Master WAN 2.2 in ComfyUI – Complete Video Generation Workflow

Video Description

In this tutorial, I walk through a complete, end-to-end WAN 2.2 workflow inside ComfyUI — the exact pipeline I use for real video generation, from short clips to long-form outputs. I explain why each technique exists, when to use it, and how everything fits together in a single, reliable workflow. 🔧 Topics covered in this video: Text-to-Video (T2V) Image-to-Video (I2V) Standard Sampling Triple-K Sampling ClownShark Sampling Sigma Shift (what it does and why it matters) Mixture of Experts (MoE) Frame Interpolation SVI long video generation Post-processing workflow NAG Self-forcing LoRAs The focus is on practical usage, not theory — helping you understand how to build a clean, scalable WAN 2.2 pipeline that actually works. If you’re experimenting with WAN 2.2, ComfyUI video workflows, or long-form AI video generation, this tutorial is designed to give you a solid foundation. 🔗 Resources GitHub (my scripts and custom nodes): https://github.com/ArcticLatent Patreon (all my free workflows + supporter tier): https://www.patreon.com/cw/ArcticLatent #comfyui #wan22 #aivideo 00:00:00 Intro 00:03:02 Sage Attention 00:04:58 Custom Nodes 00:05:48 Settings 00:14:37 Model Shift 00:17:44 Mixture of Experts 00:22:41 Self-forcing Loras 00:30:36 Model/Sigma Shift 2 00:34:22 CFG 00:35:42 Model Loaders 00:40:00 Self Forcing LoRAs 00:42:46 NAG 00:46:03 Latent 00:49:17 Standard Sampling 00:56:21 Video Output 00:58:48 Interpolation 01:01:18 Post Processing 01:08:27 TripleKsampler 01:20:02 ClownSharksampler 01:27:55 Stable Video Infinity 01:28:59 Sub-graphs 01:54:05 Final thoughts 02:01:25 Outro ⚠️ Disclaimer All visuals in this video were generated using WAN 2.2 models developed by Alibaba’s Tongyi Lab (WAN AI). I do not claim ownership of, nor authorship over, the training data used to create these models. The content shown is fully AI-generated and is presented solely for educational and demonstration purposes.