diff --git a/backend/lib/ytdlp.js b/backend/lib/ytdlp.js
index 56e5f9f..ed741a8 100644
--- a/backend/lib/ytdlp.js
+++ b/backend/lib/ytdlp.js
@@ -81,7 +81,7 @@ async function downloadTrack({ jobId, artist, rawArtist, albumArtist, title, alb
   // rawArtist is undefined for non-Japanese tracks, so fall back to mainArtist.
   const ytmArtist    = (rawArtist || mainArtist).replace(/\p{S}/gu, '').trim();
   const searchArtist = ytmArtist;
-  const query = `${ytmArtist} - ${title}`;
+  const query = `${ytmArtist} - ${title}${albumQuerySuffix}`;
 
   const baseArgs = [
     '--extract-audio',
@@ -117,7 +117,7 @@ async function downloadTrack({ jobId, artist, rawArtist, albumArtist, title, alb
 
   const YTM_SONGS_PARAM = 'EgWKAQIIAWoKEAkQBRAKEAMQBA%3D%3D';
   // Use original (raw) artist for YTM URL — better ranking since channel is in original language
-  const ytmQuery = `${rawArtist || mainArtist} - ${title}`;
+  const ytmQuery = `${rawArtist || mainArtist} - ${title}${albumQuerySuffix}`;
   const ytmUrl = `https://music.youtube.com/search?q=${encodeURIComponent(ytmQuery)}&params=${YTM_SONGS_PARAM}`;
   const expectedAudio = path.join(dir, `${fileBase}.${CONFIG.audioFmt}`);
   const hasAudio = () => fs.existsSync(expectedAudio);
@@ -167,6 +167,9 @@ async function downloadTrack({ jobId, artist, rawArtist, albumArtist, title, alb
     proc.on('close', () => resolve(channelExpanded ? [] : urls));
   });
 
+  const escRe = s => s.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
+  const JP_CHARS = /[ぁ-んァ-ン一-龯]/;
+
   const variantFilter = buildVariantFilter(title);
   const titleKws      = buildTitleKeywords(title);
   const variantPart   = variantFilter ? ` & ${variantFilter}` : '';
@@ -174,12 +177,23 @@ async function downloadTrack({ jobId, artist, rawArtist, albumArtist, title, alb
 
   // Soft channel check: first word of original artist name used in early cascade steps to avoid
   // downloading a same-title song by a different artist (e.g. wrong "Moonlight" for N!GHT).
-  // Dropped in later fallback steps so genuine mismatches (label channels etc.) still get found.
-  const escRe = s => s.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
+  // Skip when artistFirstWord is non-ASCII (Japanese kanji/kana): YouTube channel names are
+  // almost never in kanji, so the filter would reject every valid result.
   const artistFirstWord = (rawArtist || mainArtist).split(/[\s,/]+/)[0];
-  const artistChannelPart = artistFirstWord.length >= 3
+  const artistChannelPart = (artistFirstWord.length >= 3 && !JP_CHARS.test(artistFirstWord))
     ? ` & channel ~= (?i)${escRe(artistFirstWord)}` : '';
 
+  // When the title is Japanese (keywords stripped to nothing) OR when we have an album that
+  // provides better search context, include the album's first two words in the search query.
+  // This helps YTM rank the official OST/album track above fan covers and romaji-subtitle uploads.
+  // Only use album words that are ASCII (romaji/English) — Japanese album words don't help YTM ranking.
+  const hasJpTitle = JP_CHARS.test(title);
+  const albumWords = (album && album !== 'Singles' && album !== 'Unknown')
+    ? album.split(/\s+/).filter(w => /^[A-Za-z0-9]/.test(w)).slice(0, 2).join(' ')
+    : '';
+  const albumQuerySuffix = (hasJpTitle || (albumWords && titleKws.length <= 1)) && albumWords
+    ? ' ' + albumWords : '';
+
   const mf = (dur, variant, strict = false) => {
     const durPart = dur ? `duration > ${dur.min} & duration < ${dur.max} & ` : '';
     return ['--match-filter',