So before we dive too deep into the workflow itself, let me quickly explain what VACE actually is — because if you’re using this model in ComfyUI, it’s good to know what makes it special.
VACE isn’t just another video generation tool. It’s built to handle a bunch of different tasks all in one package. That’s why people are getting such insane results when combining motion, reference images, and prompts.
Here’s what it supports:
That opens up a ton of creative freedom.
Wan 2.1 FusionX VACE ComfyUI Workflow
Let me tell you — when I tested the Wan 2.1 FusionX VACE ComfyUI workflow with video input, the results were insane .
Like, I’ve tried other image-to-video tools before, but nothing compares to this.
Why This Workflow
Models You Need to Use
In my previous test Wan 2.1 FusionX ComfyUI Workflow , I used two models:
- Text-to-Video (T2V) — for generating videos from text prompts.
- Image-to-Video (I2V) — for animating reference images.
The key addition is the WanVideo Vance model node . You’ll need to:
Workflow Nodes
So yeah, let’s dig into how this workflow actually works in ComfyUI.
There are two main nodes you need to understand:
By default, the frame load is set to 125 frames , which equals about a 5-second video . If you want something longer, just increase that number. Easy enough.
Then both of those get sent through the Resize Image v2 node . This lets you set the resolution you want to generate at. For example, I used 1024×576 , but my source video wasn’t exactly that size. So I toggled “keep proportion” and selected “stretch” in preview mode. That way I could still capture all the movement clearly — like the dance steps I was testing with.
Motion Capture Setup
After resizing, I fed the output into both the Canny and DWPose ControlNet nodes.
Why both?
Because combining them gives better motion tracking — especially for things like dancing or sword movements. But yeah, it does eat up more VRAM.
If you’re running low on memory, here’s what you can do:
Still You will gets good results, just a bit less detailed.
And if you’re not using a reference image at all?
Now it’ll just copy the motion from your input video, and generate visuals based on your prompt. So the look will be AI-generated, not tied to any specific image.
Workflow Video
Common Errors (and How to Fix Them)
Okay, let’s talk about the errors I ran into while testing this workflow — and how I fixed them. A lot of people asked about these in comments, so I figured it’s worth covering step by step.
First thing to check:
Make sure sage attention is installed correctly. That’s a common bottleneck for this model.
If you don’t have it working:
Sometimes just turning that off gets things running smoother.
Another quick fix:
Also — if your setup includes the WanVideoBlockSwap node:
And here’s the most important one:
Final Thoughts
Alright, that’s pretty much everything I’ve tested so far with the Wan 2.1 FusionX VACE ComfyUI workflow .
Let me just say — this is one of the most flexible video workflows I’ve used in ComfyUI so far. Whether you’re animating a single image, swapping characters, or building entirely new scenes from motion input, it handles it all pretty damn well
