ai-video-creation v1.0.0A

Immutable. This exact content is served forever at /api/v1/blob/61f6587e6107ddf5.

---
name: ai-video-creation
description: >-
  AI-powered video creation for marketing — step-by-step workflows for
  HeyGen (AI spokespersons), Synthesia (AI avatars), Runway (generative
  video), Captions (AI editing), Opus Clip (AI clipping), ElevenLabs
  (AI voiceover), and Descript (AI editing). Use when creating demo
  videos, personalized outreach videos, social video content, ads, or
  scaling video production without a video team. Triggers on: "AI video",
  "HeyGen", "Synthesia", "Runway AI", "AI spokesperson video", "AI ads".
license: MIT
compatibility: Claude Code, Codex, GitHub Copilot, Cursor, Gemini CLI, OpenCode, Goose, Hermes, Jesse, Windsurf, Zed
metadata:
  version: "1.0.0"
  author: LeadMagic
  category: creative
  tags: [ai-video, heygen, synthesia, runway, captions, elevenlabs, video-marketing, ai-avatar]
  related_skills: [vibe-marketing, ai-content-creation, ad-creative-strategy, social-media-strategy, content-distribution, vibe-coding]
  frameworks:
    - "Gary Vaynerchuk — Volume of content wins. AI makes volume possible."
    - "HeyGen/Synthesia — AI spokesperson technology"
    - "Runway — Generative AI video"
    - "Opus Clip — AI-powered video clipping"
    - "Descript — AI-first video editing"
---

# AI Video Creation

## Overview

Video is the highest-effort, highest-impact marketing format. Traditionally,
video required a camera, lighting, editing skills, and hours of post-production.
AI flips this: you can now generate professional video from text alone. AI
spokespersons deliver personalized demos. AI editors cut and caption in minutes.
Generative AI creates B-roll from prompts. This skill covers every AI video
workflow relevant to B2B GTM — from personalized outreach to social content
at scale.

## Authoritative Foundations

- **Gary Vaynerchuk — Volume of content wins. AI makes volume possible.** — Volume of content wins. AI makes volume possible.
- **HeyGen/Synthesia — AI spokesperson technology** — AI spokesperson technology
- **Runway — Generative AI video** — Generative AI video
- **Opus Clip — AI-powered video clipping** — AI-powered video clipping
- **Descript — AI-first video editing** — AI-first video editing

## When to Use

Trigger phrases: "AI video creation", "HeyGen tutorial", "Synthesia demo",
"AI spokesperson video", "Runway for marketing", "AI video ad", "Opus Clip",
"AI video editing", "ElevenLabs voiceover", "personalized video at scale",
"AI video content"

## The AI Video Stack

| Tool | What It Does | Best For | Cost |
|---|---|---|---|
| **HeyGen** | AI spokesperson, avatar video, personalized video | Personalized demos, outreach videos | $29-89/mo |
| **Synthesia** | AI avatar video, 140+ languages | Training, onboarding, multi-language | $29-89/mo |
| **Runway** | Generative video, text-to-video, B-roll | B-roll footage, visual effects | $15-95/mo |
| **Captions** | AI video editing, auto-captions, eye contact fix | Talking-head social video | $16-25/mo |
| **Opus Clip** | AI clipping — long → short clips | Podcast/sermon clips for social | $19-29/mo |
| **Descript** | AI video editor, transcript-based editing | Full video production | $19-29/mo |
| **ElevenLabs** | AI voiceover, text-to-speech, voice cloning | Narration, ads, podcasts | $5-99/mo |
| **Pika** | Generative video, animate images | Quick video generation | $10-35/mo |

## Step-by-Step Workflows

### Workflow 1: AI Spokesperson Demo Video (1 hour)

**Use case:** Product demo, explainer video, "welcome" video for onboarding.

```
STEP 1 — Script (15 min, Claude):
"Write a 90-second product demo script for [product]. 
Audience: [ICP]. They care about [top 3 concerns].
Structure:
- 0:00-0:10: Hook — the problem they face
- 0:10-0:30: How [product] solves it (3 key features, outcomes-based)
- 0:30-0:60: Screen recording + walkthrough (you record this)
- 0:60-0:80: Social proof + results ('X teams use us, achieving Y')
- 0:80-0:90: CTA — 'Start free trial at [URL]'

Platform: spoken by AI avatar. Keep sentences short — under 15 words."

STEP 2 — Voiceover (5 min, ElevenLabs):
- Paste script into ElevenLabs Text-to-Speech
- Pick a natural voice (Adam, Bella, or similar — test 3+)
- Generate. Listen. Re-generate with pacing adjustments if needed.
- Download .mp3

STEP 3 — Avatar Video (10 min, HeyGen / Synthesia):
- Select AI avatar (matching your target audience's expectations)
- Upload script text
- Upload ElevenLabs voiceover (or use built-in TTS)
- Add: background (office, studio, or branded), on-screen text overlays
- Generate

STEP 4 — Screen Recording (15 min):
- Record product walkthrough (Loom, OBS, or QuickTime)
- Edit to match script timing

STEP 5 — Combine (15 min, Descript / CapCut):
- Layer 1: AI avatar video (talking head)
- Layer 2: Screen recording (when script references product)
- Layer 3: B-roll from Runway (generated visuals)
- Add: captions (auto-generated), background music (low, instrumental)
- Add: intro/outro branding
- Export: 1080p, 16:9
```

### Workflow 2: Personalized Outreach Video at Scale (30 min per video)

**Use case:** "Hey [first name], I made this quick video for you..."
Personalized videos get 3-5x higher reply rates than text-only outreach.

```
STEP 1 — Template Script (10 min, Claude):
"Write a 45-second personalized outreach video script. 
Personalization tokens: [first_name], [company], [pain_point].
Structure:
- 'Hey [first_name] — [relevant observation about their company/role]'
- 'The reason I'm reaching out: [pain_point is common in their industry]'
- 'We built [product] to solve exactly that. Here's what it does in 10 sec'
- 'I'd love 10 minutes to show you how it works. [Calendly] in comments'"

STEP 2 — Record Template Video (once):
- Record the template script with all personalization tokens clearly marked
- OR: Use HeyGen to generate an AI avatar video template

STEP 3 — Personalize (per prospect, 5 min):
- Research prospect: company, role, recent news, common connection
- Customize the observation: "I saw you just raised your Series A — congrats!"
- Record personalized intro (10 sec on Loom) or use HeyGen's personalization

STEP 4 — Edit & Send (5 min, Descript / Captions):
- Replace template intro with personalized intro
- Add prospect's name as on-screen text in opening frame
- Export. Upload. Send via email or LinkedIn DM.
```

### Workflow 3: Social Short-Form Video at Scale (1 hour for 5 videos)

**Use case:** LinkedIn video, TikTok, YouTube Shorts, Instagram Reels.

```
STEP 1 — Batch Scripts (15 min, Claude):
"Write 5 short-form video scripts (60 seconds each) about [topics].
Structure:
- Hook (0-3 sec): bold claim, surprising stat, or provocative question
- Body (3-50 sec): one key insight explained with example
- CTA (50-60 sec): 'Follow for more [topic] insights'

Topics:
1. [topic]
2. [topic]
..."

STEP 2 — Batch Voiceover (10 min, ElevenLabs):
- All 5 scripts → ElevenLabs. Same voice.
- Generate as a batch.
- Download .mp3 files.

STEP 3 — Batch B-Roll (15 min, Runway):
- Generate 3-5 short clips per script (prompt: B-roll concept)
- OR: Use stock footage (Pexels, Pixabay — free)

STEP 4 — Batch Edit (20 min, Captions / Descript):
- Template: voiceover + B-roll + captions + branding
- Apply template to all 5 videos
- Auto-generate captions (Captions does this natively)
- Add: semi-transparent overlay at bottom with your handle
- Add: captions with word-by-word highlighting (95%+ watch time increase)

STEP 5 — Schedule:
- Upload all 5 to scheduling tool (Buffer, Hootsuite, native schedulers)
- Post 1/day for a workweek
```

### Workflow 4: AI Video Ad Creative (2 hours for 5 variants)

```
STEP 1 — Strategy (15 min, Claude):
"Design a video ad strategy for [product] on [LinkedIn / Meta / TikTok].
Audience: [ICP]. Budget: $X. Goal: [demos / signups / leads].
Generate: 5 ad concepts with hooks, scripts, and visual direction."

STEP 2 — Scripts (15 min, Claude):
"Write 5 video ad scripts. Each: 15-30 seconds.
Variant 1: Problem/solution (hook: pain → product solves)
Variant 2: Social proof ('Join X teams...')
Variant 3: Demo teaser (show product, build curiosity)
Variant 4: Founder story ('I built this because...')
Variant 5: Objection buster ('Common belief: X. Reality: Y')"

STEP 3 — Production (60 min):
- HeyGen: Generate AI avatar delivering scripts 1, 2, 4
- Loom: Record screen for variant 3
- Runway: Generate B-roll for all variants

STEP 4 — Edit (30 min, Captions):
- Add captions (non-negotiable — 85% of social video watched on mute)
- Add branding (logo, colors, end card)
- Add background music (low, instrumental)

STEP 5 — Launch & Kill:
- Launch all 5. $50-100/day each.
- After 48 hours: kill bottom 3. Scale top 2.
- After 1 week: if CPA too high, iterate creative. Repeat.
```

### Workflow 5: Podcast/Webinar Clips (30 min)

```
STEP 1 — Upload to Opus Clip or Descript:
- Upload your podcast episode (30-60 min)
- AI auto-identifies "viral moments" — high-energy, quotable segments

STEP 2 — Review & Select (5 min):
- Opus Clip suggests 5-10 clips. Pick the best 3-5.
- Each: 30-90 seconds

STEP 3 — Polish (15 min):
- Captions: auto-generated, reviewed for accuracy (names, terms)
- Add: intro frame with episode title
- Add: end card with "Full episode → [link]"
- Adjust: start/end trim points

STEP 4 — Publish (10 min):
- 1 clip/day on LinkedIn, X/Twitter, TikTok, Instagram
- Link to full episode in comments/bio
```

## AI Video Quality Standards

**What's production-ready:**
- AI spokesperson videos with real voiceover (ElevenLabs or human)
- Screen recordings with AI narration
- Social clips from longer content
- AI-generated B-roll (abstract concepts, data visualization support)

**What needs human touch:**
- Customer testimonials (real people > AI avatars for trust)
- Complex product demos (AI can't anticipate all questions)
- Brand films (emotional storytelling needs real footage)
- Executive communications (your face > AI face for trust)

## Output Format

```
AI VIDEO PRODUCTION PLAN

Video Type: [demo / outreach / social / ad / clip]
Platform: [LinkedIn / YouTube / TikTok / email]
Target Length: [seconds]

SCRIPT: [Claude-generated script]

PRODUCTION STACK:
- Voiceover: [ElevenLabs / human / HeyGen TTS]
- Avatar: [HeyGen / Synthesia / real human]
- B-Roll: [Runway / stock / screen recording]
- Editing: [Descript / Captions / CapCut]
- Captions: [auto / reviewed]
- Music: [source / none]

PERSONALIZATION (if applicable):
- Tokens: [first_name, company, pain_point]
- Custom intro: [recorded / AI-generated]
```

## Implementation Checklist

- [ ] Script under 90 seconds (attention falls off a cliff after 90s)
- [ ] Hook in first 3 seconds (the most important 3 seconds of any video)
- [ ] Captions on every video (85% of social video watched without sound)
- [ ] Voiceover quality checked (natural pacing, correct pronunciation)
- [ ] AI avatar matches audience expectations (professional for enterprise, casual for SMB)
- [ ] Background music low enough to not compete with voice (-18dB to -24dB)
- [ ] End card with CTA (URL, follow, subscribe)
- [ ] File size optimized for platform (<100MB for email, <1GB for social)
- [ ] Thumbnail generated (Canva — face + text overlay)

## Quality Check

Before delivering, verify:

- [ ] Output matches the user's stated request
- [ ] Named frameworks or sources are reflected in the recommendation
- [ ] The deliverable is specific enough for an agent to execute
- [ ] Any assumptions, risks, or dependencies are explicit
- [ ] No unsupported claims, invented facts, or private/internal references are included

## Common Pitfalls

1. **AI avatar uncanny valley.** Poor lip-sync, dead eyes, robotic gestures =
   viewers click away in 2 seconds. Fix: ElevenLabs voiceover with natural
   pacing. Test with 5 people: "Does this feel real?"

2. **No captions.** 85% of social video is watched without sound. No captions
   = 85% of audience doesn't get the message. Fix: Captions on every video.
   Word-by-word highlighting increases watch time 95%+.

3. **Script too long.** 90 seconds is the ceiling for attention on social.
   15-30 seconds for ads. 45-60 seconds for demos. Fix: Cut your script in
   half. Then cut it again. Every word must earn its place.

4. **Generative B-roll that doesn't match.** Runway generated "a team
   collaborating in a modern office" shows 6-fingered people at impossible
   desks. Fix: Use stock footage or screen recordings for concrete scenes.
   Use Runway for abstract concepts, backgrounds, transitions.

5. **Personalized video at scale that feels mass-produced.** "Hey [first_name]"
   with an AI avatar that clearly isn't you. Fix: Record the personalized
   intro yourself (10 sec on Loom). AI handles the rest. The human touch
   is worth the 10 seconds.

6. **Ignoring platform specs.** LinkedIn: 16:9 or 1:1, under 10 min. TikTok:
   9:16 vertical, under 3 min. YouTube Shorts: 9:16, under 60 sec. Fix:
   Export in the correct aspect ratio and duration for each platform.

## Execution Artifacts

- `references/framework-notes.md` — Named frameworks and reference tables
- `templates/output-template.md` — Deliverable shell for agent output
- `scripts/check-output.py` — Lightweight deliverable validator

## Related Skills

- `vibe-marketing` — Full AI marketing stack
- `ai-content-creation` — AI for text-based content
- `ad-creative-strategy` — Creative formats per platform, testing
- `social-media-strategy` — Per-platform posting and engagement
- `content-distribution` — Distribution of video content across channels
- `vibe-coding` — AI dev tools for building video landing pages