ai-avatar-video
๐ฏSkillfrom agentspace-so/runcomfy-agent-skills
|A Claude Code skill for creating AI avatar and talking-head videos through RunComfy CLI, routing across OmniHuman, Wan 2-7, HappyHorse, and Seedance v2 models for audio-driven lip-sync and virtual presenter production.
Overview
AI Avatar & Talking Head Video is a Claude Code skill for creating audio-driven avatar and talking-head videos through the RunComfy CLI. It routes across multiple model endpoints: ByteDance OmniHuman for full-body audio-driven avatars from a single portrait plus audio file, Wan-AI Wan 2-7 for audio-driven mouth sync on portraits, HappyHorse 1.0 for text/image-to-video with in-pass audio, and Seedance v2 Pro for multi-modal cinematic generation with reference audio and subject. The skill classifies user intent (UGC voiceover, virtual presenter, dubbed product demo, lip-synced character, dialog scene) and selects the appropriate model.
Key Features
- Intent-based model selection - Automatically routes to the right model based on what you are building: OmniHuman for full-body avatar from portrait + audio, Wan 2-7 for mouth-sync on existing portraits, HappyHorse for text/image-to-video with audio, Seedance v2 Pro for cinematic multi-modal generation.
- Portrait-to-video generation - Feed a single portrait image and an audio file to produce a video where the subject speaks, sings, or gestures naturally with full head, mouth, and body movement.
- Multiple quality tiers - Choose between premium endpoints for hero-quality output and faster/cheaper tiers for iteration and drafting, with the skill guiding the selection based on your stated use case.
- Documented prompting patterns - Each model route ships with its documented prompting format and the exact
runcomfy runinvocation, removing guesswork from API parameter construction.
Who is this for?
- Marketing teams and content creators producing UGC-style voiceover videos, virtual presenter content, or dubbed product demos at scale
- Developers building automated video generation pipelines that need audio-driven talking-head output from static portrait images
- Educators and communicators who want to generate speaking avatar videos from scripts and portrait photos without filming
Same repository
agentspace-so/runcomfy-agent-skills(30 items)
Installation
npx vibeindex add agentspace-so/runcomfy-agent-skills --skill ai-avatar-videonpx skills add agentspace-so/runcomfy-agent-skills --skill ai-avatar-video~/.claude/skills/ai-avatar-video/SKILL.mdSKILL.md
More from this repository10
A RunComfy skill that generates images using Google Nano Banana 2, the flash-tier text-to-image model in the Gemini family. Optimized for rapid iteration, social thumbnails, and in-image typography with configurable resolution tiers and safety tolerance.
A smart intent-routing skill for image editing on RunComfy that selects the best model based on the editing task. Routes to Nano Banana Edit for batch edits up to 20 images, GPT Image 2 for multilingual text rewrite, Flux Kontext Pro for single-shot precise edits, or Z-Image Turbo for mask-driven inpainting.
Provides Kling 3.0 video generation on RunComfy, covering all six endpoints across three quality tiers (Standard, Pro, 4K) and two modes (text-to-video, image-to-video) for Kuaishou's third-generation cinematic video model with native synchronized audio.
Edit images with Google Nano Banana 2 on RunComfy, supporting batch edits of up to 20 images per call with strong identity preservation. Features localized edits using spatial language, background swaps, and configurable resolution up to 4K.
Generate text-to-video with Wan-AI's Wan 2.7 on RunComfy, featuring multi-reference conditioning and audio-driven lip-sync via custom audio tracks. Supports prompt expansion, negative prompts, and up to 1080p resolution through the RunComfy CLI.
Edit images with OpenAI GPT Image 2 on RunComfy, excelling at multilingual in-image text editing across any script (Latin, kana, CJK, Cyrillic, Arabic) and multi-reference composition with up to 10 input images. Ideal for identity-preserving edits and layout-precise repositioning.
Generate text-to-video with HappyHorse 1.0 on RunComfy, currently ranked #1 on Artificial Analysis Video Arena. Supports native 1080p with in-pass synchronized audio, multi-shot character consistency, and 6-language prompt support via the RunComfy CLI.
Generate cinematic short-form video with ByteDance Seedance 2.0 Pro on RunComfy, supporting multi-modal references including up to 9 images, 3 videos, and 3 audio tracks. Features native lip-synced audio generation and is ideal for brand-consistent multi-language narratives.
A RunComfy skill for generating images with Black Forest Labs' Flux 2 Klein, the distilled low-latency variant of Flux 2. Supports 9B and 4B model variants with sub-second inference for real-time art direction, rapid concepting, and multi-reference brand styling.
The foundation skill for the RunComfy platform, providing a single CLI to install, authenticate, and invoke hundreds of model endpoints including image generation, video, face-swap, lip-sync, and LoRA training.