Home
Creators
Search Recent Random
Posts
Search DMs Popular Hash Lookup Tags Random
Importer
Import FAQ
Register Favorites Keys Review DMs Logout
Community
Telegram ThePornDude
Home Creators Posts Import Register Favorites Logout

Multi-Image Qwen Edit! (Patreon)

Published:
2025-09-25 15:46:36
Edited:
2025-09-25 15:54:08
Imported:
2026-09

Downloads

Content

Video - https://youtu.be/F0tpgfnLYbU

Overview

Qwen Image Edit 2509 bring us the power of Multiple Images! Either use them as references to easily add things such as:

Multiple characters

Objects

Backgrounds

Plus, you can also use any of the images as CONTROL images too, giving you control via pre-processing such as:

Depth

Pose

Canny

Personally I've been using image 2 as the control image, so that is why I've added a preview there. For example:

but I've added the AIO Aux Preprocessor to all of them so you can mix-and-match as you like :)

Because the image gets changed automatically to 1024x1024, it can help to set the keep_proportion option, though do play with all of the options available depending on your own requirements.

The workflow is based off the ComfyUI Native implementation, but does require a few extra custom nodes to enable maximum power!

Attachment Information

You will need a few custom nodes for this this one:

KJ Nodes - https://github.com/kijai/ComfyUI-KJNodes

RGthree - https://github.com/rgthree/rgthree-comfy

Controlnet_aux - https://github.com/Fannovel16/comfyui_controlnet_aux

Quickly enable / disable the various options using the Fast Bypass node

Files

Comments

Tyler Durden

Thanks for this, it's a really slick workflow. Could you explain when to use the HiRes Fix option please?

Scott Mann

I'm getting really unsatisfactory results with this workflow... I'm using three input images. A base image in input 1 (a person), an openpose image in input 2, and a bottle on a white background in image 3. I have previews on all the images, and they look to be cropped or preprocessed perfectly (The pose is correct, etc.). Generally I just get (virtually) the same image as input 1 or it might swap the background. But it ignores the pose and just gives me back the exact same pose as the image sent to input 1. I'm using the full e4m3fn model and the Lightning 8step LORA. (I've tried the 4-step too). Haven't really changed any other settings... any idea what could be wrong?