Cap video at 30s + downscale to 480p for 10x faster processing

app.py

@@ -177,6 +177,28 @@ def generate_insights(engagement, hrf_offset=5):
     return insights
 
 
+MAX_DURATION = 30  # Only analyze first 30s — enough for any ad creative
+
+
+def trim_video(input_path: str, max_seconds: int = MAX_DURATION) -> str:
+    """Trim video to max_seconds using ffmpeg. Returns trimmed path."""
+    output_path = input_path.replace(".mp4", "_trimmed.mp4")
+    try:
+        result = subprocess.run(
+            ["ffmpeg", "-y", "-i", input_path, "-t", str(max_seconds),
+             "-c:v", "libx264", "-preset", "ultrafast", "-crf", "28",
+             "-vf", "scale='min(480,iw)':-2",  # cap width at 480px
+             "-c:a", "aac", "-b:a", "64k", output_path],
+            capture_output=True, text=True, timeout=30,
+        )
+        if result.returncode == 0 and Path(output_path).exists():
+            print(f"Trimmed to {max_seconds}s, resized to 480p: {Path(output_path).stat().st_size} bytes")
+            return output_path
+    except Exception as e:
+        print(f"Trim failed ({e}), using original")
+    return input_path
+
+
 def analyze_video(video_url: str, filename: str = "video.mp4") -> str:
     """Main analysis — uses full TRIBE v2 pipeline with WhisperX on GPU."""
     try:
@@ -193,6 +215,9 @@ def analyze_video(video_url: str, filename: str = "video.mp4") -> str:
             video_path = f.name
         print(f"Downloaded {len(resp.content)} bytes")
 
+        # Trim to MAX_DURATION seconds and downscale for faster encoding
+        video_path = trim_video(video_path)
+
         # Build events — try full pipeline (WhisperX + audio + video), fallback to audio-only
         try:
             print("Building events with full pipeline (WhisperX)...")