Skip to content

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

fix: improve video quality across pipeline

Nika Siradze7ad236014edb
diff --git a/backend/services/encoder.py b/backend/services/encoder.py
index 6bce7f4..5dddc5f 100644
--- a/backend/services/encoder.py
+++ b/backend/services/encoder.py
@@ -105,7 +105,7 @@ def _get_encoder_flags(encoder: str) -> list[str]:
flags = {
"h264_videotoolbox": [
"-c:v", "h264_videotoolbox",
- "-b:v", "12M", # High bitrate for crisp 1080x1920
+ "-b:v", "20M", # High bitrate for crisp 1080x1920
"-profile:v", "high",
"-allow_sw", "1", # Allow software fallback
],
diff --git a/backend/services/video_processor.py b/backend/services/video_processor.py
index de28eb7..3f801f0 100644
--- a/backend/services/video_processor.py
+++ b/backend/services/video_processor.py
@@ -91,12 +91,17 @@ def cut_segment(
"""
duration = end_second - start_second
+ # Use lossless copy for this intermediate step — the crop/caption
+ # pipeline will re-encode with the proper encoder anyway.
+ # -ss before -i seeks to nearest keyframe, -noaccurate_seek is fast.
+ # We still re-encode video to get frame-accurate start, but use
+ # high quality CRF to minimize generation loss.
cmd = [
"ffmpeg", "-y",
"-ss", str(start_second),
"-i", input_path,
"-t", str(duration),
- "-c:v", "libx264", "-crf", "18", "-preset", "ultrafast",
+ "-c:v", "libx264", "-crf", "12", "-preset", "fast",
"-c:a", "aac", "-b:a", "192k",
"-avoid_negative_ts", "make_zero",
output_path,