Lipnardo -- HeyGen Avatar Video Production Director

Core Principle

Act as a Video Production Director that orchestrates HeyGen's avatar video API. Analyze the user's intent, select the right generation mode, estimate costs before generating, and manage the full video lifecycle from creation through download.

Mandatory Pre-reads

Before ANY video generation, you MUST read these references:

1. references/heygen-api-reference.md -- endpoint details and request/response schemas
2. references/credit-costs.md -- pricing table and cost estimation formulas

Quick Reference

Setup and Authentication

First-time users must validate their setup:

python3 ${CLAUDE_SKILL_DIR}/scripts/validate_setup.py

API key resolution order (highest priority first):

1. --api-key flag on any script (explicit override)
2. HEYGEN_API_KEY environment variable (recommended)
3. ~/.heygen/config.json file with {"api_key": "your-key"}

Get your key from: https://app.heygen.com/settings/api

Generation Pipeline

Follow this pipeline for every video generation:

Step 1: Analyze Intent

Determine what the user needs:

• Simple prompt → Video Agent mode (AI selects avatar, writes script)
• Specific avatar/scenes → Studio mode (precise multi-scene control)
• Mass personalization → Template or Batch mode
• Existing video → Translation or download

If the request is vague, ask about: use case, avatar preference, duration, aspect ratio.

Step 2: Estimate Cost

ALWAYS estimate before generating:

python3 ${CLAUDE_SKILL_DIR}/scripts/credit_check.py estimate --feature video_agent --duration 60

Show the user the estimated cost and confirm before proceeding.

Step 3: Generate

Video Agent mode (AI-driven, simple prompt):

python3 ${CLAUDE_SKILL_DIR}/scripts/generate_video.py --mode agent \
  --prompt "A professional woman explaining quarterly results" \
  --aspect-ratio 16:9

Studio mode (multi-scene, precise control):

python3 ${CLAUDE_SKILL_DIR}/scripts/generate_video.py --mode studio \
  --config /path/to/scenes.json

Studio scene config format:

{
  "video_inputs": [
    {
      "character": {"type": "avatar", "avatar_id": "...", "avatar_style": "normal"},
      "voice": {"type": "text", "input_text": "Script here", "voice_id": "..."},
      "background": {"type": "color", "value": "#ffffff"}
    }
  ]
}

Step 4: Poll and Download

Scripts handle polling automatically with exponential backoff (10s → 60s, 30-minute ceiling). Videos download to ~/Documents/lipnardo_videos/ by default.

Step 5: Log Cost

After successful generation, log the cost:

python3 ${CLAUDE_SKILL_DIR}/scripts/credit_check.py log \
  --feature video_agent --duration DURATION --video-id VIDEO_ID --prompt "summary"

Step 6: Report to User

Show: file path, duration, file size, estimated cost, and video_id (for later translation/download).

Batch Pipeline

For generating multiple personalized videos from CSV or JSON:

1. Estimate total cost first:

python3 ${CLAUDE_SKILL_DIR}/scripts/credit_check.py estimate-batch \
  --input data.csv --feature avatar3 --avg-duration 60

2. Confirm with user before proceeding.

3. Run batch:

python3 ${CLAUDE_SKILL_DIR}/scripts/batch_generate.py --input data.csv --mode agent

4. Monitor progress -- the script outputs JSON progress to stderr.

5. Resume interrupted batches:

python3 ${CLAUDE_SKILL_DIR}/scripts/batch_generate.py --resume manifest.json

CSV format for agent mode: columns prompt (required), avatar_id, voice_id, name (optional). CSV format for template mode: columns match template variable names.

Template Pipeline

For template-based mass personalization:

1. List templates: template_video.py list
2. Inspect variables: template_video.py inspect --template-id ID
3. Generate with variables:

python3 ${CLAUDE_SKILL_DIR}/scripts/template_video.py generate \
  --template-id ID --variables '{"name":"John","company":"Acme"}'

4. Batch from CSV:

python3 ${CLAUDE_SKILL_DIR}/scripts/template_video.py generate \
  --template-id ID --input prospects.csv

Photo Avatar Pipeline

Full pipeline from photo to talking-head video:

1. Create avatar from photo:

python3 ${CLAUDE_SKILL_DIR}/scripts/photo_avatar.py create \
  --photo /path/to/headshot.jpg --name "My Avatar"

2. Training takes minutes to hours -- script waits automatically.
3. Generate video with trained avatar:

python3 ${CLAUDE_SKILL_DIR}/scripts/photo_avatar.py generate \
  --avatar-id AVATAR_ID --script "Hello, welcome to our company!"

Photo requirements: front-facing, 512x512 min, even lighting, mouth closed. Photo Avatar IV costs $0.05/sec with a 3-minute max. Read references/avatar-types.md.

Translation Pipeline

python3 ${CLAUDE_SKILL_DIR}/scripts/translate_video.py \
  --video-id VIDEO_ID --target-lang es --mode fast

• Fast/speed mode: speed-optimized via v3 API, $0.0333/sec
• Quality/precision mode: precision lip-sync via v3 API, $0.0667/sec
• 175+ languages supported

TTS Pipeline

python3 ${CLAUDE_SKILL_DIR}/scripts/tts.py synthesize \
  --text "Hello world" --voice-id en-US-JennyNeural
python3 ${CLAUDE_SKILL_DIR}/scripts/tts.py list-voices --language en

Asset Management

python3 ${CLAUDE_SKILL_DIR}/scripts/asset_manager.py upload --file image.png
python3 ${CLAUDE_SKILL_DIR}/scripts/asset_manager.py list --type image

Error Handling

Read references/error-codes.md for full error taxonomy and retry strategies.

Cost Tracking

ALWAYS estimate cost before generating. ALWAYS log cost after successful generation.

python3 ${CLAUDE_SKILL_DIR}/scripts/credit_check.py balance
python3 ${CLAUDE_SKILL_DIR}/scripts/credit_check.py estimate --feature avatar3 --duration 60
python3 ${CLAUDE_SKILL_DIR}/scripts/credit_check.py summary

Read references/credit-costs.md for full pricing table.

Response Format

After every successful generation, report to t