🤖 ComfyUI-Expert — Claude Code

A session-scoped Claude Code agent with 12 specialized skills for AI video production

ComfyUI-Expert is an agent built on Claude Code that brings 12 specialized capabilities directly into your ComfyUI workflow. From image generation and voice cloning to LoRA model training and automated publishing — all managed by a single agent that knows your context.

This is not a bundle of scripts or a one-off AI call. It is a session-scoped agent that remembers the full context of your production and connects individual skills into a coherent creative pipeline.

🎯 What the agent can do

ComfyUI-Expert covers the entire production workflow:

image generation and editing (SDXL, Flux, ControlNet, IP-Adapter)
video generation and interpolation (AnimateDiff, Wan2.1, CogVideoX, SVD)
voice cloning and audio synchronization with video lip sync
training custom LoRA models with automatic dataset preparation
upscaling, inpainting, and batch processing
export and publishing to YouTube, Civitai, or Hugging Face Hub

Each of these areas is handled by a dedicated skill with its own configuration and logic. The agent dynamically selects the appropriate skill based on your request.

⚙️ What session-scoped means

The agent’s key feature is session memory. In practice, this means:

the agent knows which models and checkpoints you prefer
it remembers quality settings and previous projects
you don’t have to reconfigure workflows every day
it preserves full production context when switching between skills

The agent runs as a Claude Code process with direct access to local ComfyUI. It reads and writes workflows, executes Python scripts, and communicates with external APIs — all from one place without manual tool switching.

🧩 12 specialized skills

The agent includes the following modules:

Image Generation — model management, seed management, prompt optimization
Image-to-Image — ControlNet, IP-Adapter, style transfer
Video Generation — AnimateDiff, Wan2.1, CogVideoX, temporal consistency
Video Editing — interpolation, upscaling, batch queue management
Voice Cloning — RVC and Tortoise TTS, lip sync synchronization
LoRA Training — dataset preparation, training, loss curve monitoring
Upscaling — Real-ESRGAN, intelligent model selection based on content
Inpainting & Outpainting — SAM segmentation, contextual filling
Audio Synchronization — beat detection, audio-driven visuals
Prompt Engineering — optimization, A/B testing, prompt library
Batch Processing — parallel queues, dynamic VRAM allocation
Publishing Pipeline — export, metadata, thumbnail generation

🚀 Typical workflow from prompt to publishing

A practical AI video creation example shows how the agent links skills into a single pipeline: it first optimizes the prompt, generates consistent keyframes with your character or LoRA model, animates them via AnimateDiff or Wan2.1, mixes dubbing and synchronizes audio with video, performs upscaling and color correction — and finally uploads the result to the chosen platform including an automatically generated thumbnail and metadata.

This entire chain, which would otherwise require switching between dozens of tools, can be managed autonomously by the agent from a single instruction.

🖥️ Local operation, no data to the cloud

The agent runs completely locally within ComfyUI. All models, data, and outputs remain on your computer — with no API generation fees and no content sent to external servers. This is a major advantage over cloud solutions, especially when working with commercial or sensitive material.

Requirements are minimal: a running ComfyUI instance, Claude Code CLI, and Python 3.10+. For video generation and LoRA training, a GPU with at least 8 GB VRAM is recommended.

📦 Installation and repository

The project is open source under the MIT license and available on GitHub. Installation is done via standard pip install and launching Claude Code with an MCP server. New skills can be added as standalone Python modules without modifying the agent core.

Repository: github.com/MCKRUZ/ComfyUI-Expert