Simple SDXL T2I & Hi Res Fix (CivitAI Metadata compatible)版本v1.1 (ID: 1049657)

This is a pretty simple workflow that is supposed to simply do 3 things:
Generate images
Upscale the images
Save the images with CivitAI compatible metadata
All while being easy on the eyes (no spaghetti)
Nothing more and nothing less, not much fancy stuff going on here

This workflow uses custom nodes to make the viewing experience better and keep the noodles at a lower count so it'd be best if you have ComfyUI Manager to easily install the nodes. There's a full (i think) list of the custom nodes used in this workflow at the bottom of this description.

Short description:

The left part of the workflow is where you'll probably spend most of your time on, it's where the parameters (steps, cfg, seed, sampler) are as well as model selection and prompting.
(I've added an extra seed node for easier access since the SD Parameter Generator is a bit too bloated for frequently changing parameters otherwise and dont use it as input for the param gen node, itll break)
There is a purple "Fast Groups Muter" node with which you can control if you want to enable/disable the A1111/Forge style Hi Res fix (2nd pass txt2img) part or the image saving part of the workflow

The middle part is just normal ksampler stuff where the magic happens.

After that comes the Upscaling using an upscaler model and then a second ksampler doing the other magic stuff. (below the ksampler is an image comparer node, you can expand it by clicking the grey circle left next to the title of the node if you want to see a comparison between hi res and original image)

On the right are the image saver nodes responsible for the CivitAI compatible metadata. The images are saved in an A1111/Forge fashion (images saved in the folder corresponding to the month they were created in; upscaled images are saved with _upscaled_ in their name)

Custom nodes in this workflow:

• https://github.com/receyuki/comfyui-prompt-reader-node (for Metadata + converter)

• https://github.com/kijai/ComfyUI-KJNodes (mainly Get/Set nodes to reduce spaghetti)

• https://github.com/rgthree/rgthree-comfy (various stuff)

• https://github.com/chibiace/ComfyUI-Chibi-Nodes (prompting)

• https://github.com/jags111/efficiency-nodes-comfyui (ksamplers)

• https://github.com/pythongosssss/ComfyUI-Custom-Scripts (for math expression)

• https://github.com/TheBill2001/comfyui-upscale-by-model (its in the name, reduces node count a bit)

描述:

训练词语: