diff --git a/scripts/slim-embedded-art.py b/scripts/slim-embedded-art.py
index 1f43ea7..bafc363 100755
--- a/scripts/slim-embedded-art.py
+++ b/scripts/slim-embedded-art.py
@@ -39,6 +39,10 @@ from PIL import Image
 ROOT = os.path.expanduser("~/Music")
 EMBED_TARGET = 80_000
 PROCESS_OVER = 100_000
+# No song has 50 KB of lyrics. The frame that set this limit held the complete text of Frank
+# McCourt's "Angela's Ashes" - 1.5 MB of memoir in an Eels track's USLT, courtesy of some tagger
+# malfunction - and AVPlayer pages through ALL of it before the first audio frame, art or not.
+LYRICS_SANE = 50_000
 
 
 def audio_region_hash(path, offset=None):
@@ -120,22 +124,33 @@ def process(path, apply):
         return None
     pics = pictures(tag)
     total = sum(len(p.data) for p in pics)
-    if total <= PROCESS_OVER:
+    junk_lyrics = [f for f in tag.getall("USLT") if len(f.text or "") > LYRICS_SANE]
+    if total <= PROCESS_OVER and not junk_lyrics:
         return None
-    largest = max(pics, key=lambda p: len(p.data))
+    largest = max(pics, key=lambda p: len(p.data)) if pics else None
     if not apply:
-        return f"would slim {total // 1024} KB of art"
+        parts = []
+        if total > PROCESS_OVER: parts.append(f"{total // 1024} KB of art")
+        if junk_lyrics: parts.append(f"{sum(len(f.text) for f in junk_lyrics) // 1024} KB of junk lyrics")
+        return "would slim " + " + ".join(parts)
 
     before = audio_region_hash(path)
     tmp = path + ".slimtmp"
     shutil.copy2(path, tmp)
     try:
-        rescued = rescue_cover(os.path.dirname(path), largest.data)
-        jpeg, note = compress_cover(largest.data)
+        note = "art untouched"
         t = ID3(tmp)
-        t.delall("APIC")
-        t.delall("PIC")
-        t.add(APIC(encoding=0, mime="image/jpeg", type=3, desc="", data=jpeg))
+        if largest is not None and total > PROCESS_OVER:
+            rescued = rescue_cover(os.path.dirname(path), largest.data)
+            jpeg, note = compress_cover(largest.data)
+            t.delall("APIC")
+            t.delall("PIC")
+            t.add(APIC(encoding=0, mime="image/jpeg", type=3, desc="", data=jpeg))
+        else:
+            rescued = False
+            jpeg = b""
+        for f in junk_lyrics:
+            t.delall(f.HashKey)
         v2 = 4 if tag.version >= (2, 4, 0) else 3
         # v1=0 REMOVES any trailing ID3v1 tag: it is a 128-byte legacy appendix fully redundant
         # with the v2 frames, and keeping it in sync is what tripped the audio guard before the
@@ -145,6 +160,8 @@ def process(path, apply):
             raise RuntimeError("audio bytes changed - keeping original")
         os.replace(tmp, path)
         note_out = f"{total // 1024} KB -> {len(jpeg) // 1024} KB ({note})"
+        if junk_lyrics:
+            note_out += f" -{sum(len(f.text) for f in junk_lyrics) // 1024} KB junk lyrics"
         return note_out + (" +rescued cover.jpg" if rescued else "")
     except Exception as e:
         if os.path.exists(tmp):