Skip to content

Open-source AI podcast clipper. Generate vertical clips with face tracking and burned-in captions. CLI, MCP server, web app.

Track a cut whose face map it scanned itself (#245)

Nika Siradze5c138e4d8571
diff --git a/backend/services/video_processor.py b/backend/services/video_processor.py
index eace9c7..4c9075f 100644
--- a/backend/services/video_processor.py
+++ b/backend/services/video_processor.py
@@ -372,7 +372,8 @@ def crop_to_vertical(
# episode-wide face_map. Global speaker→side mappings are too coarse for
# monologues and mixed-layout edits; they can pin a single-speaker clip
# to the wrong person for the whole render.
- if speakers_in_clip:
+ scanned_here = bool(face_map) and not face_map.get("speaker_mappings")
+ if speakers_in_clip or scanned_here:
result = _track_and_crop(
input_path, output_path,
width, height, target_w, target_h,
@@ -1645,8 +1646,9 @@ def _track_and_crop_inner(
# Merge very short runs (<0.8s) into their neighbors
if len(runs) > 1:
merged_runs = [runs[0]]
- for r in runs[1:]:
- if (r[1] - r[0]) < 0.8:
+ for i, r in enumerate(runs[1:], start=1):
+ shortest = 0.4 if i == len(runs) - 1 else 0.8
+ if (r[1] - r[0]) < shortest:
# Absorb into previous run
merged_runs[-1] = (merged_runs[-1][0], r[1], merged_runs[-1][2])
else:
diff --git a/tests/test_video_processor.py b/tests/test_video_processor.py
index 01bc26d..9563725 100644
--- a/tests/test_video_processor.py
+++ b/tests/test_video_processor.py
@@ -705,7 +705,10 @@ class VideoProcessorTests(unittest.TestCase):
strategy="face",
transcript_words=[{"word": "hi", "start": 0.0, "end": 0.4, "speaker": None}],
clip_start=0.0,
- face_map={"clusters": [{"center_x": 100, "crop_x": 0}], "video_width": 1920},
+ face_map={
+ "clusters": [{"center_x": 100, "crop_x": 0}], "video_width": 1920,
+ "speaker_mappings": {"SPEAKER_00": 0},
+ },
)
self.assertEqual(result, "ok.mp4")
@@ -714,6 +717,45 @@ class VideoProcessorTests(unittest.TestCase):
track_crop.assert_not_called()
runner.assert_called_once()
+ def test_crop_to_vertical_tracks_a_cut_whose_face_map_it_scanned_itself(self):
+ with mock.patch.object(vp, "get_dimensions", return_value=(1920, 1080)), \
+ mock.patch.object(vp, "_use_face_map", return_value="456") as face_map_crop, \
+ mock.patch.object(vp, "_track_and_crop", return_value="tracked.mp4") as track_crop, \
+ mock.patch.object(vp, "_run_ffmpeg_with_fallback", return_value="ok.mp4") as runner:
+ result = vp.crop_to_vertical(
+ input_path="in.mp4",
+ output_path="out.mp4",
+ strategy="speaker",
+ transcript_words=[{"word": "hi", "start": 0.0, "end": 0.4}],
+ clip_start=0.0,
+ face_map={
+ "clusters": [{"center_x": 100, "crop_x": 0}], "video_width": 1920,
+ "speaker_mappings": {},
+ },
+ )
+
+ self.assertEqual(result, "tracked.mp4")
+ track_crop.assert_called_once()
+ face_map_crop.assert_not_called()
+ runner.assert_not_called()
+
+ def test_crop_to_vertical_falls_back_to_a_scanned_face_map_when_tracking_fails(self):
+ with mock.patch.object(vp, "get_dimensions", return_value=(1920, 1080)), \
+ mock.patch.object(vp, "_use_face_map", return_value="456") as face_map_crop, \
+ mock.patch.object(vp, "_track_and_crop", return_value=None), \
+ mock.patch.object(vp, "_run_ffmpeg_with_fallback", return_value="ok.mp4"):
+ result = vp.crop_to_vertical(
+ input_path="in.mp4",
+ output_path="out.mp4",
+ strategy="speaker",
+ transcript_words=[{"word": "hi", "start": 0.0, "end": 0.4}],
+ clip_start=0.0,
+ face_map={"clusters": [{"center_x": 100, "crop_x": 0}], "video_width": 1920},
+ )
+
+ self.assertEqual(result, "ok.mp4")
+ face_map_crop.assert_called_once()
+
if __name__ == "__main__":
unittest.main()