Skip to content

Krea 2 Edit Workflow

This guide adds an optional ComfyUI edit workflow that can preserve a character's likeness. It builds on the Krea 2 setup and is only one way to use a reference image. For how reference images behave everywhere else, see Reference images.

Before you start

  • Complete ComfyUI Setup.
  • Use the Krea 2 setup from that guide.
  • Plan for at least 24 GB of VRAM.

Install the custom node

The workflow uses a node that ComfyUI does not include:

cd ComfyUI/custom_nodes
git clone https://github.com/lbouaraba/comfyui-krea2edit

Restart ComfyUI after the clone. The node pack does not need other Python packages. If Orb reports Krea2EditGroundedEncode as missing, check the folder and restart ComfyUI again.

Download the models

Use the files from the Krea 2 section of ComfyUI Setup when you already have them. The edit workflow adds this LoRA:

  1. Download Krea 2 Identity Edit.
  2. Put krea2_identity_edit_v1_2.safetensors in ComfyUI/models/loras/.
  3. Use the largest available version.

Restart or refresh ComfyUI after installing the files.

Test the workflow

  1. Open KreaEdit_Default.png in ComfyUI.
  2. Upload a source image in the empty Load Image node.
  3. Write an instruction describing the change, such as she is now wearing a red coat.
  4. Select Run and check the result.
  5. Save the output as a PNG. The PNG contains the workflow metadata.

You can also import the included PNG directly into Orb.

Import into Orb

Follow Import a ComfyUI workflow. During import:

  • Use the Krea2EditGroundedEncode node containing the positive prompt.
  • Use the empty Krea2EditGroundedEncode node for the negative prompt.
  • Leave Width and Height as None. This workflow takes its output size from the reference image.

After assigning the workflow to a style, set that style's Reference image to Character references. If it stays Off, the workflow keeps the example filename from the exported PNG, which is not present on your server.

Continue with Image Generation.