Wan ComfyUI Workflows

If you are comparing Wan ComfyUI workflows, the first thing worth knowing is that Wan is genuinely open. It is Alibaba Tongyi Lab's open-source video family, released under Apache 2.0, which is why you can run it locally on your own hardware instead of renting it by the second. The workflows on this page are the ready-made graphs for doing exactly that: load a Wan model, give it a prompt or an image, and generate a short clip.

Wan covers the jobs people actually reach for. Text-to-video turns a written prompt into motion, image-to-video animates a still you provide, and the newer S2V variants drive a portrait from an audio clip with lip-sync. Across the family you get high temporal consistency, bilingual prompts, and multiple aspect ratios at 24 frames per second, with clips that typically land around 81 frames, roughly three and a half seconds.

Versions matter when you pick a workflow. Wan 2.1 ships in a light 1.3B size that runs on about 8GB of VRAM and a 14B size that wants 24GB or more, both at 480p and 720p, plus Fun variants for camera control and depth, pose, or canny conditioning. Wan 2.2 moves to a mixture-of-experts design for better quality at the same compute, adds a hybrid TI2V-5B model that runs on a single 4090, and includes Animate for character animation and subject replacement. Higher steps and a moderate guidance scale give cleaner motion, and the workflows expose those settings so you are not guessing.

Everything here is open and editable. Run a Wan workflow on Comfy Cloud with nothing to install when you do not have the GPU for it, or download the graph and run it locally for full control. Either way you can swap models, tune the parameters, and re-run until the clip is right, with no watermark on the result.