All skills

Reel edit: split-screen tutorial

Use for SPLIT-SCREEN TUTORIAL reels - a talking-head clip + a screen recording of a demo. White "futuristic popout" style (white bg, dark text, ora…

Websites, reels and sales27 min read

From Alexander Ahdab's library.

Reel Editor (green-glass "Ambra" style)

Turn ONE raw vertical talking-head clip into a finished 9:16 reel: the speaker in the bottom half, animated green graphics/schemas in the top half, karaoke one-word subtitles, music, and by default 1–2 AI B-roll cutaways generated with the Higgsfield MCP (see "AI B-roll via Higgsfield"). Style validated against reference reels (jesserurka, emilsystems on Instagram).

All commands below assume the skill lives at .claude/skills/reel-edit-split-screen (this folder). <skill> = that folder. Work inside one project folder per reel, e.g. ~/reels/<name>/.

Stack

  • scripts/transcribe.py for TRANSCRIPTION (ElevenLabs Scribe, word-level timestamps — tightest word boundaries, best for the one-word karaoke captions). This is the DEFAULT transcriber — the API key is preconfigured in this skill's .env, no setup needed. Alternatives with the SAME CLI + output shape: scripts/transcribe_groq.py (Groq API whisper-large-v3, needs GROQ_API_KEY, near-free and very fast) and scripts/transcribe_local.py (free local Whisper, offline, slowest). Whisper-based timings (groq/local) are slightly looser than Scribe — expect an occasional extra caption-sync fix round. For the CUT use scripts/cutjoin.py (NOT a generic renderer: see ERRORS).
  • hyperframes (via npx hyperframes render) to render the HTML/GSAP cards. npx fetches it on first use; needs Node.js.
  • ffmpeg for composition + music. Local ffmpeg usually has no libass → subtitles are baked with PIL (captions.py), NOT with subtitles=.
  • Caption font: Montserrat (bundled at assets/fonts/Montserrat-VariableFont_wght.ttf, weight 900). captions.py finds it automatically.
  • Bundled assets: assets/logos/ (whatsapp, claude, openai, github), assets/bg-music.m4a (default background track — swap it for your own with MUSIC=/path/to/track.m4a).
  • Higgsfield MCP for AI-generated B-roll (Seedance 2.0, image-to-video / text-to-video). Deferred tools — load via ToolSearch before calling (search "higgsfield generate_video media_upload jobs_wait", or select: the mcp__…__generate_video, media_upload, media_confirm, jobs_wait, job_display, balance tools of the Higgsfield server; the server-name prefix varies per session).

Free, for your email

Unlock the full skill and get it by email

The rest opens right here, and a copy comes to your inbox too. One email unlocks every guide and every skill. Use your real address.

Unlocks: SCREEN-SHARE WINDOWS (this skill's addition, validated on the codex-plugin reel) · Card animation doctrine (one-prompt-perfect defaults, Aug 2026) · Screen zoom rules (punch-in, not ken-burns — user feedback Aug 2026) · Dual-source input (.mov audio + camera video) · Pipeline (run in order, inside one project folder, e.g. `~/reels/<name>/`) · DEFAULT LOOK: white "futuristic popout" (this skill's identity) · Style (fixed, except the accent color) · Sound effects (default ON since the yoom reel — user asked for them)

No spam. Reply "unsubscribe" any time and I delete your address.