4.0 bug: If you have more than one male or female face, use 3.0.
This is optimized for speed. The steps are lower because it is not nessesary to have them as high as people usually do for lower denoise upscaling.
This is a first attempt at native upscaling with FLUX NF4 that just released (native in a sense that it uses Flux instead of SDXL, SD1.5 etc.). it is a 2 iterative upscale workflow. If you get memory allocation errors, bypass the second up-scaling column so it skips the 3rd pass. I use 12gb vram for this workflow.
If you prefer using Dev i would suggest still using NF4 upscaling for a speed increase.
flux NF4 download: https://huggingface.co/lllyasviel/flux1-dev-bnb-nf4/blob/main/flux1-dev-bnb-nf4.safetensors
Got any questions you can catch me in my discord server: https://discord.gg/kt2hCJHuKp
Description
Added gender detection for male and female detailers with individual prompts
Added on/off for modules
Added logo/links for DonutsDelivery
FAQ
Comments (9)
It told me to do pip install tf-keras i did that but nothing changed,in which directory do i had to do that?
did you activate your python environment first?
something like source "(your comfyui path)/ComfyUI/venv/comfyvenv/bin/activate"
@DonutsDelivery i used anaconda powershell
@Santaonholidays shouldnt matter what directory you are in with pip installer.
@DonutsDelivery i opened anaconda powershell from the start and wrote pip install keras (dont remember exactly) still didnt worked after a bit of installation
I've noticed a problem. When you have more than one face, the face and eye detailers are working in an endless loop. Try this as an exeample "contact sheet with 4 images of a 20-year-old woman with a messy short bob cut wearing a black turtleneck"
Right, it is caused by getting the resolution from the cropped image to use for the model sampling flux node, so if there's more of the same gender it will feed multiple resolutions to the detailerdebug node and use each of them for all the inpaints. I guess someone need to do a special facedetailer node for flux xD
did i missed something? ur notes point out 3 different uscalers 1 for the whole image 1 for the face and 1 for the eyes...but u can just connect 1 upscalemodel loader... how?
you only need one upscale model. the other models are used for face detection and eye detection to crop it out.