I have some new and amazing upscale workflows available on my Patreon! Anyone joining the "Creators Lounge" tier also gets access to my Discord, for more workflows, images and ideas. Click here...
GalaxyTimeMachine AI Imagery | AI art in the form of digital images | Patreon
XLPlus_v3.5 is the latest version of my SDXL workflows.
This is the workflow used to create the example images for my latest "XLPlus_v3.5" model. It includes an LLM prompt enhancer, Portrait Master nodes, hands/face/eye detailers and a controlnet tile upscale.
In v4.0, the "KRestartSampler" node can be installed from:
https://github.com/ssitu/ComfyUI_restart_sampling
The dequality node is included in the zip file, and you just need to add it to your custom_nodes directory and restart ComfyUI.
v2.x now uses the new ComfyUI nodes for Cascade. Make sure you update it and install the models as recommended:
https://huggingface.co/stabilityai/stable-cascade/tree/main
stage_b.safetensors and stage_c.safetensors (you can use the bf16 versions) go in ComfyUI/models/unet
stage_a.safetensors goes in ComfyUI/models/vae
and this: https://huggingface.co/stabilityai/stable-cascade/blob/main/text_encoder/model.safetensors goes in the ComfyUI/models/clip/ folder to be loaded with the Load CLIP node in the workflow
The SDXL workflow includes wildcards, base+refiner stages, Ultimate SD Upscaler (using a 1.5 refined model) and a switchable face detailer.
Now with controlnet, hires fix and a switchable face detailer.
I'd recommend installing all the custom node packs shown in the resources, and also these:
https://github.com/chrisgoringe/cg-use-everywhere
https://github.com/bash-j/mikey_nodes
https://github.com/twri/sdxl_prompt_styler
https://github.com/Derfuu/Derfuu_ComfyUI_ModdedNodes
https://github.com/BadCafeCode/masquerade-nodes-comfyui
https://github.com/space-nuko/ComfyUI-OpenPose-Editor
Description
A few changes in this one. This now uses the ComfyUI Cascade checkpoints, which include the CLIP and VAE. They can be downloaded from the link below and need to go into your ComfyUI/models/checkpoints directory.
https://huggingface.co/stabilityai/stable-cascade/tree/main/comfyui_checkpoints
Version 4.0 also includes a "RestartKsampler" and a second stage C sampler, so you can play around with different samplers/schedulers on each of these. There is an SDXL face and hand detailer, which helps with subjects that are slightly further away, but isn't usually needed for closer portrait type images.
The "KRestartSampler" node can be installed from:
https://github.com/ssitu/ComfyUI_restart_sampling
FAQ
Comments (15)
in v4.0 dequality.py is missing in .zip and selecting stageloader are pointing to checkpoints folder
Thanks, it's in there now. You can just remove it or get it from the previous version if you need to :)
but all 3 stage loader still pointing to ceckpoints folder. this only happens in ver 4
@LeoDonBin Yes, have a look at the "About this version" text.
I keep getting the following error, any idea why?
Error occurred when executing VAEDecode:
Given groups=1, weight of size [384, 4, 1, 1], expected input[1, 16, 27, 41] to have 4 channels, but got 16 channels instead
I have the same error, please lmk if you find the fix!
Have you downloaded the correct ComfyUI checkpoint models?
@galaxytimemachine Yeah I downloaded the ones from "About this Version"
use the previewer.safetensor or bypass the preview node, its only for outputting to screen.
A very nice process, but it generates the same images, please tell me where to switch so that the prompt that was written is generated
In the prompt node, there is a "Mode" option which can either be "Fixed" or "Populate", change it to "Populate".
Thank you very much, that's much better)
Love the quality of this workflows, but I can only make portraits. Is it the checkpoints that does this, or is it some config on the workflow? The example of the alien in the room is cool, but I am not able to recreate it. Only male and female faces.
It's all controlled by the prompt, so should give you whatever you type. Make sure the prompt node is set to "populate" and not "fixed"
For a beginner, it's like being thrown into the deep end.