Send feedbacks on X: @adidshaft

mkreel

Creator-grade CLI for clipping YouTube moments into polished vertical reels with optional burned captions.

clip it crop it caption it ship it

mkreel takes a YouTube URL, lets you choose a segment, optionally turns it into a 1080x1920 reel, retimes English subtitles in native TypeScript, burns them into the video, and exports a ready-to-post MP4.

If you have codex or claude installed, mkreel can also act like a smart creator sidecar:

• mkreel <url> --smart suggests a clip, mode, and caption preset before the normal guided flow
• mkreel suggest <url> scans the transcript for three strong clip ideas
• mkreel package <file.mp4> generates a creator-friendly publish pack

Core rendering, validation, file handling, subtitle retiming, and final ffmpeg work all stay inside mkreel. AI only suggests safe values that mkreel validates before use.

Why It Feels Good

• Guided interactive flow with smart defaults
• Fully scriptable flags when you want automation
• Optional AI sidecar for clip ideas and publish packs
• First-run dependency bootstrapping for ffmpeg, ffprobe, and yt-dlp
• Ready-made caption presets for Shorts, Reels, and TikTok-style exports
• Clean subtitle retiming, clamping, overlap cleanup, and renumbering
• Blur-fill vertical reel layout with centered foreground video
• Predictable output naming and debug-friendly temp workspace handling

Quick Start

1. Install

Requires Node.js >=20.10.

npm install -g mkreel

Or run it without a global install:

npx mkreel --help

2. Run It

mkreel https://www.youtube.com/watch?v=YOUR_VIDEO_ID

Optional AI-assisted entry point:

mkreel https://www.youtube.com/watch?v=YOUR_VIDEO_ID --smart

If you prefer npx:

npx mkreel https://www.youtube.com/watch?v=YOUR_VIDEO_ID

Standalone AI helper commands:

mkreel suggest https://www.youtube.com/watch?v=YOUR_VIDEO_ID
mkreel package clip.mp4

3. Answer A Few Prompts

The interactive flow is intentionally compact:

1. Validate the URL
2. Ask for clip start
3. Ask for clip end
4. Ask for output mode
5. Ask whether to burn subtitles
6. Ask for a caption preset if subtitles are enabled
7. Show a concise execution summary
8. Run the pipeline
9. Print the final file path

Fast Mental Model

A Good First Command

mkreel "https://youtu.be/dQw4w9WgXcQ" \
  --start 8:43 \
  --end 11:38 \
  --mode reel \
  --subs burn \
  --subtitle-position bottom \
  --subtitle-size large \
  --subtitle-style creator \
  --output final.mp4

AI-assisted guided run:

mkreel "https://youtu.be/dQw4w9WgXcQ" --smart

Transcript-driven clip ideas:

mkreel suggest "https://youtu.be/dQw4w9WgXcQ" --ai auto

Creator publish pack for a finished clip:

mkreel package "./my-finished-clip.mp4" \
  --ai codex \
  --context "Founder interview clip about consistency and audience growth"

JSON output for automation:

mkreel suggest "<url>" --json
mkreel package "./my-finished-clip.mp4" --json

Smart Workflows

--smart

mkreel <url> --smart keeps the normal guided experience, but first tries to:

1. fetch video metadata
2. read English subtitle text with timestamps
3. ask the selected AI provider for:
   • a suggested clip range
   • reel vs original
   • a supported caption preset
   • a short reason

If the suggestion looks good, you can accept it. If not, mkreel continues with the regular manual prompts and uses the suggestion as a starting point.

If AI is unavailable, invalid, or times out, mkreel quietly falls back with a message like:

Smart suggestions weren't available, so mkreel will continue with the regular flow.

suggest

mkreel suggest <url> returns three transcript-grounded clip ideas with:

• start
• end
• label
• reason
• confidence
• captionPreset
• mode

Human mode prints a creator-friendly list. --json prints structured JSON only.

package

mkreel package <file.mp4> generates a publish pack for a finished clip:

• 5 short title ideas
• 3 thumbnail text ideas
• 1 social caption
• 5 to 10 hashtags
• 2 to 3 short hook lines

If there is matching sidecar text near the file, mkreel uses it. You can also add extra context with --context.

If you just want to preview the resolved plan without downloading or exporting:

mkreel "https://youtu.be/dQw4w9WgXcQ" \
  --start 8:43 \
  --end 11:38 \
  --mode reel \
  --subs burn \
  --dry-run

Caption Presets

Interactive caption presets are designed so you usually do not need manual tuning.

Long captions are wrapped automatically, and mkreel can reduce caption size slightly when needed to keep text inside the frame.

Core Flags

What Happens On First Run

mkreel checks for working system binaries first. If they are missing, it provisions managed versions of:

• ffmpeg
• ffprobe
• yt-dlp

Managed tools are cached for reuse. On macOS, that typically lands under:

~/Library/Caches/mkreel/bin/

This setup is quiet by default and only shows raw details when --debug is enabled.

AI Provider Detection And Fallback

AI is optional. The direct clipping/rendering path does not require codex or claude.

When you use --smart, suggest, or package, mkreel checks available providers in this order:

1. codex
2. claude

For a provider to be considered ready, mkreel verifies that:

• the CLI is installed
• its non-interactive structured-output mode is available
• the user is signed in

If --ai auto is used, mkreel picks the first ready provider. If a provider fails during the request, mkreel can fall through to the next available provider in auto mode.

All AI responses are parsed as JSON, validated with zod, normalized, and checked again against mkreel's supported values. AI cannot inject raw ffmpeg flags, shell fragments, or arbitrary rendering directives.

Reel Mode

When --mode reel is selected, mkreel exports a 1080x1920 MP4 with:

• blurred background fill
• centered foreground video
• preserved audio

The default vertical layout is based on:

[0:v]split=2[bgsrc][fgsrc];
[bgsrc]scale=1080:1920:force_original_aspect_ratio=increase,crop=1080:1920,boxblur=20:10[bg];
[fgsrc]scale=1080:-2:force_original_aspect_ratio=decrease[fg];
[bg][fg]overlay=(W-w)/2:(H-h)/2

Subtitle Behavior

If subtitles are enabled, mkreel:

1. Detects English manual captions first
2. Falls back to automatic English captions when available
3. Downloads subtitles as SRT
4. Shifts timestamps backward so the selected clip starts at 00:00:00
5. Clamps negative times safely
6. Removes cue overlap so captions stay readable
7. Drops invalid cues
8. Renumbers cues from 1
9. Burns subtitles into the final export with a styled ASS track

Default styling includes white text, black outline, bold creator captions by default, and preset-specific outline tuning.

Output Naming

Generated filenames are deterministic and based on:

• video title
• clip start
• clip end
• output mode
• subtitle status

Example:

video-title-08m43s-11m38s-reel-subbed.mp4

If the path already exists, mkreel automatically adds a numeric suffix.

Troubleshooting

Tool setup failed

Run with:

mkreel <url> --debug

No subtitles available

Interactive runs offer to continue without subtitles.

For non-interactive runs:

--subs skip

End time is before start time

mkreel validates the range before running the pipeline and exits with a clear message.

A run failed mid-pipeline

On failures, mkreel preserves the temp workspace path so you can inspect:

• downloaded media
• subtitle files
• intermediate outputs

Development

npm install
npm run typecheck
npm run build
npm test
npm run dev -- --help

Local Development Usage

npm run dev -- "https://www.youtube.com/watch?v=YOUR_VIDEO_ID"

Or build and run the compiled CLI:

npm run build
node dist/cli.js --help