Skip to main content
This guide covers how to customize various aspects of the AI YouTube Shorts Generator by modifying the source code.

​
Subtitle Styling

Subtitles are configured in Components/Subtitles.py. The TextClip configuration (lines 50-59) controls all visual aspects of the captions.

​
Font Configuration

txt_clip = TextClip(
    text,
    fontsize=dynamic_fontsize,
    color='#2699ff',
    stroke_color='black',
    stroke_width=2,
    font='Franklin-Gothic',
    method='caption',
    size=(video.w - 100, None)
)
  • font: 'Franklin-Gothic' - Change to any system font
  • fontsize: Dynamic, calculated as int(video.h * 0.065) (~6.5% of video height)
    • 1080p videos → 70px
    • 720p videos → 47px
  • color: '#2699ff' - Hex color code for text (default: blue)
  • stroke_color: 'black' - Outline color
  • stroke_width: 2 - Outline thickness in pixels

​
List Available Fonts

To see which fonts are available on your system: 
convert -list font | grep -i "font:"

​
Positioning and Margins

Subtitle position is set at line 62:
Components/Subtitles.py:62
txt_clip = txt_clip.set_position(('center', video.h - txt_clip.h - 100))
  • Horizontal: 'center' - Centered on screen
  • Vertical: video.h - txt_clip.h - 100 - 100px from bottom
  • Width margins: video.w - 100 - 50px margin on each side (line 58)

​
Highlight Selection Criteria

The AI highlight selection logic is in Components/LanguageTasks.py. You can customize what the AI considers “interesting.”

​
System Prompt

Edit the system variable (lines 27-43) to change selection criteria: 
system = """
The input contains a timestamped transcription of a video.
Select a 2-minute segment from the transcription that contains something interesting, useful, surprising, controversial, or thought-provoking.
The selected text should contain only complete sentences.
Do not cut the sentences in the middle.
The selected text should form a complete thought.
Return a JSON object with the following structure:
## Output 
[{{
    start: "Start time of the segment in seconds (number)",
    content: "The transcribed text from the selected segment (clean text only, NO timestamps)",
    end: "End time of the segment in seconds (number)"
}}]

## Input
{Transcription}
"""
Modify keywords like “interesting, useful, surprising, controversial, or thought-provoking” to focus on specific content types (e.g., “funny”, “educational”, “dramatic”).

​
Model and Temperature

Configure the OpenAI model in the GetHighlight function (lines 56-60):
Components/LanguageTasks.py:56-60
llm = ChatOpenAI(
    model="gpt-4o-mini",  # Cost-effective model
    temperature=1.0,
    api_key = api_key
)
  • model: "gpt-4o-mini" - Available models:
    • gpt-4o-mini - Fast and cost-effective (default)
    • gpt-4o - More capable, higher cost
    • gpt-3.5-turbo - Fastest, lowest cost
  • temperature: 1.0 - Controls randomness (0.0 = deterministic, 2.0 = very creative)
    • Lower (0.3-0.7): More consistent, predictable selections
    • Higher (1.0-1.5): More varied, creative selections

​
Motion Tracking Behavior

Motion tracking for screen recordings is configured in Components/FaceCrop.py.

​
Update Frequency

Control how often the crop position updates (lines 99-101):
Components/FaceCrop.py:99-101
if use_motion_tracking:
    update_interval = int(fps)  # Update once per second
    print(f"Motion tracking: updating every {update_interval} frames (~1 shift/second)")
Default is int(fps) (1 update per second). Lower values create smoother but potentially jerkier tracking. Higher values are more stable but less responsive.

​
Smoothing Factor

Adjust position smoothing at line 136:
Components/FaceCrop.py:136
# Smooth tracking (90% previous, 10% new)
smoothed_x = int(0.90 * smoothed_x + 0.10 * target_x)
  • 0.90 / 0.10: Default - Very smooth, gradual movement
  • 0.80 / 0.20: Faster response, slightly less smooth
  • 0.95 / 0.05: Extremely smooth, slower response
  • Formula: (previous_weight * old_position) + (new_weight * new_position)

​
Motion Threshold

Set minimum motion to trigger tracking (line 123):
Components/FaceCrop.py:123
motion_threshold = 2.0
Higher values (3.0-5.0) ignore subtle movements. Lower values (1.0-2.0) track more sensitively. Default 2.0 works well for most screen recordings.

​
Face Detection Sensitivity

Face detection parameters are in Components/FaceCrop.py at line 40:
Components/FaceCrop.py:40
faces = face_cascade.detectMultiScale(gray, scaleFactor=1.1, minNeighbors=8, minSize=(30, 30))
1

scaleFactor: 1.1

Image scale reduction at each detection pass. Smaller values (1.05) detect more faces but slower. Larger values (1.3) are faster but may miss faces.
2

minNeighbors: 8

Minimum neighbors required for detection. Higher values reduce false positives:
  • 5-7: More sensitive, may detect non-faces
  • 8-10: Balanced (default: 8)
  • 11-15: Very strict, may miss some faces
3

minSize: (30, 30)

Minimum face size in pixels. Increase for videos with larger faces or to ignore distant faces.

​
Video Quality Settings

Video encoding parameters are set in two locations:

​
Subtitle Output Quality

In Components/Subtitles.py (lines 73-80):
Components/Subtitles.py:73-80
final_video.write_videofile(
    output_video,
    codec='libx264',
    audio_codec='aac',
    fps=video.fps,
    preset='medium',
    bitrate='3000k'
)

​
Final Video Quality

In Components/FaceCrop.py (line 194):
Components/FaceCrop.py:194
combined_clip.write_videofile(output_filename, codec='libx264', audio_codec='aac', fps=Fps, preset='medium', bitrate='3000k')
Bitrate (bitrate='3000k'):
  • 2000k - Lower quality, smaller files
  • 3000k - Balanced (default)
  • 5000k - High quality, larger files
  • 8000k+ - Very high quality for 1080p
Preset (preset='medium'):
  • ultrafast - Fastest encoding, largest files
  • fast - Quick encoding, larger files
  • medium - Balanced (default)
  • slow - Slower encoding, better compression
  • veryslow - Best compression, slowest encoding
For batch processing, use preset='fast'. For final production, use preset='slow' with higher bitrate.

​
Example: Custom Configuration

Here’s a complete example of customizing multiple parameters:
Components/FaceCrop.py
# More responsive motion tracking
update_interval = int(fps / 2)  # Update twice per second
motion_threshold = 1.5  # Lower threshold
smoothed_x = int(0.70 * smoothed_x + 0.30 * target_x)  # 70/30 smoothing
After making changes to any Python files, restart the application for changes to take effect. No reinstallation of packages is needed.

Build docs developers (and LLMs) love

Get started for freeTalk to us