Update 26.11.24
This workflow eliminates the need for a face and skin detailer. The idea is to render directly at a significantly higher resolution and then correct any imperfections in the image using a simple ultimate upscaling process.
Additionally, image details are enhanced through the Detail Daemon and additional noise injection. In the initial samplers, the sigmas are split to allow the first steps to render with high guidance and the final steps with low guidance. This makes the images appear more natural and less AI-like.
This approach works very well but requires extremely powerful hardware and is not particularly fast. However, the resulting images are of exceptionally high quality.
Update 07.10.24
The workflow now supports GGUF models, improved ViT-L/14 CLIP, and has a significantly enhanced UltimateUpscaler thanks to the fill nodes.
Attention: The skin detailer with upscaler workflow is extremely hardware-intensive. On an RTX 3090, it takes about 10-12 minutes to generate a single image. If you want to generate images faster, please use the older workflow.
Flux is a 12 billion parameter model and it's simply amazing!!! Here’s a workflow from me that makes your face look even better, so you can create stunning portraits.
You can find the download links for the Flux model in the notes within the workflow.
Description
FAQ
Comments (7)
I tried your new workflow, Detail Daemon, and indeed, it is quite heavy. Unfortunately, I had to force-stop at the final stage of upscaling, even though it seemed like that’s where all the magic happens. I’ll stick with the previous version (Skin Detailer with Upscaler), which takes me around 5 minutes.
works like a charm
Curious. On the 4090 it's pretty fast and the result looks better on the skin as for some reason my LoRA has a plastic look to it, that said is it just me or does it modify your hair, and even the clothes around the collar of the headshot was modified and the eyes were even a different color than the original image. is this normal?
Hey. Im trying this workflow for the first time. Unfortunately every time ComfyUI finishes a generation, all i got is black images as results. Did anybody experience this problem?
This WF killed my comfyui installation.
works really good, but why there is no place to add negative prompt? and is this only good for face portraits?
After this workflow, mxtoolbox nodes stopped working for me, and they are in almost all workflow videos, reinstalling nodes does not help. Sliders, maps and other elements are not displayed, just empty boxes.