Documentation Index
Fetch the complete documentation index at: https://mintlify.com/SamurAIGPT/AI-Youtube-Shorts-Generator/llms.txt
Use this file to discover all available pages before exploring further.
The Face Crop component converts horizontal videos to vertical 9:16 format with intelligent face-centered or motion-tracked cropping.
Functions
crop_to_vertical
Crops a video to vertical 9:16 aspect ratio with intelligent face detection or motion tracking.
crop_to_vertical(input_video_path: str, output_video_path: str) -> None
Path to the input video file to crop
Path where the cropped vertical video will be saved
Cropping Modes
The function automatically detects the best cropping strategy:
Face Detection Mode
Motion Tracking Mode
When: A face is detected in the first 30 framesBehavior:
- Analyzes first 30 frames to detect faces
- Uses median face position for stability
- Applies 60px right offset to prevent cutoff
- Uses static crop position (no tracking)
- Best for: Talking head videos, interviews, vlogs
# Face detected - using face-centered crop
avg_face_x = int(sorted(face_positions)[len(face_positions) // 2])
avg_face_x += 60 # Offset to prevent right-side cutoff
x_start = max(0, min(avg_face_x - vertical_width // 2,
original_width - vertical_width))
When: No face detected (likely screen recording)Behavior:
- Scales video to show 67% of original width
- Tracks motion using optical flow (Farneback algorithm)
- Updates crop position every 1 second
- Smooth tracking (90% previous, 10% new position)
- Best for: Screen recordings, gameplay, tutorials
# No face detected - using motion tracking
target_display_width = original_width * 0.67
scale = vertical_width / target_display_width
# Update tracking once per second
if frame_count % update_interval == 0:
# Calculate optical flow
flow = cv2.calcOpticalFlowFarneback(...)
# Track motion and update crop position
Output Specifications
- Aspect Ratio: 9:16 (vertical/portrait)
- Height: Preserves original video height
- Width: Calculated as
height * 9/16
- Dimensions: Always even numbers (codec requirement)
- Codec: XVID (Windows) or mp4v (Linux/Mac)
- FPS: Preserves original frame rate
from Components.FaceCrop import crop_to_vertical
# Crop video to vertical format
input_video = "horizontal_video.mp4"
output_video = "vertical_video.mp4"
crop_to_vertical(input_video, output_video)
# Output:
# Detecting face position for static crop...
# ✓ Face detected. Using face-centered crop at x=450
# Output dimensions: 608x1080
# Processed 100/3000 frames
# Processed 200/3000 frames
# ...
# Cropping complete. Processed 3000 frames -> vertical_video.mp4
Features
Face Detection
- Uses OpenCV’s Haar Cascade classifier
- Samples first 30 frames for stable detection
- Selects largest face if multiple detected
- Uses median position to avoid outliers
- 60px right offset prevents edge cutoff
face_cascade = cv2.CascadeClassifier(
cv2.data.haarcascades + 'haarcascade_frontalface_default.xml'
)
faces = face_cascade.detectMultiScale(
gray,
scaleFactor=1.1,
minNeighbors=8,
minSize=(30, 30)
)
Motion Tracking
- Optical flow using Farneback algorithm
- Updates every 1 second (once per
fps frames)
- Focuses on significant motion (threshold: 2.0)
- Smooth tracking with 90/10 interpolation
- Prevents abrupt camera movements
flow = cv2.calcOpticalFlowFarneback(
prev_gray, curr_gray, None,
0.5, 3, 15, 3, 5, 1.2, 0
)
magnitude = np.sqrt(flow[..., 0]**2 + flow[..., 1]**2)
# Smooth tracking
smoothed_x = int(0.90 * smoothed_x + 0.10 * target_x)
Scaling and Letterboxing
For screen recordings:
- Scales to show 67% of original width
- Adds letterboxing if needed to maintain aspect ratio
- Uses LANCZOS4 interpolation for quality
The function sets a global Fps variable used by combine_videos() to ensure frame rate consistency.
combine_videos
Combines audio from one video with visuals from another, used to add original audio to cropped videos.
combine_videos(video_with_audio: str, video_without_audio: str,
output_filename: str) -> None
Path to video file containing the audio track to use
Path to video file containing the visual content to use
Path where the combined video will be saved
Output Specifications
- Video Codec: libx264 (H.264)
- Audio Codec: AAC
- Preset: medium (balanced speed/quality)
- Bitrate: 3000k
- FPS: Uses global
Fps variable from crop_to_vertical()
from Components.FaceCrop import crop_to_vertical, combine_videos
# Step 1: Crop to vertical (removes audio)
crop_to_vertical("original.mp4", "cropped_no_audio.mp4")
# Step 2: Add original audio back
combine_videos(
video_with_audio="original.mp4",
video_without_audio="cropped_no_audio.mp4",
output_filename="final_with_audio.mp4"
)
# Output:
# Combined video saved successfully as final_with_audio.mp4
Error Handling
try:
crop_to_vertical(input_path, output_path)
except Exception as e:
print(f"Cropping failed: {e}")
try:
combine_videos(audio_source, video_source, output)
except Exception as e:
print(f"Error combining video and audio: {str(e)}")
Requires OpenCV (cv2), NumPy, and MoviePy. The cropping process can be CPU/GPU intensive for long videos.
Progress Tracking
Both functions provide progress updates:
# crop_to_vertical progress
if frame_count % 100 == 0:
print(f"Processed {frame_count}/{total_frames} frames")
# combine_videos progress
# MoviePy shows built-in progress bar during write_videofile()
- Face Detection: Only samples first 30 frames for speed
- Motion Tracking: Updates once per second (not every frame)
- Smooth Tracking: Uses 90/10 interpolation to reduce jitter
- Codec Selection: Platform-specific codec for compatibility
- Dimension Alignment: Ensures even dimensions for codec compatibility
import platform
if platform.system() == 'Windows':
fourcc = cv2.VideoWriter_fourcc(*'XVID')
else:
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
- Windows: Uses XVID codec
- Linux/Mac: Uses mp4v codec
