Captions that never cover a face

Who it is for

Anyone who adds captions or headlines to vertical video and is tired of text landing on someone's face.

What it does

• 1 to 3 words per caption page, word-by-word highlight, key words in your accent color
• Big titles with one accent word in a serif italic and a small label above
• Face tracking with a filter for face-like prints on t-shirts
• Automatic placement inside the Reels safe zone, away from faces and from each other
• A final check that reports any face the text covers

In the zip: SKILL.md, scripts, brand.json, fonts, face model, requirements.txt, LICENSE and third-party notices.

Install in Claude Code

1. Unzip viralframe-face-safe-captions.zip into ~/.claude/skills/ to use it in every project, or into .claude/skills/ inside one project.
2. Check that you have viralframe-face-safe-captions/SKILL.md inside that folder.
3. Restart Claude Code and ask in plain words. Claude can install the requirements for you.

Requirements

• ffmpeg
• Python 3.10+ and pip install -r requirements.txt (numpy, opencv-python)
• Node.js 22+ for npx hyperframes
• Optional speech-to-text: faster-whisper or whisperx

Try asking

Add captions to my_reel.mp4: 3 words max per page, highlight "clients" and "strategy" in my accent color, and a title "Your ideal client." from 27.8 to 30 s. Make sure nothing covers my face.

From our test

On our test: 23 caption pages and 1 title rendered, 0 faces covered.

Licence

Code: MIT, Copyright (c) 2026 Viralframe (see LICENSE in the zip). Fonts, models and libraries from others keep their own licences: see THIRD_PARTY_NOTICES.md. Uses HyperFrames by HeyGen (Apache-2.0) as a dependency through npx; it is not included and it is not ours.