diff --git a/CHANGELOG.md b/CHANGELOG.md
index 93b256c..03b3f2b 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -67,6 +67,10 @@ All notable changes to this project are documented here. The format follows
 
 ### Fixed
 
+- An ffmpeg that dies while `materialize_mask_video` is still piping frames now
+  raises `mask ffmpeg failed (rc=N) after K of N frames: <ffmpeg stderr>` instead of
+  a bare `[Errno 32] Broken pipe`, which threw away ffmpeg's reason. A 352 934-frame
+  4K mask rebuild failed this way with nothing to go on.
 - Two objects visible at once no longer collapse onto one track, leaving one of
   them unblurred. Association had no per-frame exclusivity — nothing marked a
   track as already matched — and its center-distance fallback gate was sized by
diff --git a/redactcam/mask.py b/redactcam/mask.py
index ddcf73a..b034a48 100644
--- a/redactcam/mask.py
+++ b/redactcam/mask.py
@@ -110,18 +110,28 @@ def materialize_mask_video(timeline: BlurTimeline, output_path: Path, feather_px
     proc = subprocess.Popen(
         cmd, stdin=subprocess.PIPE, stdout=subprocess.DEVNULL, stderr=subprocess.PIPE
     )
+    sent = 0
     try:
-        for idx in range(n):
-            frame = render_mask_frame(w, h, active_boxes_at(timeline, idx), feather_px)
-            proc.stdin.write(frame.tobytes())
-        proc.stdin.close()
+        try:
+            for idx in range(n):
+                frame = render_mask_frame(w, h, active_boxes_at(timeline, idx), feather_px)
+                proc.stdin.write(frame.tobytes())
+                sent += 1
+            proc.stdin.close()
+        except BrokenPipeError:
+            # ffmpeg exited mid-stream; its exit code and stderr say why, the
+            # broken pipe on our side does not.
+            try:
+                proc.stdin.close()
+            except BrokenPipeError:
+                pass
         rc = proc.wait()
     finally:
         if proc.poll() is None:
             proc.kill()
     if rc != 0:
         err = proc.stderr.read().decode(errors="replace")[-500:] if proc.stderr else ""
-        raise RuntimeError(f"mask ffmpeg failed (rc={rc}): {err}")
+        raise RuntimeError(f"mask ffmpeg failed (rc={rc}) after {sent} of {n} frames: {err}")
     written = _count_frames(output_path)
     if written != n:
         raise RuntimeError(
diff --git a/tests/test_mask.py b/tests/test_mask.py
index 8f2606c..1643d9b 100644
--- a/tests/test_mask.py
+++ b/tests/test_mask.py
@@ -115,6 +115,21 @@ def kill(self):
         bm.materialize_mask_video(_timeline(), tmp_path / "m.mkv")
 
 
+@pytest.mark.skipif(not shutil.which("ffmpeg"), reason="ffmpeg required")
+def test_ffmpeg_dying_mid_write_reports_its_own_error(tmp_path):
+    """An ffmpeg that exits while frames are still being piped surfaces as a broken
+    pipe on the parent's write. The error must carry ffmpeg's exit code and stderr,
+    not a bare "[Errno 32] Broken pipe" that says nothing about why."""
+    out = tmp_path / "mask.mkv"
+    out.mkdir()  # ffmpeg cannot open a directory as its output, so it exits at once
+    # 512x512 gray frames (256 KB) overflow the 64 KB pipe buffer on the first write.
+    tl = _timeline(frame_width=512, frame_height=512, source_frame_count=20)
+    with pytest.raises(RuntimeError, match=r"mask ffmpeg failed \(rc=[1-9]") as exc:
+        bm.materialize_mask_video(tl, out)
+    assert "mask.mkv" in str(exc.value)  # ffmpeg's own message names the output
+    assert "of 20 frames" in str(exc.value)  # says how far the stream got
+
+
 @pytest.mark.skipif(not shutil.which("ffmpeg"), reason="ffmpeg required")
 def test_materialize_writes_exact_frame_count(tmp_path):
     out = bm.materialize_mask_video(_timeline(), tmp_path / "mask.mkv", feather_px=2)