portrait-resizer

name	portrait-resizer
description	Convert videos to 9:16 portrait format (1080x1920) for TikTok, YouTube Shorts, Instagram Reels, and Facebook Reels. Supports smart cropping (face-aware), center cropping, and letterboxing. Maintains aspect ratio and quality.
allowed-tools	Bash(ffmpeg:*)
compatibility	Requires FFmpeg and OpenCV (MediaPipe optional for face detection)
metadata	{"version":"1.0","formats":"9:16 portrait"}

Portrait Resizer

This skill converts horizontal videos to vertical 9:16 portrait format suitable for short-form content platforms.

When to Use

• User wants to create TikTok videos from horizontal content
• Converting YouTube videos to YouTube Shorts format
• Creating Instagram Reels from widescreen video
• Preparing content for Facebook Reels
• Converting any 16:9 video to 9:16 format

Supported Platforms

• TikTok: 1080x1920, 9:16, 15-60s
• YouTube Shorts: 1080x1920, 9:16, 15-60s
• Instagram Reels: 1080x1920, 9:16, 15-90s
• Facebook Reels: 1080x1920, 9:16, 15-90s

Available Scripts

scripts/resize_to_portrait.py

Convert video to 9:16 portrait format.

Usage:

python skills/portrait-resizer/scripts/resize_to_portrait.py <video_path> [options]

Options:

• --width: Target width (default: 1080)
• --height: Target height (default: 1920)
• --mode: Resize mode (smart, center, letterbox) - default: smart
• --focus-x: Focus point X coordinate (0-1, smart mode)
• --focus-y: Focus point Y coordinate (0-1, smart mode)
• --output, -o: Output video path (default: <video_path>_portrait.mp4)
• --quality: Quality preset (fast, medium, slow) - default: fast

Examples:

Basic resize to portrait:

python skills/portrait-resizer/scripts/resize_to_portrait.py video.mp4

Resize with smart cropping:

python skills/portrait-resizer/scripts/resize_to_portrait.py video.mp4 --mode smart

Resize with specific focus point:

python skills/portrait-resizer/scripts/resize_to_portrait.py video.mp4 --mode smart --focus-x 0.7 --focus-y 0.5

Center crop:

python skills/portrait-resizer/scripts/resize_to_portrait.py video.mp4 --mode center

Letterbox (padding on top/bottom):

python skills/portrait-resizer/scripts/resize_to_portrait.py video.mp4 --mode letterbox

scripts/batch_resize.py

Resize multiple videos to portrait format.

Usage:

python skills/portrait-resizer/scripts/batch_resize.py --input-dir <dir> [options]

Options:

• --input-dir: Directory with videos to resize
• --output-dir: Output directory (default: ./portrait/)
• All other options from resize_to_portrait.py

Example:

python skills/portrait-resizer/scripts/batch_resize.py --input-dir ./clips/ --output-dir ./portrait_clips/

Resize Modes

Smart Crop (Default)

Intelligently crops to focus on subjects:

• Detects faces (MediaPipe if available, otherwise OpenCV Haar)
• Samples frames across the clip to find dominant face position
• Centers crop on weighted face centroid
• Falls back to center crop if no faces detected

Center Crop

Simple center crop:

• Crops from the center of the video
• May cut off subjects on edges
• Fast and predictable
• Best when subject is always centered

Letterbox

Pillarbox/black bars:

• Maintains full width of video
• Adds padding on top and bottom
• No content loss
• Best for showing full context

Output Specifications

Default Output:

• Resolution: 1080x1920
• Aspect Ratio: 9:16
• Video Codec: H.264 (libx264)
• Audio Codec: AAC
• Container: MP4

Platform-Specific Presets:

presets = {
    'tiktok': {
        'width': 1080,
        'height': 1920,
        'bitrate': '4000k',
        'fps': 30
    },
    'youtube_shorts': {
        'width': 1080,
        'height': 1920,
        'bitrate': '4000k',
        'fps': 30
    },
    'instagram_reels': {
        'width': 1080,
        'height': 1920,
        'bitrate': '4000k',
        'fps': 30
    }
}

Output Format

{
  "success": true,
  "input_path": "video.mp4",
  "output_path": "video_portrait.mp4",
  "input_resolution": {
    "width": 1920,
    "height": 1080
  },
  "output_resolution": {
    "width": 1080,
    "height": 1920
  },
  "mode": "smart",
  "focus_point": {
    "x": 0.65,
    "y": 0.50
  },
  "quality_settings": {
    "preset": "fast",
    "crf": 23
  }
}

Smart Crop Algorithm

1. Face Detection: Uses MediaPipe (if available) or OpenCV Haar Cascades
2. Frame Sampling: Samples frames at a fixed rate for stable focus
3. Weighted Average: Chooses dominant face using size/confidence weighting
4. Fallback: Uses center crop if no faces detected

Integration with Other Skills

After resizing, you can use these skills:

• subtitle-overlay: Add captions to portrait video
• autocut-shorts: Full workflow with all steps

Common Workflow

1. User provides video file
2. Find highlights using highlight-scanner
3. Trim segments using video-trimmer
4. Resize to portrait using this skill
5. Add subtitles using subtitle-overlay
6. Export final clips

Tips

• Resize AFTER trimming (more efficient)
• Use smart crop for talking-head content
• Use center crop for action/sports content
• Use letterbox for cinematic/wide shots
• Test different focus points for best results
• Face detection works best with clear frontal faces

Performance

• Simple crop/resize: ~2-5 seconds per minute
• Smart crop with face detection: ~5-10 seconds per minute
• Batch processing: Parallel processing available

Error Handling

• Small videos: Minimum resolution checks
• Already portrait: Returns as-is or error
• Face detection failure: Falls back to center crop
• MediaPipe unavailable: Automatically uses OpenCV Haar cascade
• Codec issues: Re-encodes with standard codecs

References

• FFmpeg crop filter: https://ffmpeg.org/ffmpeg-filters.html#crop
• FFmpeg scale filter: https://ffmpeg.org/ffmpeg-filters.html#scale
• OpenCV face detection: https://docs.opencv.org/