ai-video-generator-claude（即梦视频 skill · 起号/无脸号专用）

一个跑 28 万粉账号的老外，把自己在用的 10 个即梦视频 skill 开源了，git clone 进 ~/.claude/skills 就能用。

你说一句“做个 10 秒无脸财经号开头“，它把镜头、灯光、音效、每秒时间点全写好，粘进 Higgsfield 直接出片。

rediumvex/ai-video-generator-claude

Source: https://github.com/rediumvex/ai-video-generator-claude

AI Video Generator — Claude Skill

10 prompt engineering skills that turn Claude into a cinematic video director for Seedance 2.0 on Higgsfield. You describe an idea — the skill architects a 15-25 line production-grade prompt with exact timing, camera angles, lighting setup, and sound design. Paste into Higgsfield, get studio-quality video.

This is the actual skill set I use to generate every AI video for @theromanknox (280K+ followers). Not templates. Not “try this prompt.” The real production pipeline — copied straight from my ~/.claude/skills/.

What it does

You give it a concept — “10s product launch video, dark UI, minimal” — and it gives you back:

• Complete Seedance 2.0 prompt — 15-25 lines, paste-ready, with timestamps for every second
• 2-second hook engineering — 10+ scroll-stopping patterns per style (pattern interrupt, curiosity gap, dopamine trigger)
• Camera movement direction — 15+ techniques with exact phrasing Seedance 2.0 understands (dolly, orbit, whip pan, snap zoom, rack focus)
• Lighting & mood presets — three-point, neon contrast, silhouette, golden hour, practical lights
• Sound design layers — ambient + impact + music + voice, synced to visual beats
• Platform optimization — TikTok, Instagram Reels, YouTube Shorts, LinkedIn — aspect ratio, safe zones, loop potential
• Material references — tells you exactly where to use @image1, @video1, @audio1

One idea in, finished video prompt out.

The 10 skills

For views

For business

For brand

Why “write me a cool video” doesn’t work

Most people write 2-line prompts and get generic output. Seedance 2.0 is a precision engine — it responds to specific camera language, lighting terminology, and timing breakdowns. Vague input = vague output.

These skills solve that by encoding three things into every prompt:

1. The physics of attention — the first 2 seconds determine if 50% of viewers bounce or 90% stay. Every prompt opens with a hook engineered for retention.
2. Cinematic vocabulary — Seedance 2.0 understands “dolly forward 3 feet, f/1.4, rack focus from foreground to background” better than “make it look cinematic.”
3. Sound-visual sync — every audio element (bass hit, whoosh, ASMR texture, silence) is timestamped to a visual beat. Sound is 50% of the hook.

The result: prompts that produce studio-quality video instead of generic AI slop.

Install

# Clone into your Claude skills folder
cd ~/.claude/skills
git clone https://github.com/rediumvex/ai-video-generator-claude.git ai-video-generator

Or if you use gstack:

cd ~/.claude/skills/gstack
git clone https://github.com/rediumvex/ai-video-generator-claude.git ai-video-generator

Restart Claude Code. Each skill inside skills/ is now available.

You can also use the interactive installer:

cd ai-video-generator
python install.py          # Pick which skills to install
python install.py --all    # Install everything
python install.py --list   # Preview available skills

Usage

"I need a 10s product launch video for my SaaS app, dark mode UI, minimal aesthetic"
→ SaaS Launch skill generates a complete prompt with camera, lighting, sound, timing

"Create a 6s before-after for my client's brand redesign"
→ Before-After skill builds the contrast arc: cold→warm lighting, gray→saturated color, chaos→order

"Give me a viral TikTok hook for a faceless finance channel"
→ Viral Hook + Faceless Channel skills combine for a scroll-stopping 2-second opener

"Turn this podcast clip into a cinematic video for Reels"
→ Podcast Visual skill transforms the audio into visual storytelling with kinetic typography

What’s inside each skill

Every skill is a deep engineering framework, not a prompt template. Each one contains:

• 2-second hook framework — 10+ attention-capture patterns specific to the style
• Timeline architecture — beat-by-beat breakdown for every second of a 4-15s clip
• Camera movement encyclopedia — 15+ techniques with exact Seedance 2.0 phrasing
• Lighting & mood presets — full setups (key, fill, rim, practical, color temp)
• Sound design stack — 4-layer audio architecture synced to visual beats
• Material reference system — where and how to use @image1 @video1 @audio1
• Platform optimization — TikTok, Reels, Shorts, YouTube tuning
• 5+ production-ready example prompts — 15-25 lines each, ready to paste

Who built this

Roman Knox — I build AI products that make money. Not demos.

• 📸 Instagram — @theromanknox (280K+ followers)
• 🎓 Knox Community on Skool — n8n library, Claude skills, cold outreach systems, weekly live sessions

My other Claude skills

• 📝 SEO Blog Writer — human-sounding blog posts that rank on Google and beat AI detectors
• 📱 Social Media Caption Generator — one prompt → captions for every platform

License

MIT — Copyright (c) 2026 Roman Knox

A ⭐ on the repo is the best thank-you. It helps other builders find these skills.