den"}${size:+ s=${width:+$width}x${height:+$height}}${aspect:+ aspect=$aspect}"
_ffmpeg_video="${no_gpu:+${do_deinterlace:+-filter:v yadif}${framerate:+ -r $framerate_num/$framerate_den}${size:+ -s ${width:+$width}x${height:+$height}}${aspect:+ -aspect $aspect}}"
_melt_downmix="${downmix_melt:+ac=$channels}"
_ffmpeg_downmix="${downmix_ffmpeg:+-ac $channels}"
# FIXME: how to downmix to stereo?
_oggenc_downmix="${downmix_oggenc:+${mono:+--downmix}}"
_opusenc_downmix="${downmix_opusenc:+${stereo:+--downmix-stereo}${mono:+--downmix-mono}}"
# limit (i.e. avoid peaks "clipping")
_melt_postfilters_audio="${limit:+-filter ladspa.1077}"

# format bitrate options
_melt_bitrate_theora="${do_bitrate_theora:+ vb=$bitrate_theora}"
_ffmpeg_bitrate_theora="${do_bitrate_theora:+ -b:v $bitrate_theora}"
_melt_bitrate_vp8="${do_bitrate_vp8:+ vb=$bitrate_vp8 minrate=$((bitrate_vp8/20)) maxrate=$((bitrate_vp8*12))}"
_ffmpeg_bitrate_vp8="${do_bitrate_vp8:+ -b:v $bitrate_vp8 -minrate $((bitrate_vp8/20)) -maxrate $((bitrate_vp8*12))}"
_vpxenc_bitrate_vp8="${do_bitrate_vp8:+ --target-bitrate=$((bitrate_vp8/1000))${multipass_vp8:+ --minsection-pct=5 --maxsection-pct=1200}}"
_melt_bitrate_vp9="${do_bitrate_vp9:+ vb=$bitrate_vp9 minrate=$((bitrate_vp9/20)) maxrate=$((bitrate_vp9*12))}"
_ffmpeg_bitrate_vp9="${do_bitrate_vp9:+ -b:v $bitrate_vp9 -minrate $((bitrate_vp9/20)) -maxrate $((bitrate_vp9*12))}"
_vpxenc_bitrate_vp9="${do_bitrate_vp9:+ --target-bitrate=$((bitrate_vp9/1000))${multipass_vp9:+ --minsection-pct=5 --maxsection-pct=1200}}"
_melt_bitrate_avc="${do_bitrate_avc:+ maxrate=$bitrate_avc${no_gpu:+ bufsize=$((bitrate_avc*2))}}"
_ffmpeg_bitrate_avc="${do_bitrate_avc:+ -maxrate $bitrate_avc${no_gpu:+ -bufsize $((bitrate_avc*2))}}"

# format low-level options
_melt_misc_vp8=" qmin=$qmin_vp8 qmax=$qmax_vp8 g=120 deadline=good cpu-used=$cpu_vp8${profile_vp8:+ vprofile=$profile_vp8} auto-alt-ref=1 lag-in-frames=$lag_vp8 arnr-maxframes=7 arnr-strength=5 arnr-type=centered"
_ffmpeg_misc_vp8=" -qmin $qmin_vp8 -qmax $qmax_vp8 -g 120 -deadline good -cpu-used $cpu_vp8${profile_vp8:+ -profile:v $profile_vp8} -auto-alt-ref 1 -lag-in-frames $lag_vp8 -arnr-maxframes 7 -arnr-strength 5 -arnr-type centered"
_vpxenc_misc_vp8=" --passes=${singlepass_vp8:+1}${multipass_vp8:+2} --min-q=$qmin_vp8 --max-q=$qmax_vp8 --kf-max-dist=120 --good --cpu-used=$cpu_vp8${profile_vp8:+ --profile=$profile_vp8} --token-parts=$tokenparts_log2_vp8 --auto-alt-ref=1 --lag-in-frames=$lag_vp8 --arnr-maxframes=7 --arnr-strength=5 --arnr-type=3"
_melt_misc_vp9=" qmin=$qmin_vp9 qmax=$qmax_vp9 g=120 deadline=good cpu-used=$cpu_vp9${profile_vp9:+ vprofile=$profile_vp9} auto-alt-ref=1 lag-in-frames=$lag_vp9 arnr-maxframes=7 arnr-strength=5 arnr-type=centered"
_ffmpeg_misc_vp9=" -qmin $qmin_vp9 -qmax $qmax_vp9 -g 120 -deadline good -cpu-used $cpu_vp9${profile_vp9:+ -profile:v $profile_vp9} -auto-alt-ref 1 -lag-in-frames $lag_vp9 -arnr-maxframes 7 -arnr-strength 5 -arnr-type centered"
_vpxenc_misc_vp9=" --passes=${singlepass_vp9:+1}${multipass_vp9:+2} --min-q=$qmin_vp9 --max-q=$qmax_vp9 --kf-max-dist=120 --good --cpu-used=$cpu_vp9${profile_vp9:+ --profile=$profile_vp9} --tile-columns=$tokenparts_log2_vp9 --tile-rows=$tokenparts_log2_vp9 --auto-alt-ref=1 --lag-in-frames=$lag_vp9 --arnr-maxframes=7 --arnr-strength=5 --arnr-type=3"
_melt_misc_avc="${no_gpu:+ movflags=+faststart}"
_ffmpeg_misc_avc="${no_gpu:+ -movflags +faststart}"

# format options
# emulate VP8 Constant Quality using "loose" Constrained Quality
_melt_theora="vcodec=libtheora$_melt_bitrate_theora${q_theora:+ $qkey_theora=$q_theora}"
_ffmpeg_theora="-c:v libtheora$_ffmpeg_bitrate_theora${q_theora:+ -$qkey_theora:v $q_theora}"
_melt_vp8="vcodec=$encoder_vp8${pre_vp8:+ vpreset=$pre_vp8}${preset_vp8:+ preset=$preset_vp8}$_melt_bitrate_vp8${q_vp8:+ ${no_bitrate_vp8:+vb=$((bitrate_vp8*10)) }$qkey_vp8=$q_vp8}$_melt_misc_vp8"
_ffmpeg_vp8="-c:v $encoder_vp8${pre_vp8:+ -vpre $pre_vp8}${preset_vp8:+ preset $preset_vp8}$_ffmpeg_bitrate_vp8${q_vp8:+ ${no_bitrate_vp8:+-b:v $((bitrate_vp8*10)) }-$qkey_vp8 $q_vp8}$_ffmpeg_misc_vp8"
_vpxenc_vp8="--codec=vp8$_vpxenc_bitrate_vp8 --end-usage=${no_q_vp8:+vbr}${q_vp8:+cq --cq-level=$q_vp8}$_vpxenc_misc_vp8"
_melt_vp9="vcodec=$encoder_vp9${pre_vp9:+ vpreset=$pre_vp9}${preset_vp9:+ preset=$preset_vp9}$_melt_bitrate_vp9${q_vp9:+ ${no_bitrate_vp9:+vb=0 }$qkey_vp9=$q_vp9}$_melt_misc_vp9"
_ffmpeg_vp9="-c:v $encoder_vp9${pre_vp9:+ -vpre $pre_vp9}${preset_vp9:+ -preset $preset_vp9}$_ffmpeg_bitrate_vp9${q_vp9:+ ${no_bitrate_vp9:+-b:v 0 }-$qkey_vp9 $q_vp9}$_ffmpeg_misc_vp9"
_vpxenc_vp9="--codec=vp9$_vpxenc_bitrate_vp9 --end-usage=${no_q_vp9:+vbr}${q_vp9:+cq --cq-level=$q_vp9}$_vpxenc_misc_vp9"
_melt_avc="vcodec=$encoder_avc${preset_avc:+ vpreset=$preset_avc}${profile_avc:+ vprofile=$profile_avc}${x264tune:+ tune=$x264tune}$_melt_bitrate_avc${q_avc:+ $qkey_avc=$q_avc} threads=0$_melt_misc_avc"
_ffmpeg_avc="-c:v $encoder_avc${preset_avc:+ -preset $preset_avc}${profile_avc:+ -profile:v $profile_avc}${x264tune:+ -tune $x264tune}$_ffmpeg_bitrate_avc${q_avc:+ -$qkey_avc $q_avc}$_ffmpeg_misc_avc"
_melt_pcm="$_melt_downmix acodec=pcm_s16le"
_melt_vorbis="$_melt_downmix acodec=libvorbis aq=$quality_vorbis"
_ffmpeg_vorbis="$_ffmpeg_downmix -c:a libvorbis -q:a $quality_vorbis"
_oggenc_vorbis="$_oggenc_downmix -q $quality_vorbis"
_melt_opus="$_melt_downmix acodec=libopus ab=$((channels*bitrate_opus))k${opusapp:+ application=$opusapp}"
_ffmpeg_opus="$_ffmpeg_downmix -c:a libopus -b:a $((channels*bitrate_opus))k${opusapp:+ -application $opusapp}"
_opusenc_opus="$_opusenc_downmix --bitrate $((channels*bitrate_opus))"
_melt_mp3="$_melt_downmix acodec=libmp3lame${use_lame_abr:+ aq=$quality_lame}${use_lame_cbr:+ ab=$((channels*bitrate_lame))k}"
_ffmpeg_mp3="$_ffmpeg_downmix -c:a libmp3lame${use_lame_abr:+ -q:a $quality_lame}${use_lame_cbr:+ -b:a $((channels*bitrate_lame))k}"
_melt_aac="$_melt_downmix acodec=aac ab=$((channels*bitrate_aac))k"
_ffmpeg_aac="$_ffmpeg_downmix -c:a aac -b:a $((channels*bitrate_aac))k"

# container options
_melt_stdout="-consumer avformat:pipe:1 f=yuv4mpegpipe $_melt_video pix_fmt=yuv420p an=1 audio_off=1"
_ffmpeg_stdin="-f yuv4mpegpipe -i pipe:0$_ffmpeg_gpu_filter"
_ffmpeg_rawvideo="-f rawvideo"
_melt_wav="f=wav $_melt_pcm vn=1 video_off=1"
_ffmpeg_wav_in="-f wav"
_oggenc_wav_in=
_opusenc_wav_in=
_melt_matroska_pcm="f=matroska $_melt_downmix $_melt_pcm vn=1 video_off=1"
_ffmpeg_matroska_pcm_in="-f matroska"
_melt_ogv="f=ogg $_melt_video $_melt_theora ${silence:+an=1 audio_off=1}${channels:+$_melt_vorbis}"
_ffmpeg_ogv="-f ogg $_ffmpeg_video $_ffmpeg_theora ${silence:+-an}${channels:+ $_ffmpeg_vorbis}"
_ffmpeg_ogv_in="-f ogg"
_ffmpeg_ogv_keepvideo="-f ogg $_ffmpeg_video -c:v copy ${silence:+-an}${channels:+$_ffmpeg_vorbis}"
_ffmpeg_ogg_in="-f ogg"
_ffmpeg_ogv_onlyvideo="-f ogg $_ffmpeg_video $_ffmpeg_theora -an"
_ffmpeg_ogg_vorbis="-f ogg -vn $_ffmpeg_vorbis"
_ffmpeg_ogg_opus="-f ogg -vn $_ffmpeg_opus"
_melt_webm="f=webm $_melt_video $_melt_vp8 ${silence:+an=1 audio_off=1}${channels:+$_melt_vorbis}"
_ffmpeg_webm="-f webm $_ffmpeg_video $_ffmpeg_vp8 ${silence:+-an}${channels:+ $_ffmpeg_vorbis}"
_ffmpeg_webm_in="-f webm"
_ffmpeg_webm_onlyvideo="-f webm $_ffmpeg_video $_ffmpeg_vp8 -an"
_ffmpeg_webm_keepvideo="-f webm $_ffmpeg_video -c:v copy ${silence:+-an}${channels:+$_ffmpeg_vorbis}"
_ffmpeg_webm_keepvideo_opus="-f webm $_ffmpeg_video -c:v copy ${silence:+-an}${channels:+$_ffmpeg_opus}"
_melt_webm_vp9="f=webm $_melt_video $_melt_vp9 ${silence:+an=1 audio_off=1}${channels:+$_melt_opus}"
_ffmpeg_webm_vp9="-f webm $_ffmpeg_video $_ffmpeg_vp9 ${silence:+-an}${channels:+$_ffmpeg_opus}"
_ffmpeg_webm_vp9_onlyvideo="-f webm $_ffmpeg_video $_ffmpeg_vp9 -an"
_melt_mp4="f=mp4 $_melt_video $_melt_avc ${silence:+an=1 audio_off=1}${channels:+${use_mp3:+$_melt_mp3}${use_aac:+$_melt_aac}}"
_ffmpeg_mp4="-f mp4 $_ffmpeg_video $_ffmpeg_avc ${silence:+-an}${channels:+${use_mp3:+$_ffmpeg_mp3}${use_aac:+$_ffmpeg_aac}}"
_ffmpeg_mp4_onlyvideo="-f mp4 $_ffmpeg_video $_ffmpeg_avc -an"
_ffmpeg_mp4_keepvideo="-f mp4 $_ffmpeg_video -c:v copy ${silence:+-an}${channels:+$_ffmpeg_mp3}"
_ffmpeg_mp4_keepvideo_aac="-f mp4 $_ffmpeg_video -c:v copy ${silence:+-an}${channels:+$_ffmpeg_aac}"
_melt_img="f=image2 $_melt_video"

_melt_loudness="$loudness_data"
if [ -n "$loudness" ] && [ -z "$silence$_melt_loudness" ]; then
    echo >&2 "Analyzing audio dynamics..."
    render melt $_melt_infiles \
        $audioprefilters -filter loudness $filters \
        -consumer xml:$stem.xml video_off=1 all=1
    _melt_loudness="$(perl -n \
        -e 'm!<(property) name="results">([^<]+)</\1>! and print $2' \
        $stem.xml)"
    echo >&2 "Loudness data: $_melt_loudness"
fi

if [ -n "${ogv:+$multipass_theora}${webm:+$multipass_vp8}${webm_vp9:+$multipass_vp9}${mp4:+$multipass_avc}" ]; then
    echo >&2 "Analyzing video complexity..."
    render melt $_melt_infiles $filters \
        ${ogv:+${use_melt_theora:+-consumer avformat:$stem.ogv $_melt_ogv \
            pass=1 passlogfile=${stem}_theora}} \
        ${webm:+${use_melt_vp8:+-consumer avformat:$stem.webm $_melt_webm \
            pass=1 cpu-used=4 passlogfile=${stem}_vp8}} \
        ${webm_vp9:+${use_melt_vp9:+-consumer avformat:${stem}_vp9.webm $_melt_webm_vp9 \
            pass=1 cpu-used=4 passlogfile=${stem}_vp9}} \
        ${mp4:+${use_melt_avc:+-consumer avformat:$stem.mp4 $_melt_mp4 \
            pass=1 passlogfile=${stem}_avc}} \
        ${use_any_nomelt_video:+$_melt_stdout} \
    | pee \
        ${use_ffmpeg_any:+"$(render ffmpeg-chained-echo $_ffmpeg_stdin \
            ${ogv:+${use_ffmpeg_theora:+$_ffmpeg_rawvideo $_ffmpeg_ogv -an \
                -pass 1 -passlogfile ${stem}_theora /dev/null}} \
            ${webm:+${use_ffmpeg_vp8:+$_ffmpeg_rawvideo $_ffmpeg_vp8 -an \
                -pass 1 -cpu-used 4 -passlogfile ${stem}_vp8 /dev/null}} \
            ${webm_vp9:+${use_ffmpeg_vp9:+$_ffmpeg_rawvideo $_ffmpeg_vp9 -an \
                -pass 1 -cpu-used 4 -passlogfile ${stem}_vp9 /dev/null}} \
            ${mp4:+${use_ffmpeg_avc:+$_ffmpeg_rawvideo $_ffmpeg_avc -an \
                -pass 1 -passlogfile ${stem}_avc /dev/null}})"} \
        ${webm:+${use_vpxenc:+"$(render vpxenc-chained-echo - $_vpxenc_vp8 \
            --pass=1 --cpu-used=4 --fpf=${stem}_vp8.log -o /dev/null)"}} \
        ${webm_vp9:+${use_vpxenc:+"$(render vpxenc-chained-echo - $_vpxenc_vp9 \
            --pass=1 --cpu-used=4 --fpf=${stem}_vp9.log -o /dev/null)"}}
fi

echo >&2 "Encoding video${channels:+ and audio}..."
render melt $_melt_infiles \
    ${channels:+$audioprefilters \
        ${_melt_loudness:+-filter loudness program=$lufs \
        results="$_melt_loudness"}} \
    $filters${channels:+ $_melt_postfilters_audio} \
    ${ogv:+${use_melt_theora:+-consumer avformat:$stem.ogv $_melt_ogv \
        ${multipass_theora:+pass=2 passlogfile=${stem}_theora}}} \
    ${webm:+${use_melt_vp8:+-consumer avformat:$stem.webm $_melt_webm \
        ${multipass_vp8:+pass=2 passlogfile=${stem}_vp8}}} \
    ${webm_vp9:+${use_melt_vp9:+-consumer avformat:${stem}_vp9.webm $_melt_webm_vp9 \
        ${multipass_vp9:+pass=2 passlogfile=${stem}_vp9}}} \
    ${mp4:+${use_melt_avc:+-consumer avformat:$stem.mp4 $_melt_mp4
        ${multipass_avc:+pass=2 passlogfile=${stem}_avc}}} \
    ${channels:+${use_wav:+-consumer avformat:$stem.wav \
        $_melt_wav}} \
    ${channels:+${use_matroska:+-consumer avformat:$stem.mkv \
        $_melt_matroska_pcm}} \
    ${use_any_nomelt_video:+$_melt_stdout} \
| pee \
    ${use_ffmpeg_any:+"$(render ffmpeg-chained-echo $_ffmpeg_stdin \
        ${ogv:+${use_ffmpeg_theora:+$_ffmpeg_ogv_onlyvideo \
            ${multipass_theora:+-pass 2 -passlogfile ${stem}_theora} \
            ${stem}${channels:+_silent}.ogv}} \
        ${webm:+${use_ffmpeg_vp8:+$_ffmpeg_webm_onlyvideo \
            ${multipass_vp8:+-pass 2 -passlogfile ${stem}_vp8} \
            ${stem}${channels:+_silent}.webm}} \
        ${webm_vp9:+${use_ffmpeg_vp9:+$_ffmpeg_webm_vp9_onlyvideo \
            ${multipass_vp9:+-pass 2 -passlogfile ${stem}_vp9} \
            ${stem}_vp9${channels:+_silent}.webm}} \
        ${mp4:+${use_ffmpeg_avc:+$_ffmpeg_mp4_onlyvideo \
            ${multipass_avc:+-pass 2 -passlogfile ${stem}_avc} \
            ${stem}${channels:+_silent}.mp4}})"} \
    ${webm:+${use_vpxenc:+"$(render vpxenc-chained-echo - $_vpxenc_vp8 \
        ${multipass_vp8:+--pass=2 --fpf=${stem}_vp8.log} \
        -o ${stem}${channels:+_silent}.webm)"}} \
    ${webm_vp9:+${use_vpxenc:+"$(render vpxenc-chained-echo - $_vpxenc_vp9 \
        ${multipass_vp9:+--pass=2 --fpf=${stem}_vp9.log} \
        -o ${stem}_vp9${channels:+_silent}.webm)"}}
if [ -n "${ogv:+$use_oggenc$use_ffmpeg_theora}${webm:+$use_vpxenc$use_oggenc}" ] && [ -n "$channels" ]; then
    echo >&2 "Encoding Vorbis audio and muxing with video..."
    # TODO: Encode once, muxing to containers as needed
    if [ -n "$use_oggenc" ]; then
        render oggenc $_oggenc_wav_in $_oggenc_vorbis -o - $stem.wav \
        | render ffmpeg-chained \
            ${ogv:+${use_oggenc:+$_ffmpeg_ogv_in -i ${stem}_silent.ogv \
                $_ffmpeg_ogg_in -i pipe:0 \
                $_ffmpeg_ogv_keepvideo $stem.ogv}} \
            ${webm:+${use_oggenc:+$_ffmpeg_webm_in -i ${stem}_silent.webm \
                $_ffmpeg_ogg_in -i pipe:0 \
                $_ffmpeg_webm_keepvideo $stem.webm}}
    fi
    if [ -n "${ogv:+$use_ffmpeg_theora}" ]; then
        render ffmpeg $_ffmpeg_ogv_in -i ${stem}_silent.ogv \
            $_ffmpeg_matroska_wav_in -i $stem.mkv \
            $_ffmpeg_ogv_keepvideo ${stem}.ogv
    fi
    if [ -n "${webm:+$use_vpxenc}" ]; then
        render ffmpeg $_ffmpeg_webm_in -i ${stem}_silent.webm \
            $_ffmpeg_matroska_wav_in -i $stem.mkv \
            $_ffmpeg_webm_keepvideo ${stem}.webm
    fi
fi
if [ -n "${webm_vp9:+$use_vpxenc$use_opusenc}" ] && [ -n "$channels" ]; then
    echo >&2 "Encoding Opus audio and muxing with VP9 video..."
    if [ -n "$use_opusenc" ]; then
        render opusenc $_opusenc_wav_in $_opusenc_opus -o - $stem.wav \
        | render ffmpeg-chained \
            $_ffmpeg_webm_in -i ${stem}_vp9_silent.webm \
            $_ffmpeg_ogg_in -i pipe:0 \
            $_ffmpeg_webm_keepvideo_opus ${stem}_vp9.webm
    else
        render ffmpeg $_ffmpeg_webm_in -i ${stem}_vp9_silent.webm \
            $_ffmpeg_matroska_wav_in -i $stem.mkv \
            $_ffmpeg_webm_keepvideo_opus ${stem}_vp9.webm
    fi
fi
if [ -n "${mp4:+$use_ffmpeg_avc}" ] && [ -n "$channels" ]; then
    if [ -n "$use_mp3" ]; then
        echo "Encoding MP3 audio and muxing with H.264/AVC video..."
        render ffmpeg $_ffmpeg_mp4_in -i ${stem}_silent.mp4 \
            $_ffmpeg_matroska_wav_in -i $stem.mkv \
            $_ffmpeg_mp4_keepvideo ${stem}.mp4
    else
        echo "Encoding AAC audio and muxing with H.264/AVC video..."
        render ffmpeg $_ffmpeg_mp4_in -i ${stem}_silent.mp4 \
            $_ffmpeg_matroska_wav_in -i $stem.mkv \
            $_ffmpeg_mp4_keepvideo_aac ${stem}.mp4
    fi
fi

# cleanup encoding cruft
rm -f $stem.xml $stem.wav $stem.mkv ${stem}_*.log ${stem}_*silent.webm

# JPEG preview
render melt $infile_first in=$stillframe out=$stillframe \
    -group $filters \
    -consumer avformat:$stem.jpg $_melt_img

__width="${_width:+ width=\"$_width\"}"
__height="${_height:+ height=\"$_height\"}"
# Flash object needs extra space for controllers
__heightplus=${_height:+ height=\"$((_height+4))\"}
_source_ogv="<source src=\"$stem.ogv\" type='video/ogg; codecs=\"theora, vorbis\"' />"
_source_webm="<source src=\"$stem.webm\" type='video/webm; codecs=\"vp8, vorbis\"' />"
_source_webm_vp9="<source src=\"${stem}_vp9.webm\" type='video/webm; codecs=\"vp9, opus\"' />"
# TODO: add type, with h.264 level inspected after rendering
# <https://stackoverflow.com/questions/16363167/html5-video-tag-codecs-attribute#answer-16365526>
_source_mp4="<source src=\"$stem.mp4\" type=\"video/mp4\" />"
[ -z "$flashplayer" ] || flash=yes
[ -n "$mp4" ] || [ -z "$flash" ] || error1 "Cannot enable flash when mp4 format is disabled."
_object_flash="<object$__width$__heightplus type=\"application/x-shockwave-flash\" data=\"$flashplayer.swf\">"
_param_name="<param name=\"movie\" value=\"$flashplayer.swf\" />"
_param_flashvars="<param name=\"flashvars\" value=\"image=$stem.jpg&amp;file=$stem.mp4\" />"
__ogvfile=${ogv:+open format <a href=\"$stem.ogv\">Ogg/Thera</a>}
__webmfile=${webm:+open format <a href=\"$stem.webm\">WebM/VP8</a>}
__webm_vp9_file=${vp9:+open format <a href=\"${stem}_vp9.webm\">WebM/VP9</a>}
__mp4file=${mp4:+closed format <a href=\"$stem.mp4\">MPEG-4/AVC</a>}

cat >"$stem.html" <<EOF
<!-- Video for Everybody, Kroc Camen of Camen Design -->
<video$__width$__height preload controls>
${mp4:+$_source_mp4
}${webm_vp9:+$_source_webm_vp9
}${webm:+$_source_webm
}${ogv:+$_source_ogv
}${flash:+$_object_flash
$_param_name
$_param_flashvars
}<img src="$stem.jpg"$__width$__height alt="$title"
 title="No video playback capabilities, please download the video below" />
${flash:+</object>
}</video>
<p><strong>Download Video:</strong><ul>
${webm_vp9:+<li>$__webm_vp9_file
}${webm:+<li>$__webmfile
}${ogv:+<li>$__ogvfile
}${mp4:+<li>$__mp4file
}</ul></p>
EOF