diff --git a/gnommo/cli.py b/gnommo/cli.py
index cb7d269..4602d8c 100644
--- a/gnommo/cli.py
+++ b/gnommo/cli.py
@@ -1266,6 +1266,24 @@ def _resolve_process_cache(project_path: Path, config) -> Optional[Path]:
     return p / project_path.name
 
 
+def _resolve_narration_combined(
+    project_path: Path, videos_dir: Path, config
+) -> Optional[Path]:
+    """Find narration_combined.mov: local → GnommoCache → process_cache."""
+    local = videos_dir / "narration_combined.mov"
+    if local.exists():
+        return local
+    resolved, _ = resolve_with_cache(local, project_path)
+    if resolved.exists():
+        return resolved
+    pc_root = _resolve_process_cache(project_path, config)
+    if pc_root:
+        pc_path = pc_root / "media" / "videos" / "narration_combined.mov"
+        if pc_path.exists():
+            return pc_path
+    return None
+
+
 def cmd_preprocess(
     project_path: Path,
     verbose: bool,
@@ -2501,9 +2519,13 @@ def cmd_render(
     # Load whisper transcription JSON
     # Check for narration_combined in videos.json (new workflow) or multi-segment in config (legacy)
     combined_path = videos_dir / "narration_combined.mov"
-    # Try cache fallback for combined narration
-    resolved_combined, _ = resolve_with_cache(combined_path, project_path)
-    if "narration_combined" in videos and resolved_combined.exists():
+    resolved_combined = _resolve_narration_combined(project_path, videos_dir, config)
+    if resolved_combined and resolved_combined != combined_path:
+        # File lives on external disk — point the VideoSource at the absolute path so
+        # the renderer doesn't re-resolve it via the local (missing) videos_dir.
+        if "narration_combined" in videos:
+            videos["narration_combined"].source_file = str(resolved_combined)
+    if "narration_combined" in videos and resolved_combined and resolved_combined.exists():
         # New workflow: narration_combined was created by 'gnommo concat' and is in videos.json
         # This entry has the correct volume setting from videos.json
         transcript_path = resolved_combined.with_suffix(".transcript.json")
@@ -2789,6 +2811,12 @@ def cmd_transcribe(
         video_id, video_source = result
         video_path = videos_dir / video_source.source_file
 
+        if not video_path.exists() and video_source.source_file == "narration_combined.mov":
+            found = _resolve_narration_combined(project_path, videos_dir, config)
+            if found:
+                video_path = found
+        if not video_path.exists():
+            video_path, _ = resolve_with_cache(video_path, project_path)
         if not video_path.exists():
             print(f"Error: Video not found: {video_path}", file=sys.stderr)
             return 1
@@ -3551,7 +3579,7 @@ def cmd_extract_audio(
     # Handle --combined mode: extract from narration_combined.mov
     if combined:
         videos, videos_dir = parse_videos(project_path, config)
-        combined_path = videos_dir / "narration_combined.mov"
+        combined_path = _resolve_narration_combined(project_path, videos_dir, config) or (videos_dir / "narration_combined.mov")
 
         if not combined_path.exists():
             print(
@@ -3716,7 +3744,7 @@ def cmd_master(
     videos, videos_dir = parse_videos(project_path, config)
 
     # Find narration_combined.mov
-    combined_path = videos_dir / "narration_combined.mov"
+    combined_path = _resolve_narration_combined(project_path, videos_dir, config) or (videos_dir / "narration_combined.mov")
     if not combined_path.exists():
         print(
             f"Error: narration_combined.mov not found at {combined_path}",