Skip to main content

​
Running the Tool

The AI YouTube Shorts Generator supports multiple input methods and execution modes to fit your workflow.

​
Interactive Mode

Run without arguments to enter interactive mode: 
./run.sh
You’ll be prompted to enter a YouTube URL or local file path: 
Enter YouTube video URL or local video file path:

​
Command-Line Mode

Pass the video source directly as an argument: 
./run.sh "https://youtu.be/VIDEO_ID"
The tool automatically detects whether the input is a local file path or YouTube URL by checking if the file exists on disk.

​
Resolution Selection

When downloading from YouTube, you’ll see available video streams with a 5-second auto-select timeout: 
Available video streams:
  0. Resolution: 1080p, Size: 45.2 MB, Type: Adaptive
  1. Resolution: 720p, Size: 28.1 MB, Type: Adaptive
  2. Resolution: 480p, Size: 15.3 MB, Type: Adaptive

Select resolution number (0-2) or wait 5s for auto-select...
Auto-selecting highest quality in 5 seconds...
1

Enter a number

Type 0, 1, or 2 to immediately select that resolution
2

Wait for auto-select

Wait 5 seconds to automatically select the highest quality (1080p)
3

Fallback behavior

Invalid input defaults to highest quality
The resolution selection uses a 5-second timeout implemented with select.select() in Components/YoutubeDownloader.py:32.

​
Interactive Approval Workflow

After the AI selects a highlight segment, you can review and approve it: 
============================================================
SELECTED SEGMENT DETAILS:
Time: 68s - 187s (119s duration)
============================================================

Options:
  [Enter/y] Approve and continue
  [r] Regenerate selection
  [n] Cancel

Auto-approving in 15 seconds if no input...

​
Approval Options

Press Enter or y to accept the selected segment and proceed with video generation.
Press r to have the AI select a different 2-minute segment. You can regenerate multiple times until you’re satisfied.
Regeneration calls the GPT-4o-mini API again with temperature=1.0 for varied results.
Press n to cancel the entire operation and exit.
Wait 15 seconds without input to automatically approve. Perfect for automation and batch processing.
The 15-second timeout is implemented with select.select() in main.py:122.

​
Processing Workflow

The tool follows this step-by-step process:
1

Download or Load Video

  • YouTube videos are downloaded to the videos/ directory
  • Local files are used directly from the provided path
  • Session ID is generated for tracking: session_id = str(uuid.uuid4())[:8]
2

Extract Audio

Converts video to WAV format for transcription:
audio_{session_id}.wav
3

Transcribe with Whisper

GPU-accelerated transcription extracts timestamped text segments. Approximately 30 seconds for a 5-minute video.
4

AI Highlight Selection

GPT-4o-mini analyzes the transcription to find the most engaging 2-minute segment based on criteria:
  • Interesting or useful content
  • Surprising or controversial topics
  • Thought-provoking ideas
  • Complete sentences only
5

Interactive Approval

Review the selected segment with 15-second auto-approve timeout (skipped in batch mode).
6

Extract Clip

Creates temporary clip from the selected timeframe:
temp_clip_{session_id}.mp4
7

Smart Cropping

Converts to 9:16 vertical format:
  • Face videos: Static face-centered crop (no jerky movement)
  • Screen recordings: Half-width display with smooth motion tracking (1 shift/second max)
Output: temp_cropped_{session_id}.mp4
8

Add Subtitles

Burns Franklin Gothic captions with blue text (#2699ff) and black outline into the video:
temp_subtitled_{session_id}.mp4
Font size scales dynamically: int(video.h * 0.065)
  • 1080p → 70px
  • 720p → 47px
9

Combine Audio

Merges audio track with the final video using FFmpeg.
10

Cleanup

Automatically removes all temporary files:
  • audio_{session_id}.wav
  • temp_clip_{session_id}.mp4
  • temp_cropped_{session_id}.mp4
  • temp_subtitled_{session_id}.mp4

​
Output Files

Final videos are saved with slugified, descriptive filenames: 
{video-title}_{session-id}_short.mp4

​
Filename Components

The video title is cleaned and normalized:
  • Converted to lowercase
  • Spaces and underscores replaced with hyphens
  • Invalid characters removed: < > : " / \ | ? * [ ]
  • Multiple consecutive hyphens collapsed
  • Trimmed to 80 characters max
Example: "My Awesome Video!" → "my-awesome-video"Implementation in main.py:46-58:
def clean_filename(title):
    cleaned = title.lower()
    cleaned = re.sub(r'[<>:"/\\|?*\[\]]', '', cleaned)
    cleaned = re.sub(r'[\s_]+', '-', cleaned)
    cleaned = re.sub(r'-+', '-', cleaned)
    cleaned = cleaned.strip('-')
    return cleaned[:80]

​
Example Outputs

how-to-build-a-saas-in-30-days_3f8a9b12_short.mp4
ai-coding-tips-for-beginners_7c2d4e56_short.mp4
my-startup-journey-lessons-learned_9a1b3c5d_short.mp4
All output files are saved in the current working directory unless running in Docker (then in /app/output).

​
Session IDs and Concurrent Execution

Each run generates a unique session ID for isolated processing: 
./run.sh "https://youtu.be/VIDEO1" &
./run.sh "https://youtu.be/VIDEO2" &
./run.sh "/path/to/video3.mp4" &
All instances share the same videos/ directory for downloaded content, but temporary files are isolated by session ID.
Benefits:
  • No file conflicts: Each session has unique temp files
  • Parallel processing: Run multiple videos simultaneously
  • Output traceability: Session ID in filename shows which run created each video
  • Easy debugging: Logs can be correlated by session ID
See main.py:13 for session ID generation: 
session_id = str(uuid.uuid4())[:8]
print(f"Session ID: {session_id}")

Build docs developers (and LLMs) love

Get started for freeTalk to us