What is the recommended sampler and scheduler for this workflow?

For lightning-based WAN 2.2 models, the skill uses the uni_pc sampler and beta scheduler with a two-pass approach (steps 0-2 and 2-4).

How does the skill handle image resizing?

It utilizes ImageResizeKJv2 to ensure both the start and end frames match target video dimensions exactly before encoding, preventing resolution mismatches.

Why is the Dual Hi-Lo architecture required for WAN 2.2?

Unlike previous versions, WAN 2.2 I2V was trained with separate HighNoise and LowNoise components. HighNoise establishes structure, while LowNoise refines details; using only one produces broken or low-quality output.

Does this skill support lightning models for faster generation?

Yes, it is optimized for Remix NSFW models with built-in lightning and provides specific configurations for external lightning LoRAs with GGUF or FP8 variants.

WAN 2.2 Video Workflow Generator

Name: WAN 2.2 Video Workflow Generator
Author: artokun

byartokun

•

Data Science & ML

Automates the creation of high-quality First-Last-Frame video transitions using the WAN 2.2 dual-noise architecture.

This skill empowers developers to build and execute sophisticated WAN 2.2 First-Last-Frame (FLF) video workflows within ComfyUI via Claude Code. It specifically addresses the critical Dual Hi-Lo architecture required by the 14B I2V model, automating the configuration of two-pass KSampler setups, model sampling shifts, and specialized CLIP vision encoding. By managing the complex interplay between HighNoise and LowNoise models, this skill ensures smooth, high-fidelity image-to-video transitions that maintain structural integrity and fine detail across the entire generated sequence.

Key Features

01Optimizes ModelSamplingSD3 parameters and lighting LoRA stacks

0211 GitHub stars

03Handles complex CLIP Vision encoding for both start and end reference frames

04Configures automated two-pass KSamplerAdvanced workflows for refinement

05Implements required Dual Hi-Lo split-noise architecture for WAN 2.2 14B

06Supports Remix NSFW, GGUF Q8, and Official FP8 model variants

Use Cases

01Building high-fidelity AI-driven motion content with precise frame control

02Generating seamless cinematic transitions between two static reference images

03Automating complex ComfyUI video production pipelines via natural language

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add artokun/comfyui-mcp wan-flf-video

For use in Claude.ai and ChatGPT

Download Skill