This workflow is optimized for two different scenarios
Fast and mass generating when nodes for Face Detailing and SD Upscale are turned off.
Polishing when all nodes are enabled.
Features of this workflow
LoRA: You can use 3 different LoRAs at the same time in one node.
Face Detailing: All the faces will look good. Works the same as After Detailer extension for A1111.
2 pass Ultimate SD Upscale.
Separated checkpoint for SD upscale: You can generate on one checkpoint and upscale on another checkpoint to add more details. Look at the example.
Separated prompts for base image and upscaled.
Custom filename: weight x height + checkpoint name + upscale checkpoint name + upscale model.
You can fix the seed for base image and set queue x10 to receive 10 different variations of upscaled and enhanced images.
Ways to turn off nodes
RMB on node — bypass (on menu).
Unlinking the node in the SD "Upscale ON/OFF" group.
Troubleshooting
If you are using portable ComfyUI
Move file run_gpu_updates.bat from archive to ComfyUI_windows_portable folder (near run_cpu and run_gpu)
Run it to update Comfy UI and all of your nodes.
If GIT installed to the system and added to PATH
Shut down ComfyUI server. Close browser tab with ComfyUI.
Move file troubleshooting.bat from archive to \ComfyUI\custom_nodes folder.
Run it to update all of your nodes.
Waiting for the finishing of this script.
Restart ComfyUI server.
If problem with 7x TXT Loader Concat node just add the empty text nodes to the slots.
Plans for the next versions
Add support of styles.csv from A1111 so that it can be viewed in EXIF as part of the prompt, not a style name.
XYZ-plot or something similar.
Switching the aspect ratio of an image in one click.
ControlNet full support.
2nd pass detailing for body detailing.
Regional prompting.
Description
Added ControlNet Tile for upscale.
Sampler changed form DPM++ 2M Karras to DPM++ 3M SDE Exponential for generating better images.
FAQ
Comments (22)
Any chance you'd add a feature that lets you set 10 separate prompts, it picks from them randomly or in order, so that a batch of 100+ images could be ran at once?
I'm not using random prompts. But I can suggest way to adding to the queue as many prompts as you need. Just remove the prompt, type another, set "extra options - batch count - 100" and push the "queue prompt". Repeat that.
@capybara_paws apologies, I didn't mean to suggest that you were using random prompts.
I suppose my understanding is well behind. I do not know anything about an extra options, or queue prompt. I tried doing digging, but the wording of my request is tricky.
Ideally, one click - 100 total images
20 - A red flower
20 - A blue apple
20 - A green zebra
20 - A yellow dog
20 - An orange pencil
Does that make sense? Thanks for humoring my comment
@Gradasho Will try but can't promise.
When loading the graph, the following node types were not found:
CR LoRA Stack
CR Apply LoRA Stack
CR Seed
Nodes that have failed to load will show as red on the graph.
Try to use troubleshooting instruction from the description.
Or try to install Comfy Roll pack manually https://github.com/RockOfFire/ComfyUI_Comfyroll_CustomNodes
Sweet dreams, dear Capybara... Heaven takes best of us.
Author died 09\22\2023 due accident. Information taken from author`s TG channel.
wait wtf seriously?
@spicybottlebrush333 sadly, its true. A burial ceremony was held today. He was only 31 yo.
That is so sad :(
Oh my fucking god... So sad...
R.I.P. Thanks from all community.
Жаль, что так вышло... Мы будем помнить тебя Глеб!..(
He was a good man, talented and dedicated. Did a lot and could have done so much more. Was hit by a train, they say.
Huge loss for the whole community.
Kyrie eleison.
Вечная память.. Капибаровы лапки кто знал, тот не заббудет...
Ты навсегда в наших сердцах. Ты был одним из лучших
Please help me,When I loading the graph,the following node types were not found: .Text
As far as I remember, .text is from the WAS pack. try putting troubleshooting.bat in the custom nodes folder and wait for it to load. then restart the server. It might help, but I can't guarantee it
same problem
bro 2 hour image generating.It is so slow
There are many amazing pictures by this author. Such a loss for our AI community. RIP.
Thanks to share thsu workflow, but I have an error on Upscale step :
Error occurred when executing UltimateSDUpscale: linear(): input and weight.T shapes cannot be multiplied (154x2048 and 768x320)
Any idea ?

