This workflow allows you to use the best of both worlds:
(24gb GPU is MANDATORY)
Optional: Access to 10,000 LUTs for post-processing (Yes, it's a lot!)
Place them here:ComfyUI\custom_nodes\ComfyUI_LayerStyle\lut
Key Features
Diffuser and Upscaler Options:
Use SDXL or XLL as a diffuser and SD3 as an upscaler/detailer.
Use SD3 as a diffuser and SDXL or XLL as an upscaler/detailer.
Single Workflow Solution:
Eliminate the need for multiple workflows with a simple switch toggle.
Enhanced Detailing:
Face/Eye detailer.
Before/after comparison between all passes.
Four-node setup with ON/OFF switches:
Preserve First pass breasts after upscaling
Preserve First pass Skin after upscaling
Protect background depth of field from being overly sharpened by the upscaler (preventing issues with bokeh, light leaks, and other camera lens effects).
Ensure eyes are not degraded during the first pass upscaling before face refinement.
Custom Control Panels:
Post-process control panel with ON/OFF switches.
Upscaling control panel to activate/deactivate and combine up to six different upscaling passes using model (only the last post-process second pass will increase the image size by 2x).
Additional Features
A clean, well-documented workflow with descriptions.
Thoroughly refined default settings for optimal performance and fast usability.
File dependencies are listed in the Link textbox.
https://buymeacoffee.com/designedbycrt
Let's cook!
Description
FAQ
Comments (3)
Can you please make a clean run with the correct settings and save the workflow and upload it here?
Unfortunately there are a few errors.
Your Lora: XL_SPO_lora fp32.safetensors cannot be found on the net.
You have renamed your eye detailer which is pretty stupid for people... but ok you have the right file in your info field. But your postprocess is a complete disaster, no setting fits. I have now needed 10 runs to get a run without errors. But everything that comes out after the postprocess remains white.
You've really done a lot of work on the workflow and I think it's really good... but now it also has to be able to generate images.
V1.09: Very good, thanks for the changes, it feels a bit more fluid. I have now built my Watermark Nodes into the output and I like it :)
Thank you! Learned so much from your workflow!!!