Wan 2.2 in ComfyUI: image-to-video workflow setup

This tutorial explains a Wan 2.2 image-to-video workflow in ComfyUI, adapted from its default template. The presenter uses an RTX 5070 Ti and says the workflow uses native nodes without additional custom nodes. Earlier tutorials cover ComfyUI Portable and Manager installation; optional Triton and Sage Attention setup is linked separately.

Setup covers downloading the high-noise and low-noise models, paired LightX2V four-step LoRAs, a VAE and a text encoder. Files go into the corresponding ComfyUI model folders, then a refresh makes them available in the loading nodes. The downloadable workflow includes links and placement instructions.

With LightX2V enabled, the workflow uses four steps and CFG 1. Disabling it selects 20 steps and CFG 3.5. The presenter prefers high-noise LoRA strength 2 and low-noise strength 1, reporting less unwanted slow motion with those settings.

The local AI walkthrough also covers starting images, dimensions divisible by 32, prompts and optional LoRAs. Its example uses 81 frames at 16 FPS for roughly five seconds of video. The presenter reports a generation time of about 43 seconds for that smaller portrait example, rather than a general hardware benchmark. He also warns that clips beyond eight seconds can become unusable. The tutorial demonstrates one run and does not compare settings systematically.