Hey everyone,
I’m sharing my workflow because I thought it might help others see how I create my generations. I know it can be tough when you're just starting out, especially when trying to figure out nodes and workflows, and it doesn't help when creators don’t explain things clearly. When I first started, it was frustrating to see amazing images from creators without knowing how they made them. So, I’m sharing the workflow I use for my models and my posts, hopefully it’s helpful.
This workflow was tested today (5/10/2025) on a fresh install of ComfyUI. You’ll need to download some models using ComfyUI Manager, and some from external sources. I’ll include those in the notes below.
Depending on your VRAM, you might need to tweak a few settings.
What I like about this model is that it lets you control different outputs easily through the control panel.
Let me know if you have any questions or run into issues. Hope this helps, take care!
Features
Batch Selection - Pick the best images to generate for a given batch
Turn on and off nodes easily through the control panel
Test and compare results using FaceDetailer and Upscaler so you can hone in on the best results
Default settings work!
Installation
This workflow was tested using ComfyUI's portable installation: https://github.com/comfyanonymous/ComfyUI
ComfyUI-Manager makes installing the nodes and models easier: https://github.com/Comfy-Org/ComfyUI-Manager
Make sure to update ComfyUI, Manager -> Update ComfyUI -> Restart
Drag and drop the workflow file into ComfyUI
Install Nodes by going to Manager -> Install Missing Custom Nodes
Image Filter - Is having issues with version 1.51. You can downgrade to 1.50. Manager -> Custom Node Manger -> Installed -> Switch Ver ImageFilter to 1.50 or new then 1.51, if there is an update beyond 1.51.
Download Velvet Beauty - Checkpoint. Save Comfyui/models/checkpoints/
Download 4x_NMKD-Siax_200k, ComfyUI -> Manager -> Model Manager -> "4x_NMKD-Siax_200k"
Download sdxl_vae, ComfyUI -> Manager -> Model Manger -> "sdxl_vae"
Download Eyeful_v2-Paired.pt - Eyeful | Robust eye detection for Adetailer / ComfyUI
Save to Comfyui/models/ultralytics/bbox/Eyeful_v2-Paired.pt
Considerations & Troubleshooting
Upscaler - Ultimate SD Upscale
tile_width & tile_height - may need to reduce to 512 or lower if having VRAM issues.
denoise - may need to adjust if the generation is changing your original image a lot or not changing at all. If to much change, reduce the number. If needing more change, increase
Upscale_by - may need to reduce if having VRAM issues, 1.25 or 1.5.
Face Detailer
1st Face Detailer is the face + eyes
2nd Face Detailer is eyes only. Sometimes certain models need both and sometimes they only need the 1st Face Detailer. Turn the 2nd Face Detailer on and off to see what provides the best results.
Image Filter (Batch Selector)
Change Empty Latent Image to 2+ and enable Image Filter to select the best images to generate. Version 1.51 had an issue with selecting images. 1.50 and older versions appear to work.