Workflow for XL/Illustrious/NoobAI Models
Any feedback would be appreciated.
See the Article for a Guide:
https://civarchive.com/articles/17339 (probably doesn't work rn)
https://civarchive.com/articles/17339
Key Features:
LoRA
TIPO Prompting
Detail Enhancers
ControlNet Upscale
IPAdapter
HiresFix
OpenPose
Wildcards
ControlNET
Color Match
FBCNN
Regional Prompting
Inpaint
Outpaint
Workflows included:
馃煡Advanced: comprehensive text to image workflow
馃煥Basic: more basic version of the Standard workflow
馃煢Standard: more basic version of the Advanced workflow
馃煩Detailer: detailing previously generated or existing images
Experiment and enjoy!
Description
Update from V22鈫扸23
(This is not an exhaustive changelog)
General:
General improvements and bug fixes
Adjusted default values
Added CLIP Skip toggle
Updated NSFW detection model
T2I:
Fixed dependency issues
FAQ
Comments (27)
Everytime i use the upscaler (The SD one, built in the workflow),the nod upscale 5 images at once! I dont know why...I just generating one image btw...
The behavior you're seeing is not a bug. Even though you're sending only one input image to the node, Ultimate SD Upscale doesn't upscale the entire image in one go. Instead, it works by tiling. It splits your image into overlapping tiles, upscales each tile, and then stitches them back together. That's why it can feel slow since it's a resource intensive task.
Oh...I get it,thanks for he answer.
I use your basic I2T, v23 is even better, thank you. I still have the odd case where the NSFW adetailer turns nipples into small vaginas. Especially if only one nipple is visible. Should I lower the threshold or is there something I should put into the wildcard? thanks!
nm, i found the civitai page for the detector. it really should be used only on anime, illustrious seems best. And the author suggests if the wrong 'thing' replaces the proper thing, increase crop size. i set it to 5.0 from 3.0 and it seemed better. Using an anime illustrious checkpoint. thx!
Hello, thanks for the workflow! It has a nice presentation and efficiency.
I'd only recommend 1 thing for V24 : Image Filter node (with a On/Off toggle).
Example : Generate 8 pictures, Sound notifications when it's done and you can then pick which one you want to save.
Hello, I am new to using the workflow. I have already learned how to use it a little, but I still find it difficult to use more than two character loRAS in the same image. All their features get mixed up and the image looks very bad. I would like to know how I can fix that and be able to use up to five characters at the same time.
You would need to use a workflow with regional prompting, 2 or more prompts for the same image.
why was lora manager removed from the basict21? it was so good already having it default on there
What do you mean by removed? The Basic鈥疶2I never had the LoRA Manager custom nodes, but I might add them to it in the next updated version.
This workflow is really great, but your documentation is even better. Thanks for putting this together!
Comfy-Core unavailable when using the EasyWAN ComfyUI suite as base.
Hello, thanks for sharing, but i got a "Image Saver
'NoneType' object is not iterable" with BasicT2I_V24, any idea ? I'm not used to comfyui, sry, best regards
Did you make sure to check if you downloaded the correct Image Saver custom node, or if there are any conflicting custom nodes installed as well?
Thank you for your continued efforts not just for the workflow but for the guide and links to resources. Most comprehensive workflow and info I've found on the entire site!
I use the drag and drop feature to drag pngs into the workspace so it loads my workflow (im using your basic workflow). The drag and drop now loads like a generic workflow that doesn't have all the nodes anymore. Wtf. It was working perfectly like a day ago. Any idea what happened?
I am facing the same problem right now.
It's caused by a change in how ComfyUI parses image metadata. There is a workaround for the meantime. Save the images with a different file extension (e.g., .webp or .jpg). Loading them will trigger the custom node's own loading logic (unless you've disabled that in settings), and they should load correctly again.
@Legendaer聽I appreciate your reply. Hopefully this will be resolved soon. That feature saved so much time.
Your workflow is awesome by the way I've been spending a lot of time with it and can't wait to see the update.
Also you can use online ComfyUI Metadata Viewer to withdraw the whole workflow in json.
Excuse me, may I ask if there is a good way to replace the generated character's pupils with another pupil image? For example, when I generate Arknights' Entelechia, her special pupils cannot be correctly generated, so I plan to replace them with a separate pupil image.
Flux.1 Kontext and Qwen Image do such things without a lot of effort
@qek聽Ok,thank you.
Why does Regional prompting ignore prompts written for masks? It only uses the regular prompt. Although the log says that it recognized prompts from Regional prompting.
Thank you for your feedback. It seems like the updated nodes currently convert the colors to masks unreliably, resulting in empty masks (no mask->no region for the prompt). I will rework that part for V26 soon.