#!/usr/bin/env bash scriptVersion="1.0.032" if [ -z "$lidarrUrl" ] || [ -z "$lidarrApiKey" ]; then lidarrUrlBase="$(cat /config/config.xml | xq | jq -r .Config.UrlBase)" if [ "$lidarrUrlBase" = "null" ]; then lidarrUrlBase="" else lidarrUrlBase="/$(echo "$lidarrUrlBase" | sed "s/\///g")" fi lidarrApiKey="$(cat /config/config.xml | xq | jq -r .Config.ApiKey)" lidarrUrl="http://127.0.0.1:8686${lidarrUrlBase}" fi agent="lidarr-extended ( https://github.com/RandomNinjaAtk/docker-lidarr-extended )" musicbrainzMirror=https://musicbrainz.org # Debugging Settings #addFeaturedVideoArtists=true if [ "$dlClientSource" = "tidal" ] || [ "$dlClientSource" = "both" ]; then sourcePreference=tidal fi log () { m_time=`date "+%F %T"` echo $m_time" :: Extended Video :: "$1 } verifyApiAccess () { until false do lidarrTest=$(wget --timeout=0 -q -O - "$lidarrUrl/api/v1/system/status?apikey=${lidarrApiKey}" | jq -r .appName) if [ "$lidarrTest" == "Lidarr" ]; then lidarrVersion=$(wget --timeout=0 -q -O - "$lidarrUrl/api/v1/system/status?apikey=${lidarrApiKey}" | jq -r .version) log "Lidarr Version: $lidarrVersion" break else log "Lidarr is not ready, sleeping until valid response..." sleep 1 fi done } log "-----------------------------------------------------------------------------" log "|~) _ ._ _| _ ._ _ |\ |o._ o _ |~|_|_|" log "|~\(_|| |(_|(_)| | || \||| |_|(_||~| | |<" log "Presents: lidarr-extended ($scriptVersion)" log "Docker Version: $dockerVersion" log "May the vidz be with you!" log "-----------------------------------------------------------------------------" log "Donate: https://github.com/sponsors/RandomNinjaAtk" log "Project: https://github.com/RandomNinjaAtk/docker-lidarr-extended" log "Support: https://github.com/RandomNinjaAtk/docker-lidarr-extended/discussions" log "-----------------------------------------------------------------------------" sleep 5 log "" log "Lift off in..."; sleep 0.5 log "5"; sleep 1 log "4"; sleep 1 log "3"; sleep 1 log "2"; sleep 1 log "1"; sleep 1 Configuration () { processstartid="$(ps -A -o pid,cmd|grep "start_video.sh" | grep -v grep | head -n 1 | awk '{print $1}')" processdownloadid="$(ps -A -o pid,cmd|grep "video.sh" | grep -v grep | head -n 1 | awk '{print $1}')" log "To kill script, use the following command:" log "kill -9 $processstartid" log "kill -9 $processdownloadid" sleep 2 verifyApiAccess downloadPath="$downloadPath/videos" log "CONFIG :: Download Location :: $downloadPath" log "CONFIG :: Music Video Location :: $videoPath" log "CONFIG :: Subtitle Language set to: $youtubeSubtitleLanguage" log "CONFIG :: Upgrading yt-dlp to the latest version..." pip install yt-dlp --upgrade --no-cache-dir &>/dev/null log "CONFIG :: Complete" } CacheMusicbrainzRecords () { log "$processCount of $lidarrArtistIdsCount :: MBZDB :: $lidarrArtistName :: Processing..." log "$processCount of $lidarrArtistIdsCount :: MBZDB :: $lidarrArtistName :: Checking Musicbrainz for recordings..." musicbrainzArtistRecordings=$(curl -s -A "$agent" "$musicbrainzMirror/ws/2/recording?artist=$lidarrArtistMusicbrainzId&limit=1&offset=0&fmt=json") sleep 1 musicbrainzArtistRecordingsCount=$(echo "$musicbrainzArtistRecordings" | jq -r '."recording-count"') log "$processCount of $lidarrArtistIdsCount :: MBZDB :: $lidarrArtistName :: $musicbrainzArtistRecordingsCount recordings found..." if [ ! -d /config/extended/cache/musicbrainz ]; then mkdir -p /config/extended/cache/musicbrainz chmod 777 /config/extended/cache/musicbrainz chown abc:abc /config/extended/cache/musicbrainz fi if [ -f "/config/extended/cache/musicbrainz/$lidarrArtistId--$lidarrArtistMusicbrainzId--recordings.json" ]; then if ! [[ $(find "/config/extended/cache/musicbrainz" -type f -name "$lidarrArtistId--$lidarrArtistMusicbrainzId--recordings.json" -mtime +7 -print) ]]; then log "$processCount of $lidarrArtistIdsCount :: MBZDB :: $lidarrArtistName :: Previously cached, skipping..." return else log "$processCount of $lidarrArtistIdsCount :: MBZDB :: $lidarrArtistName :: Previously cached, data needs to be updated..." rm "/config/extended/cache/musicbrainz/$lidarrArtistId--$lidarrArtistMusicbrainzId--recordings.json" fi musicbrainzArtistDownloadedRecordingsCount=$(cat "/config/extended/cache/musicbrainz/$lidarrArtistId--$lidarrArtistMusicbrainzId--recordings.json" | jq -r .id | wc -l) if [ $musicbrainzArtistRecordingsCount -ne $musicbrainzArtistDownloadedRecordingsCount ]; then log "$processCount of $lidarrArtistIdsCount :: MBZDB :: $lidarrArtistName :: Previously cached, data needs to be updated..." rm "/config/extended/cache/musicbrainz/$lidarrArtistId--$lidarrArtistMusicbrainzId--recordings.json" fi if ! cat "/config/extended/cache/musicbrainz/$lidarrArtistId--$lidarrArtistMusicbrainzId--recordings.json" | grep -i "artist-credit" | read; then log "$processCount of $lidarrArtistIdsCount :: MBZDB :: $lidarrArtistName :: Previously cached, data needs to be updated..." rm "/config/extended/cache/musicbrainz/$lidarrArtistId--$lidarrArtistMusicbrainzId--recordings.json" fi fi if [ ! -f "/config/extended/cache/musicbrainz/$lidarrArtistId--$lidarrArtistMusicbrainzId--recordings.json" ]; then offsetcount=$(( $musicbrainzArtistRecordingsCount / 100 )) for ((i=0;i<=$offsetcount;i++)); do if [ $i != 0 ]; then offset=$(( $i * 100 )) dlnumber=$(( $offset + 100)) else offset=0 dlnumber=$(( $offset + 100)) fi log "$processCount of $lidarrArtistIdsCount :: MBZDB :: $lidarrArtistName :: Downloading page $i... ($offset - $dlnumber Results)" curl -s -A "$agent" "$musicbrainzMirror/ws/2/recording?artist=$lidarrArtistMusicbrainzId&inc=artist-credits+url-rels+recording-rels+release-rels+release-group-rels&limit=100&offset=$offset&fmt=json" | jq -r ".recordings[]" >> "/config/extended/cache/musicbrainz/$lidarrArtistId--$lidarrArtistMusicbrainzId--recordings.json" sleep 1 done fi } TidalClientSetup () { log "TIDAL :: Verifying tidal-dl configuration" touch /config/xdg/.tidal-dl.log if [ -f /config/xdg/.tidal-dl.json ]; then rm /config/xdg/.tidal-dl.json fi if [ ! -f /config/xdg/.tidal-dl.json ]; then log "TIDAL :: No default config found, importing default config \"tidal.json\"" if [ -f /config/extended/scripts/tidal-dl.json ]; then cp /config/extended/scripts/tidal-dl.json /config/xdg/.tidal-dl.json chmod 777 -R /config/xdg/ fi fi TidaldlStatusCheck tidal-dl -o $downloadPath/incomplete if [ ! -f /config/xdg/.tidal-dl.token.json ]; then TidaldlStatusCheck log "TIDAL :: ERROR :: Downgrade tidal-dl for workaround..." pip install tidal-dl==2022.3.4.2 --no-cache-dir &>/dev/null TidaldlStatusCheck log "TIDAL :: ERROR :: Loading client for required authentication, please authenticate, then exit the client..." tidal-dl fi if [ ! -d "$downloadPath/incomplete" ]; then mkdir -p $downloadPath/incomplete chmod 777 $downloadPath/incomplete chown abc:abc $downloadPath/incomplete fi TidaldlStatusCheck log "TIDAL :: Upgrade tidal-dl to newer version..." pip install tidal-dl==2022.07.06.1 --no-cache-dir &>/dev/null } TidalClientTest () { log "TIDAL :: tidal-dl client setup verification..." TidaldlStatusCheck tidal-dl -o $downloadPath/incomplete -l "166356219" &>/dev/null downloadCount=$(find $downloadPath/incomplete/ -type f -regex ".*/.*\.\(flac\|opus\|m4a\|mp3\)" | wc -l) if [ $downloadCount -le 0 ]; then if [ -f /config/xdg/.tidal-dl.token.json ]; then rm /config/xdg/.tidal-dl.token.json fi log "TIDAL :: ERROR :: Download failed" log "TIDAL :: ERROR :: You will need to re-authenticate on next script run..." log "TIDAL :: ERROR :: Exiting..." rm -rf $downloadPath/incomplete/* exit else rm -rf $downloadPath/incomplete/* log "TIDAL :: Successfully Verified" fi } TidaldlStatusCheck () { until false do running=no if ps aux | grep "tidal-dl" | grep -v "grep" | read; then running=yes log "STATUS :: TIDAL-DL :: BUSY :: Pausing/waiting for all active tidal-dl tasks to end..." sleep 2 continue fi break done } ImvdbCache () { if [ -z "$artistImvdbSlug" ]; then return fi if [ ! -d "/config/extended/cache/imvdb" ]; then log "$processCount of $lidarrArtistIdsCount :: IMVDB :: $lidarrArtistName :: Creating Cache Folder..." mkdir -p "/config/extended/cache/imvdb" chmod 777 "/config/extended/cache/imvdb" chown abc:abc "/config/extended/cache/imvdb" fi log "$processCount of $lidarrArtistIdsCount :: IMVDB :: $lidarrArtistName :: Caching Records..." if [ ! -f /config/extended/cache/imvdb/$artistImvdbSlug ]; then log "$processCount of $lidarrArtistIdsCount :: IMVDB :: $lidarrArtistName :: Recording Artist Slug into cache" echo -n "$lidarrArtistName" > /config/extended/cache/imvdb/$artistImvdbSlug fi artistImvdbVideoUrls=$(curl -s "https://imvdb.com/n/$artistImvdbSlug" | grep "$artistImvdbSlug" | grep -Eoi ']+>' | grep -Eo 'href="[^\"]+"' | grep -Eo '(http|https)://[^"]+' | grep -i ".com/video/$artistImvdbSlug/" | sed "s%/[0-9]$%%g" | sort -u) artistImvdbVideoUrlsCount=$(echo "$artistImvdbVideoUrls" | wc -l) sleep 0.5 imvdbProcessCount=0 for imvdbVideoUrl in $(echo "$artistImvdbVideoUrls"); do imvdbProcessCount=$(( $imvdbProcessCount + 1 )) imvdbVideoUrlSlug=$(basename "$imvdbVideoUrl") imvdbVideoData="/config/extended/cache/imvdb/$lidarrArtistMusicbrainzId--$imvdbVideoUrlSlug.json" find /config/extended/cache/imvdb -type f -iname "*--[0-9]*[0-9].json" -delete #echo "$imvdbVideoUrl :: $imvdbVideoUrlSlug :: $imvdbVideoId" log "$processCount of $lidarrArtistIdsCount :: IMVDB :: $lidarrArtistName :: $imvdbProcessCount of $artistImvdbVideoUrlsCount :: Caching video data..." if [ -f "$imvdbVideoData" ]; then if [ ! -s "$imvdbVideoData" ]; then rm "$imvdbVideoData" fi fi if [ ! -f "$imvdbVideoData" ]; then count=0 until false; do count=$(( $count + 1 )) #echo "$count" if [ ! -f "$imvdbVideoData" ]; then imvdbVideoId=$(curl -s "$imvdbVideoUrl" | grep "

ID:" | grep -o "[[:digit:]]*") imvdbVideoJsonUrl="https://imvdb.com/api/v1/video/$imvdbVideoId?include=sources,countries,featured,credits,bts,popularity" log "$processCount of $lidarrArtistIdsCount :: IMVDB :: $lidarrArtistName :: $imvdbProcessCount of $artistImvdbVideoUrlsCount :: Downloading Video data" curl -s "$imvdbVideoJsonUrl" -o "$imvdbVideoData" sleep 0.5 fi if [ -f "$imvdbVideoData" ]; then if [ ! -s "$imvdbVideoData" ]; then rm "$imvdbVideoData" if [ $count = 2 ]; then log "$processCount of $lidarrArtistIdsCount :: IMVDB :: $lidarrArtistName :: $imvdbProcessCount of $artistImvdbVideoUrlsCount :: Download Failed, skipping..." break fi elif jq -e . >/dev/null 2>&1 <<<"$(cat "$imvdbVideoData")"; then break else rm "$imvdbVideoData" fi fi done else log "$processCount of $lidarrArtistIdsCount :: IMVDB :: $lidarrArtistName :: $imvdbProcessCount of $artistImvdbVideoUrlsCount :: Video Data already downloaded" fi done } DownloadVideo () { if [ -d "$downloadPath/incomplete" ]; then rm -rf "$downloadPath/incomplete" fi if [ ! -d "$downloadPath/incomplete" ]; then mkdir -p "$downloadPath/incomplete" chmod 777 "$downloadPath/incomplete" chown abc:abc "$downloadPath/incomplete" fi if echo "$1" | grep -i "youtube" | read; then if [ ! -z "$cookiesFile" ]; then yt-dlp --cookies "$cookiesFile" -o "$downloadPath/incomplete/${2}${3}" --embed-subs --sub-lang $youtubeSubtitleLanguage --merge-output-format mkv --remux-video mkv --no-mtime --geo-bypass "$1" &>/dev/null else yt-dlp -o "$downloadPath/incomplete/${2}${3}" --embed-subs --sub-lang $youtubeSubtitleLanguage --merge-output-format mkv --remux-video mkv --no-mtime --geo-bypass "$1" &>/dev/null fi if [ -f "$downloadPath/incomplete/${2}${3}.mkv" ]; then chmod 666 "$downloadPath/incomplete/${2}${3}.mkv" chown abc:abc "$downloadPath/incomplete/${2}${3}.mkv" downloadFailed=false else downloadFailed=true fi fi if echo "$1" | grep -i "tidal" | read; then TidalClientTest sleep 1 TidaldlStatusCheck tidal-dl -o "$downloadPath/incomplete" -l "$1" &>/dev/null find "$downloadPath/incomplete" -type f -exec mv "{}" "$downloadPath/incomplete"/ \; find "$downloadPath/incomplete" -mindepth 1 -type d -exec rm -rf "{}" \; &>/dev/null find "$downloadPath/incomplete" -type f -regex ".*/.*\.\(mkv\|mp4\)" -print0 | while IFS= read -r -d '' video; do file="${video}" filenoext="${file%.*}" filename="$(basename "$video")" extension="${filename##*.}" filenamenoext="${filename%.*}" mv "$file" "$downloadPath/incomplete/${2}${3}.mp4" done if [ -f "$downloadPath/incomplete/${2}${3}.mp4" ]; then chmod 666 "$downloadPath/incomplete/${2}${3}.mp4" chown abc:abc "$downloadPath/incomplete/${2}${3}.mp4" downloadFailed=false else downloadFailed=true fi fi } DownloadThumb () { curl -s "$1" -o "$downloadPath/incomplete/${2}${3}.jpg" chmod 666 "$downloadPath/incomplete/${2}${3}.jpg" chown abc:abc "$downloadPath/incomplete/${2}${3}.jpg" } VideoProcessWithSMA () { find "$downloadPath/incomplete" -type f -regex ".*/.*\.\(mkv\|mp4\)" -print0 | while IFS= read -r -d '' video; do count=$(($count+1)) file="${video}" filenoext="${file%.*}" filename="$(basename "$video")" extension="${filename##*.}" filenamenoext="${filename%.*}" if python3 /usr/local/sma/manual.py --config "/config/extended/scripts/sma.ini" -i "$file" -nt &>/dev/null; then sleep 0.01 log "$processCount of $lidarrArtistIdsCount :: $1 :: $lidarrArtistName :: $2 :: Processed with SMA..." rm /usr/local/sma/config/*log* else log "$processCount of $lidarrArtistIdsCount :: $1 :: $lidarrArtistName :: $2 :: ERROR: SMA Processing Error" rm "$video" log "$processCount of $lidarrArtistIdsCount :: $1 :: $lidarrArtistName :: $2 :: INFO: deleted: $filename" fi done } VideoTagProcess () { find "$downloadPath/incomplete" -type f -regex ".*/.*\.\(mkv\|mp4\)" -print0 | while IFS= read -r -d '' video; do count=$(($count+1)) file="${video}" filenoext="${file%.*}" filename="$(basename "$video")" extension="${filename##*.}" filenamenoext="${filename%.*}" artistGenres="" OLDIFS="$IFS" IFS=$'\n' artistGenres=($(echo $lidarrArtistData | jq -r ".genres[]")) IFS="$OLDIFS" if [ ! -z "$artistGenres" ]; then for genre in ${!artistGenres[@]}; do artistGenre="${artistGenres[$genre]}" OUT=$OUT"$artistGenre / " done genre="${OUT%???}" else genre="" fi mv "$filenoext.mkv" "$filenoext-temp.mkv" log "$processCount of $lidarrArtistIdsCount :: $4 :: $lidarrArtistName :: ${1}${2} $3 :: Tagging file" ffmpeg -y \ -i "$filenoext-temp.mkv" \ -c copy \ -metadata TITLE="${1}" \ -metadata DATE_RELEASE="$3" \ -metadata DATE="$3" \ -metadata YEAR="$3" \ -metadata GENRE="$genre" \ -metadata ARTIST="$lidarrArtistName" \ -metadata ALBUMARTIST="$lidarrArtistName" \ -metadata ENCODED_BY="lidarr-extended" \ -attach "$downloadPath/incomplete/${1}${2}.jpg" -metadata:s:t mimetype=image/jpeg \ "$filenoext.mkv" &>/dev/null rm "$filenoext-temp.mkv" chmod 666 "$filenoext.mkv" chown abc:abc "$filenoext.mkv" done } VideoNfoWriter () { log "$processCount of $lidarrArtistIdsCount :: $7 :: $lidarrArtistName :: ${3} :: Writing NFO" nfo="$downloadPath/incomplete/${1}${2}.nfo" if [ -f "$nfo" ]; then rm "$nfo" fi echo "" >> "$nfo" echo " ${3}${4}" >> "$nfo" echo " " >> "$nfo" echo " " >> "$nfo" echo " " >> "$nfo" artistGenres="" OLDIFS="$IFS" IFS=$'\n' artistGenres=($(echo $lidarrArtistData | jq -r ".genres[]")) IFS="$OLDIFS" if [ ! -z "$artistGenres" ]; then for genre in ${!artistGenres[@]}; do artistGenre="${artistGenres[$genre]}" echo " $artistGenre" >> "$nfo" done fi echo " " >> "$nfo" echo " $6" >> "$nfo" if [ "$5" = "musicbrainz" ]; then OLDIFS="$IFS" IFS=$'\n' for artistName in $(echo "$musicbrainzVideoArtistCreditsNames"); do echo " $artistName" >> "$nfo" done IFS="$OLDIFS" fi if [ "$5" = "imvdb" ]; then echo " $lidarrArtistName" >> "$nfo" for featuredArtistSlug in $(echo "$imvdbVideoFeaturedArtistsSlug"); do if [ -f /config/extended/cache/imvdb/$featuredArtistSlug ]; then featuredArtistName="$(cat /config/extended/cache/imvdb/$featuredArtistSlug)" echo " $featuredArtistName" >> "$nfo" fi done fi echo " " >> "$nfo" echo " $lidarrArtistName" >> "$nfo" echo " $lidarrArtistMusicbrainzId" >> "$nfo" echo " " >> "$nfo" echo " ${1}${2}.jpg" >> "$nfo" echo "" >> "$nfo" chmod 666 "$nfo" chown abc:abc "$nfo" } LidarrTaskStatusCheck () { alerted=no until false do taskCount=$(curl -s "$lidarrUrl/api/v1/command?apikey=${lidarrApiKey}" | jq -r .[].status | grep -v completed | grep -v failed | wc -l) if [ "$taskCount" -ge "1" ]; then if [ "$alerted" = "no" ]; then alerted=yes log "STATUS :: LIDARR BUSY :: Pausing/waiting for all active Lidarr tasks to end..." fi sleep 2 else break fi done } AddFeaturedVideoArtists () { if [ "$addFeaturedVideoArtists" != "true" ]; then log "-----------------------------------------------------------------------------" log "Add Featured Music Video Artists to Lidarr :: DISABLED" log "-----------------------------------------------------------------------------" return fi log "-----------------------------------------------------------------------------" log "Add Featured Music Video Artists to Lidarr :: ENABLED" log "-----------------------------------------------------------------------------" lidarrArtistsData="$(curl -s "$lidarrUrl/api/v1/artist?apikey=${lidarrApiKey}" | jq -r ".[]")" artistImvdbUrl=$(echo $lidarrArtistsData | jq -r '.links[] | select(.name=="imvdb") | .url') videoArtists=$(ls /config/extended/cache/imvdb/ | grep -Ev ".*--.*") videoArtistsCount=$(ls /config/extended/cache/imvdb/ | grep -Ev ".*--.*" | wc -l) if [ "$videoArtistsCount" == "0" ]; then log "$videoArtistsCount Artists found for processing, skipping..." return fi loopCount=0 for slug in $(echo $videoArtists); do loopCount=$(( $loopCount + 1)) artistName="$(cat /config/extended/cache/imvdb/$slug)" if echo "$artistImvdbUrl" | grep -i "^https://imvdb.com/n/${slug}$" | read; then log "$loopCount of $videoArtistsCount :: $artistName :: Already added to Lidarr, skipping..." continue fi log "$loopCount of $videoArtistsCount :: $artistName :: Processing url :: https://imvdb.com/n/$slug" query_data=$(curl -s -A "$agent" "https://musicbrainz.org/ws/2/url?query=url:%22https://imvdb.com/n/$slug%22&fmt=json") count=$(echo "$query_data" | jq -r ".count") if [ "$count" != "0" ]; then musicbrainzArtistId="$(echo "$query_data" | jq -r ".urls[].\"relation-list\"[].relations[].artist.id")" sleep 1 else log "$loopCount of $videoArtistsCount :: $artistName :: ERROR : Musicbrainz ID Not Found, skipping..." continue fi data=$(curl -s "$lidarrUrl/api/v1/search?term=lidarr%3A$musicbrainzArtistId" -H "X-Api-Key: $lidarrApiKey" | jq -r ".[]") artistName="$(echo "$data" | jq -r ".artist.artistName")" foreignId="$(echo "$data" | jq -r ".foreignId")" data=$(curl -s "$lidarrUrl/api/v1/rootFolder" -H "X-Api-Key: $lidarrApiKey" | jq -r ".[]") path="$(echo "$data" | jq -r ".path")" qualityProfileId="$(echo "$data" | jq -r ".defaultQualityProfileId")" metadataProfileId="$(echo "$data" | jq -r ".defaultMetadataProfileId")" data="{ \"artistName\": \"$artistName\", \"foreignArtistId\": \"$foreignId\", \"qualityProfileId\": $qualityProfileId, \"metadataProfileId\": $metadataProfileId, \"monitored\":true, \"monitor\":\"all\", \"rootFolderPath\": \"$path\", \"addOptions\":{\"searchForMissingAlbums\":false} }" if echo "$lidarrArtistIds" | grep "^${musicbrainzArtistId}$" | read; then log "$loopCount of $videoArtistsCount :: $artistName :: Already in Lidarr ($musicbrainzArtistId), skipping..." continue fi log "$loopCount of $videoArtistsCount :: $artistName :: Adding $artistName to Lidarr ($musicbrainzArtistId)..." LidarrTaskStatusCheck lidarrAddArtist=$(curl -s "$lidarrUrl/api/v1/artist" -X POST -H 'Content-Type: application/json' -H "X-Api-Key: $lidarrApiKey" --data-raw "$data") done } Configuration TidalClientSetup AddFeaturedVideoArtists log "-----------------------------------------------------------------------------" log "Finding Videos" log "-----------------------------------------------------------------------------" lidarrArtists=$(wget --timeout=0 -q -O - "$lidarrUrl/api/v1/artist?apikey=$lidarrApiKey" | jq -r .[]) lidarrArtistIds=$(echo $lidarrArtists | jq -r .id) lidarrArtistIdsCount=$(echo "$lidarrArtistIds" | wc -l) processCount=0 for lidarrArtistId in $(echo $lidarrArtistIds); do processCount=$(( $processCount + 1)) lidarrArtistData=$(echo $lidarrArtists | jq -r "select(.id==$lidarrArtistId)") lidarrArtistName=$(echo $lidarrArtistData | jq -r .artistName) lidarrArtistMusicbrainzId=$(echo $lidarrArtistData | jq -r .foreignArtistId) lidarrArtistPath="$(echo "${lidarrArtistData}" | jq -r " .path")" lidarrArtistFolder="$(basename "${lidarrArtistPath}")" lidarrArtistFolder="$(echo "$lidarrArtistFolder" | sed "s/ (.*)$//g" | sed "s/\.$//g")" # Plex Sanitization, remove disambiguation lidarrArtistNameSanitized="$(echo "$lidarrArtistFolder" | sed 's% (.*)$%%g')" artistImvdbUrl=$(echo $lidarrArtistData | jq -r '.links[] | select(.name=="imvdb") | .url') artistImvdbSlug=$(basename "$artistImvdbUrl") downloadedVideoCount=0 if [ "$lidarrArtistName" == "Various Artists" ]; then log "$processCount of $lidarrArtistIdsCount :: $lidarrArtistName :: Skipping, not processed by design..." continue fi if [ -d /config/extended/logs/video/complete ]; then find /config/extended/logs/video/complete -type f -mtime +7 -delete # Remove Files older than 7 days to allow re-processing artist for new videos if [ -f "/config/extended/logs/video/complete/$lidarrArtistFolder" ]; then downloadedVideoCount=$(find "$videoPath/$lidarrArtistFolder" -type f -iname "*.mkv" | wc -l) log "$processCount of $lidarrArtistIdsCount :: $lidarrArtistName :: All $downloadedVideoCount Artist Music Videos previously downloaded, skipping..." continue fi fi CacheMusicbrainzRecords ImvdbCache log "$processCount of $lidarrArtistIdsCount :: MBZDB :: $lidarrArtistName :: Checking records for videos..." musicbrainzArtistVideoRecordings=$(cat "/config/extended/cache/musicbrainz/$lidarrArtistId--$lidarrArtistMusicbrainzId--recordings.json" | jq -r "select(.video==true)") musicbrainzArtistVideoRecordingsCount=$(echo "$musicbrainzArtistVideoRecordings" | jq -r .id | wc -l) log "$processCount of $lidarrArtistIdsCount :: MBZDB :: $lidarrArtistName :: $musicbrainzArtistVideoRecordingsCount videos found..." musicbrainzArtistVideoRecordingsDataWithUrl=$(echo "$musicbrainzArtistVideoRecordings" | jq -r "select(.relations[].url)" | jq -s "." | jq -r "unique | .[] | select(.disambiguation | test(\"official\";\"i\"))") musicbrainzArtistVideoRecordingsDataWithUrlIds=$(echo "$musicbrainzArtistVideoRecordingsDataWithUrl" | jq -r ".id") musicbrainzArtistVideoRecordingsDataWithUrlIdsCount=$(echo "$musicbrainzArtistVideoRecordingsDataWithUrlIds" | wc -l) log "$processCount of $lidarrArtistIdsCount :: MBZDB :: $lidarrArtistName :: $musicbrainzArtistVideoRecordingsDataWithUrlIdsCount \"Official\" videos found with URL..." if [ $musicbrainzArtistVideoRecordingsDataWithUrlIdsCount = 0 ]; then log "$processCount of $lidarrArtistIdsCount :: MBZDB :: $lidarrArtistName :: ERROR :: No vidoes with URLs to process, skipping..." else for musicbrainzVideoId in $(echo "$musicbrainzArtistVideoRecordingsDataWithUrlIds"); do musicbrainzVideoRecordingData=$(echo "$musicbrainzArtistVideoRecordingsDataWithUrl" | jq -r "select(.id==\"$musicbrainzVideoId\")") musicbrainzVideoTitle="$(echo "$musicbrainzVideoRecordingData" | jq -r .title)" musicbrainzVideoTitleClean="$(echo "$musicbrainzVideoTitle" | sed -e "s/[^[:alpha:][:digit:]$^&_+=()'%;{},.@#]/ /g" -e "s/ */ /g" | sed 's/^[.]*//' | sed 's/[.]*$//g' | sed 's/^ *//g' | sed 's/ *$//g')" musicbrainzVideoArtistCredits="$(echo "$musicbrainzVideoRecordingData" | jq -r ".\"artist-credit\"[]")" musicbrainzVideoArtistCreditsNames="$(echo "$musicbrainzVideoArtistCredits" | jq -r ".artist.name")" musicbrainzVideoArtistCreditId="$(echo "$musicbrainzVideoArtistCredits" | jq -r ".artist.id" | head -n1)" musicbrainzVideoDisambiguation="" musicbrainzVideoDisambiguation="$(echo "$musicbrainzVideoRecordingData" | jq -r .disambiguation)" if [ ! -z "$musicbrainzVideoDisambiguation" ]; then musicbrainzVideoDisambiguation=" ($musicbrainzVideoDisambiguation)" musicbrainzVideoDisambiguationClean=" ($(echo "$musicbrainzVideoDisambiguation" | sed -e "s%[^[:alpha:][:digit:]]% %g" -e "s/ */ /g" | sed 's/^[.]*//' | sed 's/[.]*$//g' | sed 's/^ *//g' | sed 's/ *$//g'))" else musicbrainzVideoDisambiguation="" musicbrainzVideoDisambiguationClean="" fi musicbrainzVideoRelations="$(echo "$musicbrainzVideoRecordingData" | jq -r .relations[].url.resource)" if [ $sourcePreference = tidal ]; then if echo "$musicbrainzVideoRelations" | grep -i "tidal" | read; then videoDownloadUrl="$(echo "$musicbrainzVideoRelations" | grep -i "tidal" | head -n1)" else videoDownloadUrl="$(echo "$musicbrainzVideoRelations" | grep -i "youtube" | head -n1)" fi else videoDownloadUrl="$(echo "$musicbrainzVideoRelations" | grep -i "youtube" | head -n1)" fi log "$processCount of $lidarrArtistIdsCount :: MBZDB :: $lidarrArtistName :: ${musicbrainzVideoTitle}${musicbrainzVideoDisambiguation} :: $videoDownloadUrl..." if echo "$musicbrainzVideoDisambiguation" | grep -i "lyric" | read; then plexVideoType="-lyrics" videoDisambiguationTitle=" (lyric)" else plexVideoType="-video" videoDisambiguationTitle="" fi if [ -d "$videoPath/$lidarrArtistFolder" ]; then if [[ -n $(find "$videoPath/$lidarrArtistFolder" -iname "${musicbrainzVideoTitleClean}${plexVideoType}.*") ]]; then log "$processCount of $lidarrArtistIdsCount :: MBZDB :: $lidarrArtistName :: ${musicbrainzVideoTitle}${musicbrainzVideoDisambiguation} :: Previously Downloaded, skipping..." continue fi fi if [ "$musicbrainzVideoArtistCreditId" != "$lidarrArtistMusicbrainzId" ]; then log "$processCount of $lidarrArtistIdsCount :: MBZDB :: $lidarrArtistName :: ${musicbrainzVideoTitle}${musicbrainzVideoDisambiguation} :: First artist does not match album arist, skipping..." continue fi if echo "$videoDownloadUrl" | grep -i "tidal" | read; then videoId="$(echo "$videoDownloadUrl" | grep -o '[[:digit:]]*')" videoData="$(curl -s "https://api.tidal.com/v1/videos/$videoId?countryCode=$tidalCountryCode" -H 'x-tidal-token: CzET4vdadNUFQ5JU' | jq -r)" videoDate="$(echo "$videoData" | jq -r ".releaseDate")" videoYear="${videoDate:0:4}" videoImageId="$(echo "$videoData" | jq -r ".imageId")" videoImageIdFix="$(echo "$videoImageId" | sed "s/-/\//g")" videoThumbnail="https://resources.tidal.com/images/$videoImageIdFix/750x500.jpg" fi if echo "$videoDownloadUrl" | grep -i "youtube" | read; then if find /config -type f -name "cookies.txt" | read; then cookiesFile="$(find /config -type f -name "cookies.txt" | head -n1)" log "$processCount of $lidarrArtistIdsCount :: YT-DLP :: $lidarrArtistName :: Cookies File Found!" videoData="$(yt-dlp --cookies "$cookiesFile" -j "$videoDownloadUrl")" else log "$processCount of $lidarrArtistIdsCount :: YT-DLP :: $lidarrArtistName :: Cookies File Not Found!" cookiesFile="" videoData="$(yt-dlp -j "$videoDownloadUrl")" fi videoThumbnail="$(echo "$videoData" | jq -r .thumbnail)" videoUploadDate="$(echo "$videoData" | jq -r .upload_date)" videoYear="${videoUploadDate:0:4}" fi DownloadVideo "$videoDownloadUrl" "$musicbrainzVideoTitleClean" "$plexVideoType" "MBZDB" if [ "$downloadFailed" = "true" ]; then log "$processCount of $lidarrArtistIdsCount :: MBZDB :: $lidarrArtistName :: ${musicbrainzVideoTitle}${musicbrainzVideoDisambiguation} :: Download failed, skipping..." continue fi DownloadThumb "$videoThumbnail" "$musicbrainzVideoTitleClean" "$plexVideoType" "MBZDB" VideoProcessWithSMA "MBZDB" "$musicbrainzVideoTitle" VideoTagProcess "$musicbrainzVideoTitleClean" "$plexVideoType" "$videoYear" "MBZDB" VideoNfoWriter "$musicbrainzVideoTitleClean" "$plexVideoType" "$musicbrainzVideoTitle" "" "musicbrainz" "$videoYear" "MBZDB" if [ ! -d "$videoPath/$lidarrArtistFolder" ]; then mkdir -p "$videoPath/$lidarrArtistFolder" chmod 777 "$videoPath/$lidarrArtistFolder" chown abc:abc "$videoPath/$lidarrArtistFolder" fi mv $downloadPath/incomplete/* "$videoPath/$lidarrArtistFolder"/ done fi if [ -z "$artistImvdbSlug" ]; then log "$processCount of $lidarrArtistIdsCount :: IMVDB :: $lidarrArtistName :: No IMVDB artist link found, skipping..." continue fi imvdbArtistVideoCount=$(ls /config/extended/cache/imvdb/$lidarrArtistMusicbrainzId--*.json | wc -l) if [ $imvdbArtistVideoCount = 0 ]; then log "$processCount of $lidarrArtistIdsCount :: IMVDB :: $lidarrArtistName :: No videos found, skipping..." continue else log "$processCount of $lidarrArtistIdsCount :: IMVDB :: $lidarrArtistName :: Processing $imvdbArtistVideoCount Videos!" fi find /config/extended/cache/imvdb -type f -empty -delete # delete empty files imvdbProcessCount=0 for imvdbVideoData in $(ls /config/extended/cache/imvdb/$lidarrArtistMusicbrainzId--*.json); do imvdbProcessCount=$(( $imvdbProcessCount + 1 )) imvdbVideoTitle="$(cat "$imvdbVideoData" | jq -r .song_title)" videoTitleClean="$(echo "$imvdbVideoTitle" | sed -e "s/[^[:alpha:][:digit:]$^&_+=()'%;{},.@#]/ /g" -e "s/ */ /g" | sed 's/^[.]*//' | sed 's/[.]*$//g' | sed 's/^ *//g' | sed 's/ *$//g')" imvdbVideoYear="$(cat "$imvdbVideoData" | jq -r .year)" imvdbVideoImage="$(cat "$imvdbVideoData" | jq -r .image.o)" imvdbVideoArtistsSlug="$(cat "$imvdbVideoData" | jq -r .artists[].slug)" echo "$lidarrArtistName" > /config/extended/cache/imvdb/$imvdbVideoArtistsSlug imvdbVideoFeaturedArtistsSlug="$(cat "$imvdbVideoData" | jq -r .featured_artists[].slug)" imvdbVideoYoutubeId="$(cat "$imvdbVideoData" | jq -r ".sources[] | select(.is_primary==true) | select(.source==\"youtube\") | .source_data")" #"/config/extended/cache/musicbrainz/$lidarrArtistId--$lidarrArtistMusicbrainzId--recordings.json" #echo "$imvdbVideoTitle :: $imvdbVideoYear :: $imvdbVideoYoutubeId :: $imvdbVideoArtistsSlug" if [ -z "$imvdbVideoYoutubeId" ]; then continue fi videoDownloadUrl="https://www.youtube.com/watch?v=$imvdbVideoYoutubeId" plexVideoType="-video" if [ ! -z "$imvdbVideoFeaturedArtistsSlug" ]; then for featuredArtistSlug in $(echo "$imvdbVideoFeaturedArtistsSlug"); do if [ -f /config/extended/cache/imvdb/$featuredArtistSlug ]; then featuredArtistName="$(cat /config/extended/cache/imvdb/$featuredArtistSlug)" else query_data=$(curl -s -A "$agent" "https://musicbrainz.org/ws/2/url?query=url:%22https://imvdb.com/n/$featuredArtistSlug%22&fmt=json") count=$(echo "$query_data" | jq -r ".count") if [ "$count" != "0" ]; then featuredArtistName="$(echo "$query_data" | jq -r ".urls[].\"relation-list\"[].relations[].artist.name")" echo -n "$featuredArtistName" > /config/extended/cache/imvdb/$featuredArtistSlug sleep 1 fi fi find /config/extended/cache/imvdb -type f -empty -delete # delete empty files if [ -z "$featuredArtistName" ]; then continue fi done fi if [ -d "$videoPath/$lidarrArtistFolder" ]; then if [[ -n $(find "$videoPath/$lidarrArtistFolder" -iname "${videoTitleClean}${plexVideoType}.*") ]]; then log "$processCount of $lidarrArtistIdsCount :: IMVDB :: $lidarrArtistName :: ${imvdbVideoTitle} :: Previously Downloaded, skipping..." continue fi fi if find /config -type f -name "cookies.txt" | read; then cookiesFile="$(find /config -type f -name "cookies.txt" | head -n1)" log "$processCount of $lidarrArtistIdsCount :: YT-DLP :: $lidarrArtistName :: Cookies File Found!" videoData="$(yt-dlp --cookies "$cookiesFile" -j "$videoDownloadUrl")" else log "$processCount of $lidarrArtistIdsCount :: YT-DLP :: $lidarrArtistName :: Cookies File Not Found!" cookiesFile="" videoData="$(yt-dlp -j "$videoDownloadUrl")" fi videoThumbnail="$imvdbVideoImage" videoUploadDate="$(echo "$videoData" | jq -r .upload_date)" videoYear="${videoUploadDate:0:4}" log "$processCount of $lidarrArtistIdsCount :: IMVDB :: $lidarrArtistName :: ${imvdbVideoTitle} :: $videoDownloadUrl..." DownloadVideo "$videoDownloadUrl" "$videoTitleClean" "$plexVideoType" "IMVDB" if [ "$downloadFailed" = "true" ]; then log "$processCount of $lidarrArtistIdsCount :: IMVDB :: $lidarrArtistName :: ${imvdbVideoTitle} :: Download failed, skipping..." continue fi DownloadThumb "$imvdbVideoImage" "$videoTitleClean" "$plexVideoType" "IMVDB" VideoProcessWithSMA "IMVDB" "$imvdbVideoTitle" VideoTagProcess "$videoTitleClean" "$plexVideoType" "$videoYear" "IMVDB" VideoNfoWriter "$videoTitleClean" "$plexVideoType" "$imvdbVideoTitle" "" "imvdb" "$videoYear" "IMVDB" if [ ! -d "$videoPath/$lidarrArtistFolder" ]; then mkdir -p "$videoPath/$lidarrArtistFolder" chmod 777 "$videoPath/$lidarrArtistFolder" chown abc:abc "$videoPath/$lidarrArtistFolder" fi mv $downloadPath/incomplete/* "$videoPath/$lidarrArtistFolder"/ done if [ ! -d /config/extended/logs/video ]; then mkdir -p /config/extended/logs/video chmod 777 /config/extended/logs/video chown abc:abc /config/extended/logs/video fi if [ ! -d /config/extended/logs/video/complete ]; then mkdir -p /config/extended/logs/video/complete chmod 777 /config/extended/logs/video/complete chown abc:abc /config/extended/logs/video/complete fi touch "/config/extended/logs/video/complete/$lidarrArtistFolder" done #CacheMusicbrainzRecords #ImvdbCache exit