Simple Inpainting Workflow for Z Image Turbo (ComfyUI)

Note: If you use the attached workflow you will need to install the “Image Compare” node via ComfyUI Manager.

I wanted to share a very simple inpainting method I’ve been experimenting using with Z Image Turbo. I’m still fairly new to both ComfyUI and Z Image, so this workflow is intentionally basic. When I was learning, I struggled to find anything that focused on just straightforward inpainting without extra complexity.

One thing to note: since this workflow does not use a separate latent image node, the generation size will default to the size of the image you load. That said, for quick edits and simple inpainting, this setup has worked well.

Example of the workflow:

Before:

After (Masking could have probably been better on my half here for this):

Manual Setup Steps (Updated to set both Max Scale and Batching)

If you prefer to build it yourself instead of importing a workflow, here are the steps:

Base setup

• Start with the ZiT Image to Text template.

• Delete or disconnect:

  • EmptySD3LatentImage

Add the inpainting nodes

• Add a Load Image node

• Add a VAE Encode (for inpainting) node

• Add a Repeat Latent Batch node

• Add a ImageScaleToMaxDimension node (to control input/output sizes)

Connections

• Load Image → VAE Encode (for inpainting)

  • Image → Pixels

  • Mask → Mask

• Load Image → ImageScaleToMaxDimension

  • Image → Pixels

• ImageScaleToMaxDimension → VAE Encode (for inpainting)

  • Image → image

• Load VAE → VAE Encode (for inpainting)

  • VAE  → vae

• VAE Encode (for inpainting) → Repeat Latent Batch

  • Latent → Samples

• Repeat Latent Batch → KSampler

  • Latent → latent_image

Continue as normal

• From KSampler, continue the workflow into VAE Decode and your usual preview/output nodes.

Optional: Image Comparison

(Optional – install the “Image Compare” node via ComfyUI Manager)

• Connect:

  • image_a → IMAGE from ImageScaleToMaxDimension

  • image_b → IMAGE from VAE Decode

This allows you to quickly compare the original image to the inpainted result.

Update: I attached a slightly more updated, corrected workflow. I added the ability to set the resolution of the file and ability to batch. The attached workflow was updated. I had also had a Lora set for testing that also can confuse people It was not needed for this.

Instructions should be updated to reflect a better version of the workflow. I consider all of this to be a kind of WIP as I too am new to all of this with ComfyUI.