branded-reel-pipeline

Branded Reel Pipeline

Reusable framework that produces a premium short-form brand reel end to end. It is the premium layer on top of media/short-form-video-production (ideation, hook, retention, captions craft). Use those for the creative craft; use this for the premium branded assembly with real on-brand visuals and live data.

The proven stack (do not deviate without reason)

1. Script — hook-first, retention-structured. Borrow craft from content-creation (Hook→Tension→Payload→CTA) and short-form-video-production. Beat table: t | OST | VO.
2. Visuals + music — higgsfield-generate skill. GPT Image 2 for hero frames (9:16, 2k, brand palette), sonilo_music for the bespoke bed. Frames are premium connective tissue; live screens carry conversion.
3. Voiceover — ElevenLabs eleven_v3, stability 0.30, default voice id cjVigY5qzO86Huf0OWal ("Eric"). Key from environment. Calm-operator delivery.
4. Live-data capture — reference/live-data-capture.py (Playwright). Parameterized: URL + auth method + selector / API path. The differentiator: real dashboards, real numbers count up on screen.
5. Assembly — Remotion with brand-preset tokens (templates/remotion-template/). Parameterized by aspect ratio. Count-ups, curve-draws, glass device frames, the audio-reactive orb. 1080x1920 @ 30fps for 9:16.
6. Render + verify — remotion render → ffprobe confirms duration/resolution/codec.

Input: the brief

A reel is driven by a brief object:

{
  "subject": "Our AI agent that posts daily and tracks every number",
  "brand_preset": "my-brand",
  "format": "9:16",
  "mode": "gated",
  "live_data_source": {
    "url": "https://your-dashboard.example.com/api/analytics?days=30",
    "auth": { "method": "session_token", "mint": "node scripts/mint-session-token.js" },
    "capture": "json",
    "selector": null
  },
  "script": null
}

• subject (required): what the reel is about.
• brand_preset (required): a dir under presets/ matching your brand.
• format: 9:16 (default) | 1:1 | 16:9.
• mode: gated (default) | auto.
• live_data_source (optional): { url, auth, capture: json|screenshot, selector }.
• script (optional): pre-written beat table; if null the pipeline writes one in Stage 1.