Because this website heavily downscales the previews from 8k i have cropped a few at the full resolution. The images are huge.
This workflow is capable of creating HD quality 8k images, 44mb+ i don't know how much more it can do, my 12gb card is limited there.
You are not limtied to 8k, anything lower works just as well.
Features
- XL or 1.5 base generation - LCM used can be removed.
- 2 base models - 2 extras for use with enhancers
- Full upscale with 1.5
- Full subject cutout and enhance
- Decent hand fixing with meshgraphormer
- Full face with hair cutout enhance
- inner face enhance
- face features enhance (eyes, nose, mouth)
- Aux detector to enhance something else (breasts, clothing, etc)
- All enhancement steps use cutouts to avoid full scale renders
- Any relevant areas have optional inpainting
- Any relevant areas have optional IP adaptor Refs
- All areas have separate prompts/Lora
Updates
V1.0
- fixed faces not working for multiple people
- fixed padding issues that came up rarely but caused improper blending
- fixed face detection drop size being too large, made drop size a main option due to its usefulness
- face parts enhance each option on its own now. Can do just 1 eye, or just mouth.
- added custom subject paste back to avoid bad paste orders on some photos.
- added 1 extra aux detailer because we all know why you need 2 :P
- added reactor face swap, it's not great on 8k but decent on others
- added model upscale option
- added Downscale for all previews, so my browser would stop freaking out. Turns out having 10 8k previews makes it freak out.
Beta 1 - release
Default detectors needed
Full head
https://civarchive.com/models/326568/anzhcs-head-segmentation-prototype-or-yolov8-or-adetailer-model
hand detector
https://civarchive.com/models/329458/hand-detailersegmentation-adetailer
Control nets used
hand controlnet - makes bare hands
https://civarchive.com/models/262788/controlnet-handrefiner
This can be used to give gloves.
https://civarchive.com/models/9868/controlnet-pre-trained-difference-models
Description
V1.0
- fixed faces not working for multiple people
- fixed padding issues that came up rarely but caused improper blending
- fixed face detection drop size being too large, made drop size a main option due to its usefulness
- face parts enhance each option on its own now. Can do just 1 eye, or just mouth.
- added custom subject paste back to avoid bad paste orders on some photos.
- added 1 extra aux detailer because we all know why you need 2 :P
- added reactor face swap, it's not great on 8k but decent on others
- added model upscale option
- added Downscale for all previews, so my browser would stop freaking out. Turns out having 10 8k previews makes it freak out.
ton of other changes. i think its working really well.