# AI Talking Head

> Specialized skill for AI talking head and lip-sync video generation. Use when you need presenter videos, UGC-style content, or lip-synced avatars. Triggers on: talking head, presenter video, lip sync, UGC video. Outputs professional talking head videos.

- Skill: `majiayu000/ai-talking-head` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds add majiayu000/ai-talking-head`
- Raw SKILL.md: https://api.skillmd.com/api/skills/majiayu000/ai-talking-head/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: majiayu000 (https://skillmd.com/u/majiayu000)
- Updated: 2026-09-09
- Page: https://skillmd.com/skills/majiayu000/ai-talking-head

---


# AI Talking Head

Generate talking head videos, presenter content, and lip-synced videos.

**Use this skill when:** You need a person (real or AI) talking to camera.
**Route here from:** ai-creative-workflow, ai-creative-strategist, or direct requests.

---

## Why This Skill Exists

**The problem:** Talking head videos are the most persuasive content format but:
1. Recording yourself is time-consuming and requires confidence
2. Professional presenters are expensive ($500-5000+ per video)
3. UGC creators charge $100-500 per post and may not match your brand
4. Iterating on scripts means re-filming everything
5. Scaling personalized video is nearly impossible manually

**The solution:** AI talking heads that:
- Generate professional presenter videos in minutes
- Let you iterate on scripts without re-recording
- Create unlimited variants for A/B testing
- Maintain consistent brand presenter identity
- Scale personalized outreach cost-effectively

**The game-changer:** Combining avatar generation + lip-sync lets you:
- Create a consistent "brand spokesperson"
- Update any script without re-filming
- Test multiple presenter styles quickly
- Produce video content at 10x the speed

---

## Presenter Style Exploration (Before Generation)

**Critical insight from ai-creative-strategist:** Don't generate with one style and hope it works. Explore genuinely DIFFERENT presenter styles first.

### The Style Exploration Process

**STEP 1: GENERATE 4-5 DIFFERENT PRESENTER STYLES**

This is NOT: Same person with different clothes
This IS: Fundamentally different presenter archetypes that each tell a different story

```
[YOUR BRAND] - Style Exploration

Generate presenter concepts for these 5 directions:

1. CORPORATE AUTHORITY
   - Demographic: 35-50, professional appearance
   - Setting: Modern office, corporate environment
   - Wardrobe: Business professional, suit/blazer
   - Energy: Confident, measured, authoritative
   - Vibe: "Trust the expert"

2. RELATABLE FRIEND
   - Demographic: 25-40, approachable look
   - Setting: Home office, kitchen, casual space
   - Wardrobe: Smart casual, comfortable
   - Energy: Warm, conversational, genuine
   - Vibe: "Let me share what worked for me"

3. ENERGETIC CREATOR
   - Demographic: 22-35, creator aesthetic
   - Setting: Ring light setup, content studio
   - Wardrobe: Trendy casual, branded
   - Energy: High, dynamic, enthusiastic
   - Vibe: "You HAVE to try this"

4. EXPERT EDUCATOR
   - Demographic: 30-55, credible appearance
   - Setting: Study, library, professional backdrop
   - Wardrobe: Smart casual, glasses optional
   - Energy: Calm, explanatory, helpful
   - Vibe: "Let me explain how this works"

5. LIFESTYLE ASPIRATIONAL
   - Demographic: 28-45, aspirational look
   - Setting: Beautiful home, travel location, luxury
   - Wardrobe: Elevated casual, tasteful
   - Energy: Relaxed confidence, success aura
   - Vibe: "This is what my life looks like"
```

**STEP 2: IDENTIFY WINNER**

After generating style exploration:
```
REVIEW each presenter style:

Which presenter:
- Best matches brand voice?
- Would audience trust most?
- Fits the content type?
- Has right energy level?
- Would work across multiple videos?

WINNER: [Selected style]
BECAUSE: [Why this style wins for this brand/use case]
```

**STEP 3: EXTRACT PRESENTER PRINCIPLES**

Once winner identified:
```
WINNING STYLE EXTRACTION

Demographics:
- Age range: [X-X]
- Gender: [if specific]
- Ethnicity: [if specific]
- Overall look: [descriptors]

Environment:
- Primary setting: [where they present from]
- Background elements: [what's visible]
- Lighting style: [natural/studio/mixed]

Wardrobe:
- Style: [formal/casual/etc.]
- Colors: [palette]
- Accessories: [if any]

Delivery:
- Energy level: [1-10]
- Speaking pace: [slow/medium/fast]
- Hand gestures: [minimal/moderate/expressive]
- Eye contact: [direct to camera always]

Audio:
- Voice tone: [warm/authoritative/energetic]
- Pacing: [conversational/punchy/measured]
```

**STEP 4: APPLY ACROSS CONTENT**

Use extracted principles for:
- All future videos maintain consistency
- Same presenter = brand recognition
- Variations in script, not in presenter

---

## Presenter Archetype Deep Dives

### Corporate Authority

**When to use:** B2B, financial services, healthcare, enterprise SaaS, professional services

**Visual Formula:**
```
[Man/Woman] in [30s-50s], [silver/dark hair], wearing [tailored blazer/suit],
in [modern glass office/conference room with city view], [warm professional lighting],
[confident composed expression], [seated at desk OR standing with slight lean],
[direct eye contact with camera], [subtle hand gestures], corporate executive style
```

**Setting Options:**
- Corner office with city view
- Modern conference room
- Executive desk with minimal decor
- Standing at presentation screen
- Seated in designer chair

**Wardrobe Options:**
- Tailored navy blazer over white shirt
- Grey suit, no tie (modern)
- Classic suit with subtle tie
- Blazer over turtleneck (thought leader)
- Professional dress (solid colors)

**Energy Markers:**
- Measured pace
- Deliberate movements
- Confident pauses
- Minimal but purposeful gestures
- Assured vocal tone

---

### Relatable Friend (UGC Style)

**When to use:** DTC brands, consumer products, wellness, beauty, lifestyle

**Visual Formula:**
```
[Friendly man/woman] in [25-40s], wearing [casual but put-together outfit],
in [bright modern apartment/kitchen/home office], [natural window light],
[genuine warm smile], [relaxed comfortable posture], [talking to camera like
a friend], [natural hand movements], authentic UGC creator style
```

**Setting Options:**
- Bright kitchen counter
- Cozy living room couch
- Home office with plants
- Bedroom getting-ready setup
- Outdoor patio/balcony

**Wardrobe Options:**
- Cozy sweater/cardigan
- Simple t-shirt
- Casual button-down
- Loungewear (if brand appropriate)
- Athleisure

**Energy Markers:**
- Conversational rhythm
- Natural pauses ("honestly?", "okay so...")
- Expressive facial reactions
- Genuine enthusiasm without over-selling
- Relatable body language

**UGC Script Patterns:**
```
DISCOVERY: "Okay so I found this [product] and I'm obsessed..."
REVIEW: "So I've been using [product] for [time] and here's my honest take..."
COMPARISON: "I used to use [old product] but then I tried [new product]..."
TRANSFORMATION: "Before [product] I was [problem]. Now? [result]."
```

---

### Energetic Creator

**When to use:** Gen-Z products, entertainment, gaming, trendy DTC, social apps

**Visual Formula:**
```
[Young energetic creator] in [22-35], [colorful trendy outfit], in [content
studio with ring light/neon lights], [bright dynamic lighting], [animated
expressions], [lots of movement and gestures], [high energy delivery],
[fast-paced enthusiastic style], YouTube/TikTok creator aesthetic
```

**Setting Options:**
- Ring light setup visible
- LED/neon accent lighting
- Streaming/gaming setup
- Colorful backdrop
- Outdoor action setting

**Wardrobe Options:**
- Graphic tees
- Bold colors
- Branded merch
- Trendy streetwear
- Statement accessories

**Energy Markers:**
- Fast-paced delivery
- Big expressions
- Lots of hand movement
- Pattern interrupts
- Enthusiasm at 10

**Creator Script Patterns:**
```
HOOK: "STOP scrolling. This is important."
REVEAL: "I literally just discovered [thing] and I'm freaking out."
CHALLENGE: "I bet you can't guess what [product] does."
REACTION: "[reaction to trying product]... WAIT what?!"
```

---

### Expert Educator

**When to use:** Online courses, professional services, B2B explainers, tutorials

**Visual Formula:**
```
[Knowledgeable expert] in [30s-55], [smart casual or academic style],
in [home study/office with books/whiteboard], [balanced lighting],
[thoughtful composed expression], [explaining with purposeful gestures],
[patient instructive tone], educator/thought leader style
```

**Setting Options:**
- Study with bookshelves
- Office with credentials visible
- Whiteboard/screen behind
- Standing at presentation
- Desk with relevant props

**Wardrobe Options:**
- Button-down shirt
- Blazer over casual shirt
- Sweater over collared shirt
- Glasses (authority signal)
- Minimal accessories

**Energy Markers:**
- Patient pace
- Teaching rhythm
- Logical structure
- Illustrative gestures
- "Here's what matters" moments

---

### Lifestyle Aspirational

**When to use:** Luxury brands, high-ticket services, aspirational DTC, travel, real estate

**Visual Formula:**
```
[Elegant successful person] in [30s-50s], [elevated casual attire],
in [beautiful interior/scenic location], [golden hour OR designer lighting],
[relaxed confident demeanor], [speaking with quiet confidence], [minimal
but graceful movement], aspirational lifestyle aesthetic
```

**Setting Options:**
- Designer living room
- Travel location (balcony view)
- Luxury car interior
- High-end restaurant/hotel
- Yacht/beach/resort

**Wardrobe Options:**
- Designer casual
- Linen/natural fabrics
- Neutral luxury palette
- Subtle jewelry/watch
- Effortlessly elegant

**Energy Markers:**
- Relaxed confidence
- No rushing
- "I have time" energy
- Subtle smile
- Quiet success vibes

---

## Video Model Roster (Quality Winners)

Generate presenter videos with ALL THREE models, present outputs for selection:

| Model | Owner | Speed | Strengths |
|-------|-------|-------|-----------|
| **Sora 2** | openai | ~80s | Excellent general quality, good faces |
| **Veo 3.1** | google | ~130s | Native audio generation, natural movement |
| **Kling v2.5 Turbo Pro** | kwaivgi | ~155s | Best for people/motion, most realistic |

**Strategy:** Run same prompt through all 3 models → User picks best output.

### Model Selection Guide

```
FOR MAXIMUM REALISM (people quality):
    → Kling v2.5 Turbo Pro (best faces, most natural movement)

FOR SPEED + QUALITY BALANCE:
    → Sora 2 (fastest, still good quality)

FOR BUILT-IN AUDIO:
    → Veo 3.1 (generates audio with video)

FOR UGC AUTHENTICITY:
    → Kling v2.5 (handles casual movements well)

FOR CORPORATE/FORMAL:
    → Sora 2 or Kling v2.5 (cleaner, more controlled)
```

---

## Lip-Sync Model

For adding speech to existing videos:

| Model | Use | Cost | Speed | Quality |
|-------|-----|------|-------|---------|
| **Kling Lip-Sync** | Add voiceover to any video | ~$0.20 | ~1min | Excellent |

**When to use Lip-Sync:**
- You have a great presenter video but need different script
- Client wants to change messaging after video generation
- Creating personalized versions of same base video
- Adding voiceover to product demo videos
- Dubbing content for different languages

---

## Use Cases Deep Dive

### 1. Lip-Sync Overlay

**Best for:** Adding voiceover to existing video, dubbing, personalization

**Input Requirements:**
- Video with visible face (front-facing works best)
- Audio file (MP3, WAV) OR text script

**Workflow:**
```json
{
  "model_owner": "kwaivgi",
  "model_name": "kling-lip-sync",
  "Prefer": "wait",
  "input": {
    "video": "https://... (source video URL)",
    "audio": "https://... (audio file URL)"
  }
}
```

**Or with text (uses built-in TTS):**
```json
{
  "input": {
    "video": "https://... (source video URL)",
    "text": "Script text to speak"
  }
}
```

**Quality Tips:**
- Source video should have face visible 70%+ of time
- Forward-facing shots work better than profiles
- Avoid videos with heavy face movement/turning
- Audio should be clear without background noise
- Script pacing should match natural speech

---

### 2. AI Presenter Generation

**Best for:** Creating presenter content from scratch, brand spokesperson

**Multi-Model Workflow:**

```json
// Sora 2
{
  "model_owner": "openai",
  "model_name": "sora-2",
  "input": {
    "prompt": "[presenter prompt]",
    "aspect_ratio": "16:9",
    "duration": 5
  }
}

// Veo 3.1 (with native audio)
{
  "model_owner": "google",
  "model_name": "veo-3.1",
  "input": {
    "prompt": "[presenter prompt]",
    "aspect_ratio": "16:9",
    "generate_audio": true
  }
}

// Kling v2.5
{
  "model_owner": "kwaivgi",
  "model_name": "kling-v2.5-turbo-pro",
  "input": {
    "prompt": "[presenter prompt]",
    "aspect_ratio": "16:9",
    "duration": 5
  }
}
```

**Then add lip-sync if specific script needed:**
```json
{
  "model_owner": "kwaivgi",
  "model_name": "kling-lip-sync",
  "input": {
    "video": "[generated video URL]",
    "text": "[script text]"
  }
}
```

---

### 3. UGC-Style Content

**Best for:** Authentic testimonials, product reviews, social proof

**The UGC Formula:**
```
[Relatable person] + [Casual setting] + [Natural lighting] +
[Authentic delivery] + [Genuine reaction] = Believable UGC
```

**Prompt Template:**
```
Friendly [demographic] sitting in [casual setting], natural window light,
holding/showing [product], genuine excited expression, talking directly to
camera like filming a selfie video, authentic UGC testimonial style, casual
comfortable body language, 5 seconds
```

**UGC Authenticity Markers:**
- Slightly imperfect framing
- Natural lighting (not studio)
- Casual wardrobe
- Real reactions, not posed
- Personal space as backdrop
- Eye contact with camera

---

### 4. Personal Brand Series

**Best for:** Thought leaders, course creators, coaches, consultants

**Consistency Formula:**
```
ESTABLISH ONCE, USE FOREVER:
- Same presenter appearance
- Same setting/background
- Same wardrobe style
- Same energy level
- Same lighting setup

Only change: Script and specific content
```

**Series Prompt Template:**
```
[Consistent presenter description - use same each time], [same setting],
[same lighting], [same wardrobe style], [same energy], discussing [new topic],
[consistent delivery style], 5 seconds
```

---

## Script Mastery

### Duration Calculation

| Word Count | Duration | Use Case |
|------------|----------|----------|
| 15 words | ~5 seconds | Social hook |
| 30 words | ~10 seconds | Instagram Reel |
| 45 words | ~15 seconds | TikTok optimal |
| 60 words | ~20 seconds | Short testimonial |
| 90 words | ~30 seconds | Product explainer |
| 150 words | ~60 seconds | Full testimonial |

**Rule:** ~150 words per minute at natural conversational pace

### Script Structures

**HOOK-VALUE-CTA (15-30 seconds):**
```
Hook (0-3 sec): [Attention-grabber - question, statement, or pattern interrupt]
Value (3-20 sec): [Main message, benefit, or story]
CTA (20-30 sec): [Clear next step]
```

**PROBLEM-AGITATE-SOLVE (30-60 seconds):**
```
Problem (0-10 sec): [Name the pain point]
Agitate (10-30 sec): [Make them feel it]
Solve (30-60 sec): [Present the solution + CTA]
```

**BEFORE-AFTER (15-30 seconds):**
```
Before (0-10 sec): [Life before product/solution]
After (10-25 sec): [Transformation/result]
CTA (25-30 sec): [How to get same result]
```

### Tone Templates

**Professional/Corporate:**
```
"[Name] here with [Company]. Today I want to share how [product/insight]
can help you [achieve outcome]. Here's what you need to know..."
```

**Casual/UGC:**
```
"Okay so I've been using [product] for [time] and honestly? I'm obsessed.
Here's why [specific benefit]. If you [problem], you need this."
```

**Expert/Educational:**
```
"One thing I see people get wrong about [topic] is [misconception].
Here's what actually works: [insight]. Let me show you..."
```

**Energetic/Sales:**
```
"Stop what you're doing. [Product] just changed everything. I'm serious -
[result] in [timeframe]. You HAVE to try this."
```

**Aspirational:**
```
"[Casual opening]. I wanted to share something that's completely transformed
[area of life]. [Product] gave me [result]. Here's how it works..."
```

---

## Platform-Specific Optimization

### TikTok/Reels (9:16)

**Specs:**
- Aspect Ratio: 9:16 (vertical)
- Duration: 15-30 seconds optimal
- Safe Zone: Keep face/text center 60%

**Style Adjustments:**
```
→ Higher energy delivery
→ Faster pacing
→ Hook in first 1-2 seconds
→ Pattern interrupts
→ Jump cuts acceptable
→ Casual/authentic feel
```

**Prompt Modifier:**
```
...[base prompt], filmed vertically like TikTok/Reels content,
energetic creator style, direct eye contact with camera
```

### YouTube (16:9)

**Specs:**
- Aspect Ratio: 16:9 (landscape)
- Duration: 30-120 seconds
- Safe Zone: Standard letterbox

**Style Adjustments:**
```
→ More measured pacing
→ Can be longer form
→ More professional setups accepted
→ Room for B-roll integration
→ Intro/outro structure
```

**Prompt Modifier:**
```
...[base prompt], widescreen YouTube style, professional yet engaging,
room for graphics/lower thirds
```

### LinkedIn (1:1 or 16:9)

**Specs:**
- Aspect Ratio: 1:1 (square) or 16:9
- Duration: 30-60 seconds optimal
- Tone: Professional but personal

**Style Adjustments:**
```
→ Professional appearance
→ Business-appropriate setting
→ Thought leadership tone
→ Value-first messaging
→ Credibility signals
```

**Prompt Modifier:**
```
...[base prompt], professional LinkedIn style, credible expert appearance,
business casual in modern office environment
```

### Instagram Stories (9:16)

**Specs:**
- Aspect Ratio: 9:16
- Duration: 15 seconds max per segment
- Ephemeral feel

**Style Adjustments:**
```
→ Casual, in-the-moment feel
→ Can be "rougher" quality
→ Direct audience address
→ Personal/behind-scenes vibe
→ Clear single message per story
```

### Ads (Various)

**Facebook/Instagram Ads:**
- 1:1, 4:5, or 9:16
- 15-30 second optimal
- Hook in 0-3 seconds
- Clear CTA

**YouTube Ads:**
- 16:9
- 15-30 second (skippable) or 6 second (bumper)
- Brand visible throughout

---

## Audio & Voice Considerations

### When Using Veo 3.1 Native Audio

**Strengths:**
- Generates synchronized audio with video
- Natural ambient sounds
- Speech that matches lip movement
- Good for establishing scenes

**Limitations:**
- Less control over specific script
- Audio quality varies
- May need post-processing

### When Adding Lip-Sync

**Best Practices:**
- Use high-quality audio recording
- Match energy level to video presenter
- Pace script to natural speaking rhythm
- Allow for breath pauses
- Keep sentences short (easier sync)

### Voice-Over Tips

**If recording your own VO for lip-sync:**
```
□ Record in quiet environment
□ Use consistent distance from mic
□ Match energy to presenter style
□ Natural pauses between sentences
□ Clear enunciation
□ Export as MP3 or WAV
```

**If using TTS (text input):**
```
□ Use punctuation for natural pauses
□ Write phonetically for tricky words
□ Keep sentences conversational length
□ Test different phrasings
□ Consider adding "..." for pauses
```

---

## Execution Workflow

### Step 1: Clarify Requirements

Before generating:
```
□ What's the use case? (UGC, corporate, educational, etc.)
□ What platform? (TikTok, YouTube, LinkedIn, ads)
□ What aspect ratio? (9:16, 16:9, 1:1)
□ What duration? (and word count)
□ What presenter style? (see archetypes)
□ What's the script/message?
□ Need lip-sync to specific audio?
```

### Step 2: Style Selection

If not predefined:
```
□ Generate style exploration with 4-5 different presenter styles
□ Present options to user
□ Extract principles from winner
□ Document for consistency
```

### Step 3: Construct Prompt

Use this formula:
```
[PRESENTER DESCRIPTION] + [SETTING] + [LIGHTING] +
[EXPRESSION/ENERGY] + [ACTION] + [STYLE MODIFIER] + [DURATION]
```

### Step 4: Multi-Model Generation

```
Run same prompt through:
1. Sora 2 (~80s)
2. Veo 3.1 (~130s)
3. Kling v2.5 (~155s)

Present all three to user for selection.
```

### Step 5: Add Lip-Sync (If Needed)

If specific script delivery required:
```
1. User approves video from Step 4
2. Run through Kling Lip-Sync
3. Input: selected video + audio/text
4. Output: synced talking head
```

### Step 6: Deliver & Iterate

```markdown
## Talking Head Video Options

**Style:** [Archetype used]
**Platform:** [Target platform]
**Duration:** [X seconds]

### Option 1: Sora 2
[video URL]
Notes: [quality assessment]

### Option 2: Veo 3.1 (with audio)
[video URL]
Notes: [quality assessment]

### Option 3: Kling v2.5
[video URL]
Notes: [quality assessment]

**Select preferred video for lip-sync or final delivery.**
```

---

## Quality Checklist

### Technical Quality
- [ ] Face clearly visible throughout
- [ ] No uncanny valley artifacts
- [ ] Consistent appearance (no morphing)
- [ ] Smooth natural movement
- [ ] Appropriate resolution for platform

### Presenter Quality
- [ ] Matches intended archetype
- [ ] Expression appropriate for message
- [ ] Energy level fits content type
- [ ] Wardrobe matches brand/context
- [ ] Setting supports message

### Lip-Sync Quality (if applicable)
- [ ] Mouth movement matches audio
- [ ] Natural speech rhythm
- [ ] No obvious desync
- [ ] Head movement doesn't break sync
- [ ] Audio quality clear

### Content Quality
- [ ] Script delivered clearly
- [ ] Pacing appropriate for platform
- [ ] Hook captures attention
- [ ] Message comes through
- [ ] CTA clear (if applicable)

---

## Common Issues & Solutions

| Issue | Cause | Solution |
|-------|-------|----------|
| Uncanny valley feel | Model limitations | Use Kling v2.5 for most realistic faces |
| Face morphing mid-video | Long duration | Keep videos shorter (5-10 sec), extend with cuts |
| Lip-sync drift | Audio/video mismatch | Use shorter scripts, clear enunciation |
| Wrong energy level | Prompt too vague | Be explicit about energy: "calm" vs "enthusiastic" |
| Generic stock presenter | No specific direction | Add detailed demographic and style descriptors |
| Setting doesn't match | Prompt conflict | Prioritize setting description, remove conflicts |
| Awkward hand movement | Unspecified gestures | Add gesture direction or specify "minimal movement" |
| Bad lighting | Missing lighting prompt | Always include lighting: "warm natural light" |
| Doesn't look like brand | No style consistency | Create and use presenter spec document |
| Audio quality poor | TTS limitations | Use recorded audio instead of text input |

---

## Output Format

### Style Exploration Output
```markdown
## Presenter Style Exploration

**Brand/Project:** [Name]
**Use Case:** [What videos will be used for]

### Style 1: Corporate Authority
[video URL or generation]
- Demographic: [specifics]
- Setting: [description]
- Energy: [level]

### Style 2: Relatable Friend
[video URL or generation]
- Demographic: [specifics]
- Setting: [description]
- Energy: [level]

[...continue for all 5 styles...]

**Recommendation:** Style [X] best fits because [reasons]
**Feedback needed:** Which direction resonates?
```

### Generated Video Output
```markdown
## Talking Head Video Generated

**Style:** [Archetype]
**Platform:** [Target]
**Duration:** [X seconds]

### Model Outputs:

**Sora 2:** [URL]
**Veo 3.1:** [URL] (includes audio)
**Kling v2.5:** [URL]

**Prompt Used:**
> [full prompt for reference]

**Next Steps:**
- [ ] Select preferred video
- [ ] Add lip-sync to specific script (if needed)
- [ ] Request variation
- [ ] Approve for use
```

### Lip-Sync Output
```markdown
## Lip-Sync Video Delivered

**Source Video:** [URL]
**Script:** "[excerpt...]"
**Duration:** [X seconds]

**Final Video:** [URL]

**Quality Check:**
- ✓ Sync accuracy
- ✓ Natural rhythm
- ✓ Audio clarity
- ✓ Expression match

**Options:**
- [ ] Approve and use
- [ ] Adjust script and resync
- [ ] Try different source video
```

---

## Pipeline Integration

```
TALKING HEAD PIPELINE

┌─────────────────────────────────────────┐
│  Request arrives (direct or routed)     │
│  → Clarify: platform, duration, style   │
│  → Determine: generation vs lip-sync    │
└─────────────────────────────────────────┘
                    │
        ┌───────────┴───────────┐
        ▼                       ▼
┌──────────────────┐   ┌──────────────────┐
│  Style Undefined │   │  Style Defined   │
│  → Run style     │   │  → Skip to       │
│    exploration   │   │    generation    │
└──────────────────┘   └──────────────────┘
                    │
                    ▼
┌─────────────────────────────────────────┐
│  ai-talking-head (THIS SKILL)           │
│  → Multi-model generation               │
│  → Present options                      │
│  → Add lip-sync if needed               │
│  → Quality check                        │
└─────────────────────────────────────────┘
                    │
                    ▼
┌─────────────────────────────────────────┐
│  Delivery                               │
│  → Platform-optimized output            │
│  → Ready for ads/social/content         │
└─────────────────────────────────────────┘
```

---

## Handoff Protocols

### Receiving from ai-creative-workflow
```yaml
Receive:
  use_case: "talking head" | "UGC" | "presenter" | "lip-sync"
  platform: "[target platform]"
  aspect_ratio: "[ratio]"
  duration: "[seconds]"
  style: "[archetype or custom]"
  script: "[text]"
  audio_url: "[if lip-sync with audio]"
  video_url: "[if lip-sync to existing]"
```

### Returning to Workflow
```yaml
Return:
  status: "complete" | "needs_selection" | "needs_iteration"
  deliverables:
    - video_url: "[URL]"
      model: "[which model]"
      has_audio: true | false
      duration: "[seconds]"
  feedback_needed: "[any questions]"
```

### Receiving Video from ai-product-video
```yaml
Receive for lip-sync:
  video_url: "[product video URL]"
  aspect_ratio: "[ratio]"
  script: "[voiceover text]"
  audio_url: "[optional, if pre-recorded]"
```

---

## Tips from Experience

### What Works

1. **Consistency beats variety** — Same presenter across videos builds recognition
2. **Kling v2.5 for faces** — Most realistic human generation
3. **Shorter is safer** — 5-10 second clips avoid quality degradation
4. **Explicit energy levels** — "calm and measured" vs "enthusiastic and dynamic"
5. **Multi-model approach** — Always generate with 2-3 models, let user pick
6. **Lip-sync extends value** — One good video can become many scripts

### What Doesn't Work

1. **Vague presenter description** — "A person talking" = generic results
2. **Long continuous takes** — Quality degrades after 10-15 seconds
3. **Ignoring setting** — Presenter without context looks artificial
4. **Skipping style exploration** — First idea rarely best for brand
5. **Mismatched energy** — Corporate script + UGC style = awkward
6. **Complex movements** — Walking + talking + gesturing = artifacts

### The 80/20

80% of talking head success comes from:
1. Clear presenter archetype selection
2. Matching energy to platform
3. Short, punchy scripts
4. Using Kling v2.5 for realism

Get these four right, and you'll get good results.

---

## Quick Reference

| Task | Model | Process |
|------|-------|---------|
| Generate presenter video | All 3 models | Multi-model, user picks |
| Add speech to existing video | Kling Lip-Sync | Direct, ~1min |
| Presenter + specific script | Generate → Lip-Sync | Two-step |
| Video with built-in audio | Veo 3.1 | Single generation |
| Most realistic face | Kling v2.5 | Single or multi-model |
| Fastest generation | Sora 2 | Single generation |
| UGC style | Kling v2.5 | Handles casual movement best |