avatar-video

name	avatar-video
user-invocable	true
description	Generates talking avatar videos by chaining image generation (nano-banana), voice synthesis (MiniMax Speech via fal.ai), and lip-sync animation (OmniHuman 1.5 or Infinite Talk via fal.ai). Use when asked to "create a talking avatar", "generate a video with speech", "make an avatar say something", "lip-sync", or any video generation task combining portrait + voice + animation.

avatar-video

Full pipeline: text script → portrait image → voice audio → lip-synced video.

/init - First-Time Setup

~/.claude/skills/avatar-video/scripts/setup.sh

This creates ~/.avatar-video/.env with your FAL_KEY. Get one at https://fal.ai → Dashboard → API Keys.

nano-banana must also be installed (see /nano-banana skill).

Quick Reference

Full Pipeline (text → video)

# Premium quality (OmniHuman 1.5) — ~$3.68 for 30s
~/.claude/skills/avatar-video/scripts/generate-avatar.sh \
  "Hi, I'm your your app assistant. I'm here to help you navigate your VA disability claim." \
  --quality premium

# Budget quality (Infinite Talk) — ~$1.88 for 30s
~/.claude/skills/avatar-video/scripts/generate-avatar.sh \
  "Welcome to our service" \
  --quality budget

Individual Steps

# 1. Generate portrait (uses nano-banana)
nano-banana "professional headshot of a woman, studio lighting, neutral background" -s 1K -o portrait

# 2. Generate voice (MiniMax Speech)
~/.claude/skills/avatar-video/scripts/minimax-tts.sh \
  "Hello, welcome to your app" \
  --voice Friendly_Person \
  --output speech.mp3

# 3a. Lip-sync with OmniHuman 1.5 (premium)
~/.claude/skills/avatar-video/scripts/omnihuman-lipsync.sh \
  --image portrait.png \
  --audio speech.mp3 \
  --output avatar-video.mp4

# 3b. Lip-sync with Infinite Talk (budget)
~/.claude/skills/avatar-video/scripts/infinitetalk-lipsync.sh \
  --image portrait.png \
  --audio speech.mp3 \
  --output avatar-video.mp4

Orchestrator Options

Flag	Default	Description
--quality	premium	premium (OmniHuman) or budget (Infinite Talk)
--engine	auto	omnihuman or infinitetalk (overrides --quality)
--voice	Friendly_Person	MiniMax voice ID
--voice-speed	1.0	Speech speed (0.5-2.0)
--image	-	Skip image gen, use existing portrait
--audio	-	Skip voice gen, use existing audio
--output	avatar-{timestamp}	Output filename (no extension)
--dir	current directory	Output directory
--image-prompt	auto	Custom prompt for portrait generation

MiniMax Speech Voices

Built-in voices (pass to --voice):

Voice ID	Description
Wise_Woman	Mature, authoritative female
Friendly_Person	Warm, approachable (default)
Deep_Voice_Man	Deep, resonant male
Calm_Woman	Soothing, relaxed female
Casual_Guy	Informal, conversational male
Lively_Girl	Energetic, youthful female

Custom voice cloning: provide a voice sample URL as --voice <url_to_6sec_sample>.

Engine Comparison

Feature	OmniHuman 1.5	Infinite Talk
Cost	$0.12/s (lip-sync)	$0.03/s (480p), $0.06/s (720p)
30s video	~$3.60	~$0.90-$1.80
Max length	30s audio	10 minutes
Quality	Best — full body motion, gestures	Good — solid lip-sync
Resolution	720p/1080p	480p/720p
Best for	Short premium content, ads	Long-form, bulk, prototyping

Full Pipeline Costs (30-second video)

Step	Tool	Cost
Image	nano-banana (Gemini Flash)	~$0.05
Voice	MiniMax Speech 2.6 Turbo	~$0.03
Lip-sync (budget)	Infinite Talk 720p	~$1.80
Lip-sync (premium)	OmniHuman 1.5	~$3.60
Total (budget)		~$1.88
Total (premium)		~$3.68

Cost Tracking

Every generation logs to ~/.avatar-video/costs.json. View summary:

~/.claude/skills/avatar-video/scripts/generate-avatar.sh --costs

API Keys

Service	Key	Source
MiniMax Speech, OmniHuman, Infinite Talk	FAL_KEY	~/.avatar-video/.env
nano-banana (image gen)	GEMINI_API_KEY	~/.nano-banana/.env

All fal.ai services use a single key. Get one at https://fal.ai.

Troubleshooting

• "FAL_KEY not found" — Run setup: ~/.claude/skills/avatar-video/scripts/setup.sh
• "Audio must be under 30s" (OmniHuman) — Use Infinite Talk for longer audio, or trim script
• Upload failed — Check FAL_KEY is valid; fal.ai file upload requires auth
• Queue timeout — OmniHuman can take 2-5 min; the script polls automatically
• nano-banana not found — Install via /nano-banana skill init