ComfyUI Workflow for Animating Parts of an Image: Motion Brush

This Motion Brush workflow allows you to add animations to specific parts of a still image. It literally works by allowing users to “paint” an area or subject, then choose a direction and add an intensity.

In short, given a still image and an area you choose, the workflow will output an mp4 video file that animated the area you chose. Here is an example below:-

We can upload the above image into our ComfyUI motion brush workflow to animate the car

Recommended User Level: Advanced or Expert

One Time Workflow Setup

Let's get the hard work out of the way, this is a one time set up and once you have done it, your custom nodes will persist the next time you launch a machine.

1. Download the workflow here Motion Brush Workflow
2. Launch a ThinkDiffusion Turbo machine.
3. Drag and drop the motion brush workflow .json file into your ComfyUI machine workspace.
4. If there are red coloured nodes, download the missing the custom nodes using ComfyUI manager:

• ComfyUI Path Helper
• MarasIT Nodes
• KJNodes
• Mikey Nodes
• AnimateDiff
• AnimateDiff Evolved
• IPAdapter plus
• ComfyUI Frame Interpolation
• Rgthree

Recommended File – Download these files if not available in your Workflow’s node and restart the machine after uploading.

Download Files Here: https://drive.google.com/drive/folders/1-UzbFUw_tsT0IULuAau82xIJAGGdURCN?usp=drive_link

How to use the workflow

Now that the hard work is out of the way, let's get creative and animate some images. For most use cases, you only need to do the steps that are marked required change

1. Go to Inits and Conds Group Node

Step	Description	Default / Recommended Values	Required Change
Select an image that you want to animate	The first step of the workflow, which allows you to upload an image that you want to animate		YES
Set the image resize and set True in keep_proportion	Keeping the aspect ratio and proportion of image while image is processed throughout the workflow
Set Frame Count	Input your desired frames that will be processed.		YES
Set the load checkpoint	Set the base model for the workflow Choose photonLCM if realistic image Choose everyjourneyLCM if anime image
Set the load VAE	Set a variable auto encoder for your neural network model
Type positive prompt in the text prompt node like moving, waving simple verb, etc.	Type simple verb words in text prompt which serves as a basic movement for your image		YES
Open the mask editor in image and inpaint the area that you want animate	Inpaint an image’s area where you want to animate		YES

2. Go to Latent Mix Mask Group Node

Step	Description / Impact	Default / Recommended Values	Required Change
Set your desired settings in Latent Mix mask for the mask grow settings, blur radius, etc	A node which you tweak the settings for the mask blur on how does it expand, how it grows, how strong its blur		YES
Set the remap mask range	Set the value of animate in comparison to the original		YES
Check the mask preview	Check the mask preview if how vast is the expand of mask blur and how strong is the mask blur

3. Go to IpAdapter Group Node

Step	Description / Impact	Default / Recommended Values	Required Change
Set the clip vision	CLIP vision models are used to encode images.
Set the ipadapater	Adds image prompting capabilities to a diffusion model.
Set the lora loader	Set a lora model

4. Go to ControlNet Group Node

Step	Description / Impact	Default / Recommended Values	Required Change
Set the control_net	It empowers precise artistic and structural control in generating images
Set for the weight	You can play the strength value for controlnet		YES

5. Go to Output Group Node

Step	Description / Impact	Default / Recommended Values	Required Change
Set the motion model.	Set the base model of animatediff
Set the motion Lora	Set the motion lora for animate
Set the multival	Multival is the strength of motion lora. You can set if how clear is motion lora will be applied to the mask area. MultiVal Max is 1.500 and you can play with this value but above 1.500 may display video artifacts.		YES
Set your desired upscale model	You can select the upscale type either pixel or latent. Pixel generate faster while latent generates slower but add more details and good for higher resolution.		YES
Check the video combine node settings	Check for the loop count and frame rate		YES
Check the Film VFI for its multiplier	Set your desired multipler		YES

You can now click queue prompt to animate your still image!

More Examples

Download the input and output files here input / output.

If you’re having issues with installation or slow hardware, you can try any of these workflows on a more powerful GPU in your browser with ThinkDiffusion.

If you enjoy ComfyUI and you want to test out creating awesome animations, then feel free to check out this AnimateDiff tutorial here.