Kai Video Production

Video Production Pipeline

Kai root note: knowledge/, harness/, and scripts/ paths in this skill live in the Kai install, not the user's project. Resolve them against the first ancestor directory of this SKILL.md that contains a knowledge/ folder (the Kai plugin root, ~/.claude/kai, or the kai-cmo-harness repo). MARKETING.md, memory/, and any output files live in the current project. If a referenced scripts/ command is not available in this install, say so, skip it, and continue with the file-based guidance — never fabricate its output.

Complete video production from concept to final MP4. Combines two capabilities:

1. Script generation - Platform-optimized scripts (TikTok, YouTube, Reels)
2. Video production - AI-powered rendering with voiceovers, music, transitions

Prerequisites Check

Required:

• Node.js 18+ (for Remotion)
• FFmpeg (for video encoding)

Recommended for AI features:

• Modal account ($30/month free compute) OR
• RunPod account (pay-per-second)
• ElevenLabs API key (optional, for premium voices)

Run /video-setup for interactive configuration.

Part 1: Script Generation (existing kai-video skill)

Phase 0: Load Product Context

Check if MARKETING.md exists in the project root (same directory as CLAUDE.md, README.md, package.json).

If it exists: Read it — skip product discovery questions.

If it does NOT exist: Auto-explore the codebase to create it in the project root (next to CLAUDE.md). Do NOT ask the user what the product is. Read CLAUDE.md, README.md, PROJECT.md, package.json, landing pages, and any project files. Search for email/ad/analytics config. Then create MARKETING.md using the template from /kai-email-system. Present draft to user for confirmation.

Phase 1: Video Discovery

Read from MARKETING.md. Only ask about things not covered there:

1. Topic/product — what's the video about?
2. Platform(s) — TikTok, YouTube Shorts, Reels, YouTube long-form?
3. Format — talking head, screen recording, b-roll, animation, slides?
4. Length — short-form (15-60s) or long-form (5-15min)?
5. Goal — awareness, education, conversion, engagement?
6. Production — script only, or full video render?

Phase 2: Script Production

Load these before writing:

• knowledge/playbooks/video-content-creation.md
• knowledge/channels/tiktok-algorithm.md (if TikTok)
• knowledge/channels/youtube.md (if YouTube)
• knowledge/channels/instagram.md (if Reels)

Use hook formulas, quality gates from existing kai-video skill.

Output scripts to workspace/video/scripts/

Part 2: Video Production (new capability)

Production Options

1. Script only - Stop after Phase 2, user records manually
2. Automated production - Continue to render pipeline
3. Hybrid - Generate some assets (voiceover, slides), user provides demos

Video Templates

Available templates:

• product-demo - Marketing videos (title, problem, solution, demo, CTA)
• explainer - Educational content (title, overview, sections, recap)
• demo-walkthrough - Screen recordings with narration
• social-short - 15-60s vertical video for TikTok/Reels/Shorts
• testimonial - Customer quotes + visuals
• announcement - Product launches, releases, news

Template structure:

projects/{name}/
├── project.json          # Project state, scenes, assets
├── VOICEOVER-SCRIPT.md   # Full narration script
├── src/
│   ├── Root.tsx          # Remotion composition
│   ├── scenes/           # Scene components
│   └── config/           # Timing, brand, assets
├── public/
│   ├── audio/            # Voiceovers, music, SFX
│   ├── demos/            # Screen recordings
│   ├── images/           # Screenshots, graphics
│   └── videos/           # External video clips
└── CLAUDE.md             # Auto-generated status

Production Pipeline

Preflight Variant Lab

Before asset creation or rendering, generate lightweight variants and score them as a dry run:

• 3 hook variants with platform, expected first-frame visual, and viewer promise
• 2 structure variants for scene order or pacing
• 2 CTA variants matched to the campaign goal
• 1 risk note per variant for unsupported claims, brand mismatch, or approval needs

Write results to workspace/video/projects/{name}/PREFLIGHT-VARIANTS.md.

Use this scoring table:

Pick one winner for production and keep rejected variants in the same file as the kill list. User approval is required before moving from variant lab to full render.

Step 1: Scene Planning

Decompose script into scenes:

{
  "scenes": [
    {
      "id": 1,
      "type": "title",
      "duration": 5,
      "visual": "slide",
      "narration": "Introducing KaiCalls...",
      "status": "ready"
    },
    {
      "id": 2,
      "type": "demo",
      "duration": 20,
      "visual": "demos/call-flow.mp4",
      "narration": "Watch how it handles incoming calls...",
      "status": "asset-needed"
    }
  ]
}

Scene types:

• title - Branded opening (slide)
• overview - Key points (slide with bullets)
• demo - Screen recording (Playwright or external)
• split-demo - Side-by-side comparison
• problem - Pain point visualization
• solution - How product solves it
• feature - Specific feature highlight
• stats - Numbers, metrics (animated)
• testimonial - Customer quote + photo
• cta - Call to action (slide)
• credits - Team, thank you

Step 2: Asset Creation

For each scene requiring assets:

Slide assets (auto-generated):

• Title cards with brand colors/fonts
• Bullet point slides
• Stat visualization
• CTA screens

Demo assets (user/AI created):

• Option 1: Browser recording with Playwright

  /record-demo
  # Opens browser, user performs actions, saves MP4
  

• Option 2: External screen recording
  • User provides MP4/MOV
  • Moved to public/demos/
• Option 3: AI-generated demo
  • Script actions → Playwright automation → recording

  /record-demo --script demo-script.md --auto
  

Image assets:

• AI generation - FLUX.2 Klein for backgrounds, graphics

  python tools/flux2.py --prompt "Tech background" --cloud modal
  

• AI editing - Style transfer, backgrounds, effects

  python tools/image_edit.py --input photo.jpg --style cyberpunk
  

• Screenshots - Playwright capture
• External - User-provided images

Step 3: Audio Production

Voiceover generation:

Option 1: AI voiceover (Qwen3-TTS, free)

python tools/voiceover.py --provider qwen3 \\
  --speaker Ryan --script VOICEOVER-SCRIPT.md \\
  --scene-dir public/audio/scenes --json

9 speakers available:

• Ryan, Brad, Ava, Lily (male/female, neutral/warm/energetic)
• Emily, Sam, Alex, Kevin, Zoe

Option 2: ElevenLabs (premium)

python tools/voiceover.py --provider elevenlabs \\
  --voice-id {ID} --script VOICEOVER-SCRIPT.md

Option 3: Voice cloning (custom brand voice)

/voice-clone
# Records sample, creates cloned voice, saves to brand

Music generation:

Option 1: ACE-Step (free, precise control)

python tools/music_gen.py --preset corporate-bg \\
  --duration 120 --bpm 90 --key "D Minor"

8 scene presets:

• corporate-bg, upbeat-intro, dramatic-reveal, ambient-subtle
• tension-build, inspirational-montage, tech-minimal, celebration-end

Option 2: ElevenLabs music

python tools/music.py --prompt "Upbeat corporate" --duration 120

Sound effects:

python tools/sfx.py --preset whoosh  # Transitions
python tools/sfx.py --preset pop     # UI interactions
python tools/sfx.py --preset success # Completion

Step 4: Scene Review

Launch Remotion Studio for visual verification:

cd projects/{name}
npm run studio  # Opens localhost:3000

Review checklist per scene:

• Visual renders correctly
• Timing matches narration
• Transitions smooth
• Text readable
• Brand colors accurate
• Audio synced

Iterative refinement:

• Adjust timing in config.ts
• Tweak styles in scene components
• Replace assets if needed
• Re-sync audio

Step 5: Design Polish

For slide-based scenes:

Load knowledge/frameworks/design/frontend-design.md and refine:

• Typography hierarchy
• Color contrast
• Animation timing
• Layout composition
• Visual hierarchy

Run design review:

/design --scene {scene-id}
# Opens focused design session with frontend-design skill

Step 6: Rendering

Preview render (fast, lower quality):

npm run render -- --quality=low

Final render (production quality):

npm run render
# Outputs: out/{name}.mp4

Render options:

• --codec=h264 (default, widely compatible)
• --codec=h265 (smaller file size)
• --resolution=1080p (default)
• --resolution=720p (smaller file)
• --fps=30 (default)
• --fps=60 (smoother motion)

Step 7: Post-Production Utilities

Add music to existing video:

python tools/addmusic.py --input video.mp4 \\
  --prompt "Subtle ambient" --output final.mp4

Redub with different voice:

python tools/redub.py --input video.mp4 \\
  --voice-id {ID} --output dubbed.mp4

Remove watermarks:

python tools/dewatermark.py --input video.mp4 \\
  --preset sora --output clean.mp4 --cloud modal

Upscale resolution:

python tools/upscale.py --input video.mp4 \\
  --scale 2x --output hd.mp4 --cloud modal

Multi-Session Project Tracking

Project Lifecycle

planning → assets → review → audio → editing → rendering → complete

project.json Schema

{
  "name": "kaicalls-demo-v1",
  "template": "product-demo",
  "brand": "kaicalls",
  "created": "2026-03-28T15:00:00Z",
  "updated": "2026-03-28T16:30:00Z",
  "phase": "audio",
  "scenes": [
    {
      "id": 1,
      "type": "title",
      "duration": 5,
      "visual": "slide",
      "narration": "Introducing KaiCalls",
      "status": "ready"
    },
    {
      "id": 2,
      "type": "demo",
      "duration": 20,
      "visual": "demos/call-handling.mp4",
      "narration": "Watch how it handles calls...",
      "status": "asset-present"
    }
  ],
  "audio": {
    "voiceover": {
      "file": "audio/voiceover.mp3",
      "status": "present",
      "provider": "qwen3",
      "speaker": "Ryan"
    },
    "music": {
      "file": "audio/background.mp3",
      "status": "needed",
      "preset": "corporate-bg"
    }
  },
  "estimates": {
    "totalDurationSeconds": 90
  },
  "sessions": [
    {
      "date": "2026-03-28",
      "phase": "planning",
      "summary": "Created project, defined scenes"
    },
    {
      "date": "2026-03-28",
      "phase": "assets",
      "summary": "Recorded demo video"
    }
  ]
}

Intent vs Reality Reconciliation

On project resume:

1. Scan filesystem - Check what assets actually exist
2. Compare to intent - What does project.json expect?
3. Update statuses:
   • asset-needed + file exists → asset-present
   • ready + file missing → asset-missing
4. Flag discrepancies - Alert user to missing/unexpected files
5. Regenerate CLAUDE.md - Auto-generated status document

CLAUDE.md example:

# Project: kaicalls-demo-v1

**Template:** product-demo | **Brand:** kaicalls | **Phase:** audio
**Last Updated:** 30 minutes ago

## Current Status

All demo assets are present. Voiceover generated. Ready for final review.

## Scenes

| # | Scene | Type | Visual | Status |
|---|-------|------|--------|--------|
| 1 | Title | title | slide | ✅ Ready |
| 2 | Demo | demo | demos/call-handling.mp4 | ✅ Present |
| 3 | Stats | stats | slide | ✅ Ready |
| 4 | CTA | cta | slide | ✅ Ready |

## Audio

- Voiceover: ✅ Generated (Qwen3-TTS, Ryan voice)
- Music: ⬜ Optional background music

## Next Actions

1. **Preview in Remotion Studio**
   ```bash
   cd projects/kaicalls-demo-v1 && npm run studio

2. Render final video

   npm run render
   

Session History

• 2026-03-28: Created project, defined scenes (planning)
• 2026-03-28: Recorded demo video (assets)
• 2026-03-28: Generated voiceover (audio)

### Session Continuity

**First session:** Planning + scene definition  
**Second session:** Asset creation  
**Third session:** Audio generation  
**Fourth session:** Review + polish  
**Fifth session:** Final render

**Between sessions:**
- project.json tracks state
- CLAUDE.md provides instant context
- Automatic reconciliation on resume
- No context loss

---

## Brand Integration

### Brand Profiles

brands/{brand}/ ├── brand.json # Colors, fonts, typography ├── voice.json # Voice settings (speaker, style) └── assets/ # Logo, backgrounds, graphics

### brand.json Example

```json
{
  "name": "KaiCalls",
  "colors": {
    "primary": "#3B82F6",
    "secondary": "#10B981",
    "background": "#111827",
    "text": "#F9FAFB"
  },
  "fonts": {
    "heading": {
      "family": "Inter",
      "weight": 700
    },
    "body": {
      "family": "Inter",
      "weight": 400
    }
  },
  "logo": "assets/logo.svg",
  "style": "modern-tech"
}

voice.json Example

{
  "provider": "qwen3",
  "speaker": "Ryan",
  "tone": "professional-warm",
  "pace": "moderate",
  "elevenlabs": {
    "voice_id": "YOUR_VOICE_ID",
    "stability": 0.75,
    "similarity_boost": 0.75
  }
}

Auto-application:

• Brand colors applied to all slides
• Brand fonts used throughout
• Logo appears on title/credits
• Voice settings used for all voiceovers

Transition Library

7 custom transitions + 4 Remotion official:

Usage in scenes:

import { glitch, slide } from '@/lib/transitions';

<Sequence from={0} durationInFrames={150}>
  <Title />
</Sequence>
<Sequence from={150} durationInFrames={600}>
  <Demo transition={glitch({ duration: 30 })} />
</Sequence>

Cloud GPU Integration

Modal (Recommended)

Setup:

/video-setup
# Deploys all tools automatically
# $30/month free compute on Starter plan

Tools deployed:

• qwen3_tts - Text-to-speech (~$0.01/video)
• flux2 - Image generation (~$0.02/image)
• music_gen - Music generation (~$0.05/track)
• sadtalker - Talking head video (~$0.10/video)
• ltx2 - AI video generation (~$0.23/clip)
• image_edit - Image editing (~$0.03/image)
• upscale - Image upscaling (~$0.01/image)
• dewatermark - Watermark removal (~$0.10/video)

Typical monthly cost: $1-3 for a few videos (within free tier)

RunPod (Alternative)

Setup:

python tools/qwen3_tts.py --setup
python tools/flux2.py --setup
# ... repeat for each tool

Pay-per-second, no minimums

Quality Gates

Before rendering:

1. Script approved - Hooks strong, no AI slop
2. All assets present - No missing demos/images
3. Timing validated - Scene durations match narration
4. Audio synced - Voiceover aligns with visuals
5. Brand consistent - Colors, fonts, logo correct
6. Transitions smooth - No jarring cuts
7. Text readable - Contrast sufficient, size appropriate
8. Preview reviewed - Remotion Studio walkthrough complete

After rendering:

9. Resolution correct - 1080p or as specified
10. Audio levels balanced - Voiceover audible over music
11. File size reasonable - Under platform limits
12. Format compatible - Works on target platform

Output Organization

workspace/video/
├── _video-projects.md              # Index of all projects
├── projects/
│   ├── kaicalls-demo-v1/
│   │   ├── project.json
│   │   ├── VOICEOVER-SCRIPT.md
│   │   ├── CLAUDE.md
│   │   ├── src/
│   │   ├── public/
│   │   └── out/
│   │       └── kaicalls-demo-v1.mp4
│   └── tropibot-launch/
│       └── ...
├── scripts/                         # Script-only outputs
│   ├── tiktok/
│   ├── youtube/
│   └── reels/
└── _production-guide.md

Commands Integration

/kai-video (entry point)

Script only:

/kai-video script --platform tiktok --topic "KaiCalls demo"
# Generates script, stops

Full production:

/kai-video produce --template product-demo --brand kaicalls
# Full pipeline: script → assets → audio → render

Resume project:

/kai-video resume kaicalls-demo-v1
# Picks up where left off

/video-setup (one-time)

/video-setup
# Interactive setup:
# 1. Cloud GPU provider (Modal/RunPod)
# 2. API keys (ElevenLabs optional)
# 3. Tool deployment
# 4. Verification

/record-demo (asset creation)

/record-demo --url https://app.kaicalls.com --output call-demo
# Opens browser, records interactions, saves MP4

/generate-voiceover (audio creation)

/generate-voiceover --project kaicalls-demo-v1
# Reads VOICEOVER-SCRIPT.md, generates audio per scene

/scene-review (quality check)

/scene-review --project kaicalls-demo-v1
# Opens Remotion Studio, walks through each scene

/design (visual refinement)

/design --scene 3 --project kaicalls-demo-v1
# Focused design session for specific scene

/voice-clone (brand voice)

/voice-clone --brand kaicalls
# Records sample, creates cloned voice

Cost Estimates

Full video production (60-90 seconds):

With premium features:

• ElevenLabs voiceover: +$0.30
• AI video clips (LTX-2): +$0.23/clip
• Talking head (SadTalker): +$0.10

Free tier covers: 100+ videos/month on Modal Starter plan ($30 free compute)

Examples & Templates

Load these when starting production:

For product demos:

• templates/product-demo/ - Marketing video structure
• examples/digital-samba-skill-demo/ - Finished product demo

For explainers:

• templates/explainer/ - Educational content structure
• examples/schlumbergera/ - Sprint review example

For social shorts:

• templates/social-short/ - Vertical video optimized
• Hook-first structure, 15-60 seconds

All examples include:

• Full source code
• Rendered MP4 output
• Production notes
• Lessons learned

Key Principles

1. Script first, production second - Never skip script quality for production speed
2. Multi-session by default - Projects span days/weeks, track state meticulously
3. Intent vs reality - Filesystem is truth, reconcile on every resume
4. Brand consistency - Auto-apply brand profiles, don't re-specify colors/fonts
5. Quality gates mandatory - Review before render, verify after
6. Cost-conscious - Prefer free/cheap AI tools, Modal free tier sufficient
7. Platform-optimized - Different outputs for TikTok vs YouTube vs Reels

Evolution

This skill combines:

• Original kai-video script generation (March 2026)
• claude-code-video-toolkit production pipeline (March 2026)
• Multi-session project tracking from video toolkit
• Brand integration from marketing harness

Continuous improvement:

• Track what works per platform
• Refine templates based on performance
• Add new transitions, scenes, effects
• Optimize costs and render times

Feedback loop:

• Video performs well → template promoted
• Script resonates → hook formula added
• Production bottleneck → tool improved
• Cost spike → cheaper alternative found

Integration with Other Skills

Combines with:

• kai-content-calendar - Schedule video releases
• kai-social - Auto-generate social posts from video
• kai-repurpose - Extract clips for multiple platforms
• kai-analytics - Track video performance
• kai-brand - Maintain brand consistency

Feeds into:

• Product launches (video + email + ads)
• Content marketing (blog + video + social)
• Sales enablement (demo videos)
• Customer success (tutorial videos)

Example workflow:

# 1. Plan content
/kai-content-calendar --month April

# 2. Produce video
/kai-video produce --template product-demo --brand kaicalls

# 3. Repurpose for social
/kai-repurpose --source projects/kaicalls-demo-v1/out/video.mp4

# 4. Schedule posts
/kai-social --schedule --video

# 5. Track performance
/kai-analytics --video-performance

Troubleshooting

"Asset not found" error:

• Check project.json paths match filesystem
• Run intent reconciliation (automatic on resume)
• Verify file moved to correct directory

"Audio out of sync":

• Adjust timing in config.ts
• Regenerate voiceover with updated script
• Use Remotion Studio to visually verify sync

"Render fails":

• Check FFmpeg installed (ffmpeg -version)
• Verify all assets exist
• Run preview render first (--quality=low)

"Tool timeout on Modal":

• Check Modal account has compute credits
• Verify tool deployed (modal app list)
• Try RunPod alternative

"Brand colors not applied":

• Verify brand.json in brands/ directory
• Check brand name matches in project.json
• Regenerate src/config/brand.ts

Next Steps After This Skill

When user says "create video":

1. Clarify scope:

   • Script only? Or full production?
   • Which platform(s)?
   • Timeline?

2. Choose path:

   • Script only: Use original kai-video workflow
   • Full production: Use production pipeline

3. For production:

   • Check if setup complete (/video-setup)
   • Select template
   • Create project
   • Guide through pipeline

4. Track progress:

   • Update project.json after each step
   • Regenerate CLAUDE.md
   • Flag blockers clearly

5. On resume:

   • Scan filesystem
   • Reconcile intent vs reality
   • Present current status
   • Suggest next action

This skill is the union of script generation (marketing focus) and video production (technical execution). Use the appropriate parts based on user needs. Always start with script quality — production amplifies good scripts, it doesn't fix bad ones.