Sign In

SDXL 0.9 Iterative Workflow

Updated: Jul 10, 2023
Jul 10, 2023
Base Model
SDXL 0.9


This workflow and supporting custom node will support iterating over the SDXL 0.9 Base Model + Refiner Model combo, as well as perform a Hires. Fix (approximation) to improve on the quality of the generation. This is more of an experimentation workflow than one that will produce amazing, ultrarealistic images. It allows you to run essentially unlimited number of iterations on an image.

What does it do?

One of the common methods being tested for SDXL 0.9 image generation is running the base model for ~80% of the total steps, and then running the refiner model for the remaining ~20% of the total steps. This process can then be repeated multiple times to adjust the image. The problem is that currently this would require you to duplicate the nodes for every iteration of this. To clean this up, I created an iterative KSampler that supports multiple models (ie. Base & Refiner) as well as separate positive and negative prompt inputs for each. This node technically isn't limited to just SDXL, but was built with that in mind.

Note: I have not found the "best" values for iterations, denoising strength, samplers, base model %, etc. Feel free to try changing these values and seeing what effect it has!

How to install it?

  1. Download the workflow zip file

  2. Extract the zip file

  3. Move the extracted iterative node python file to your custom_nodes folder: ComfyUI/custom_nodes/

  4. Launch (or relaunch) ComfyUI

  5. Load the workflow by pressing the Load button and selecting the extracted workflow json file

  6. Select Queue Prompt to generate an image

Quality of life add-ons

I highly suggest downloading this ComfyUI Manger extension as well for ComfyUI as it will make downloading other extensions MUCH easier moving forwards. In addition to that, if there are any dependencies that you do not already have fulfilled, it can auto-download those for you.

Download URL: