ReelForge

Drop a raw talking head video. Get back a finished Instagram reel.

Captions, cuts, branding — done. Runs as a macOS menu bar app.

---

What It Does

1. You pick a video from the menu bar app (or drop it in the input/ folder)
2. Groq Whisper transcribes every word with millisecond-accurate timestamps
3. Claude reads the transcript and decides what to cut — pauses, false starts, dead air
4. Remotion renders the final reel with animated word-by-word captions and delivery-based emphasis
5. You get a finished .mp4 auto-named from the content, ready to post

---

Quick Start

Prerequisites

• Node.js 20+
• Rust toolchain (for the Tauri menu bar app)
• ffmpeg + ffprobe installed
• API keys for Groq (Whisper) and Anthropic (Claude)

Setup

git clone https://github.com/phillhendry/reelforge.git
cd reelforge
npm install

# Add your API keys
cp .env.example .env
# Edit .env — add GROQ_API_KEY and ANTHROPIC_API_KEY

Run the Menu Bar App

npm run tauri:dev

Click the tray icon to open the UI. Pick a video, hit render.

Or Run Without Tauri

# Start the web server
npm run server
# Open http://localhost:3100

# Or use the CLI directly
npm run process -- ./input/my-video.mp4

---

Menu Bar App

ReelForge runs as a macOS system tray app built with Tauri:

• Click the menu bar icon to toggle the popover
• No dock icon — stays out of your way
• Spawns the Express server automatically
• Open input/output folders in Finder from the UI
• Build a standalone .app with npm run tauri:build

---

Features

AI Editing

Claude analyses the transcript and produces an edit decision list:

• Cuts pauses >800ms, false starts, repeated phrases
• Keeps natural speech rhythm — single filler words in flowing sentences stay
• Snaps cuts to word boundaries with 80ms/150ms padding so nothing clips
• Fade-to-black ending with 400ms tail after the last word

Smart Captions

Uses Remotion's native @remotion/captions system:

• createTikTokStyleCaptions() for word grouping
• Each caption page gets its own <Sequence> for frame-perfect timing
• Delivery-based emphasis — words spoken slower than median pace or after a pause get a bigger visual pop (larger scale, snappier spring, subtle highlight retention)

Caption Styles

Style	Description
highlight	Accent background highlights the active word
bounce	Active word scales up with a spring animation
typewriter	Words fade in as they're spoken
minimal	Active word is white, others are dimmed

Output Presets

Preset	Dimensions	Use Case
instagramReel	1080x1920 (9:16)	Instagram Reels, TikTok
youtubeShort	1080x1920 (9:16)	YouTube Shorts
instagramSquare	1080x1080 (1:1)	Instagram Feed
landscape	1920x1080 (16:9)	YouTube, LinkedIn

Multi-Job Support

• Queue multiple renders — each gets its own progress bar
• Cancel any active job mid-render
• Auto-watch mode processes files dropped in input/ (max 2 concurrent)
• Toggle auto-watch on/off from the UI
• Clear finished jobs individually or all at once

Content-Based Filenames

Output files are auto-named from the AI analysis summary:

speaker-shares-tips-on-design-systems_2026-03-19T14-30-00.mp4

---

CLI

# Full edit + captions
npm run process -- ./input/raw-take.mp4

# Captions only — no AI editing
npm run process -- ./input/raw-take.mp4 --no-cuts

# Different caption style
npm run process -- ./input/raw-take.mp4 --caption-style=bounce

# YouTube Shorts preset
npm run process -- ./input/raw-take.mp4 --preset=youtubeShort

# Save debug files (transcript + EDL as JSON)
npm run process -- ./input/raw-take.mp4 --debug

---

API

Start the server with npm run server (port 3100).

Method	Endpoint	Description
POST	/render	Start a render (used by the UI)
POST	/render/talking-head	Start a render with full options
POST	/render/carousel	Render carousel slides as a reel
GET	/status/:jobId	Poll job progress
GET	/jobs	List all jobs
POST	/cancel/:jobId	Cancel an active job
DELETE	/jobs	Clear all finished jobs
DELETE	/jobs/:jobId	Clear a single finished job
POST	/auto-watch	Toggle auto-processing { "enabled": true }
GET	/files	List videos in input directory
POST	/open-folder/input	Open input folder in Finder
POST	/open-folder/output	Open output folder in Finder
GET	/health	Health check

---

Architecture

input/video.mp4
       │
       ▼
  ┌──────────┐
  │  Probe   │  ffprobe → dimensions, duration, fps
  └────┬─────┘
       ▼
  ┌──────────────┐
  │  Transcribe  │  Groq Whisper → Remotion Caption[]
  └──────┬───────┘
       ▼
  ┌──────────────┐
  │   Analyse    │  Claude → edit decision list
  └──────┬───────┘
       ▼
  ┌──────────────┐
  │  Pad + Remap │  Pad EDL for playback, remap captions
  └──────┬───────┘
       ▼
  ┌──────────────┐
  │  Emphasis    │  Detect stressed words from delivery
  └──────┬───────┘
       ▼
  ┌──────────────┐
  │   Render     │  Remotion SSR → MP4
  └──────┬───────┘
       ▼
output/content-summary_2026-03-19.mp4

Stack

• Remotion — React-based video rendering with native caption support
• Groq Whisper — Fast cloud transcription with word-level timestamps
• Claude — AI editing decisions (not content generation)
• Express — HTTP server + file watcher
• Tauri — macOS menu bar app wrapper (~5MB binary)

---

Brand Configuration

Set in .env or pass via API:

BRAND_PRIMARY_COLOR=#ffffff
BRAND_ACCENT_COLOR=#E8FF00
BRAND_FONT_FAMILY=Inter

---

Building the App

# Development (hot-reload)
npm run tauri:dev

# Production build
npm run tauri:build

Output:

• src-tauri/target/release/bundle/macos/ReelForge.app
• src-tauri/target/release/bundle/dmg/ReelForge_1.0.0_aarch64.dmg

---

License

MIT. Remotion is free for individuals and companies up to 3 people.

Built by Phill. Blame Claude.