SDXL conditioning can contain image size! This workflow takes this into account, guiding generation to:
Look like higher resolution images
Keep objects in frame
Guide image composition to make sense
Installation and dependencies
Install WAS Node Suite custom nodes;
Instal ComfyMath custom nodes;
Download and open this workflow.
How to use
Select image for img2img;
Choose to resize or not;
(optional) Choose Conditioning Scale. It can make your output look like bigger, higher resolution image;
Queue Prompt.
Pro Tip: You can set denoise to 1.0 to use the workflow as usual txt2img, but with size guiding benefits.
Note
I made some research and find out that v1.0 version of the workflow was total bullshit 😅. If you downloaded previous version, don't use it. Better update.
Description
Total rework.
Conditioning parameters are actually helping to get better picture now
Introducing Conditioning Scale
Single node from ComfyMath intsead of giant math group
Crop forces resize
Compact layout
FAQ
Comments (5)
Thanks a million! works perfectly
doesn't work for me, it ignores the prompts and brings me the same image with a bit of noise, what am i doing wrong?
Hard to tell. Maybe denoise is too low. Maybe VAE is wrong.
I've moved the denoise all over the place and tried 3 different VAE, still the same
Maybe it's the model thing? I didn't test it on anything but base SDXL. I'm not sure it even works on anything but base SDXL, because finetunes might totally ignore these extra conditionings.