My first attempt at making a slider.
Do you like an art style but wish it had thicker, more expressive lines? Have a style with thick lines that you wish was lineless? This is the slider for you!
Negative weights = Thicker Lines
Positive weights = Thinner Lines
Recommended settings:
For thick lines, -.5
For thin lines, .5
-.5 through +.5 seems to be the sweet spot for most uses.
The effectiveness will vary based on the line weight of your checkpoint and/or style loras. Experiment at your own risk. Checkpoints with thicker lines need less strength, unless you want dummy thicc lines.
I initially wanted the positive weights to be thicker lines, but the training reversed that, so negative is thicker. I'll invert the polarity fix that in a future update.
Description
Trained on NoobAI XL blah blah whatever the alphabet soup of the original NoobAI is called,
but it works just fine for Illustrious.
For best results, don't use Nova Checkpoints.
FAQ
Comments (50)
Wohooo! Amazing work! ✨
Thanks! :)
Didn't even tried it yet, but i already know that's what i was missing for my pixelart stuff 💜
Let me know if it works out for you! :)
Lora is really awesome :O Super!
Glad you like it! :)
Solves a problem I've been trying to fix for ages. No noticeable style drift, just solid and predictable results. Thanks!
i wanted this! any chance for an Illustrious version?
It should work exactly the same for illustrious
It works on IL. Saved my a**
great results, thanks!!
nice work man!
really good work
Excellent! Thank you so much. Tested with ComfyUI and some checkpoins using Illustrious with not problems, great results.
how do use this in comfyUI
@wbmasteruwu693 Just load it as a regular LoRA. The strength value controls the line weight, and no trigger word is needed. Thin lines work best with values between 0.1 and 0.5, while thick lines can be achieved using negative values from -0.1 to -0.5. Try generating manga art or lineart, adjusting the LoRA strength across different ranges to clearly see the differences.
@kozkiizen beat me to it. Thanks for answering! :)
@kozkiizen I see, but I have tested it and it seems not really working with WAI-illus checkpoint
@wbmasteruwu693 Are you mixing it with other loras? What weights are you trying?
@Bridgewalker I am mixing it with lazy embedding and another lora Ado character. I am trying to make the line thicker using -0.2 to -0.5. However, I did try to only use the linewight lora by itself to no avail.
I like both sliders! Do you have any ideas for other sliders? I don't understand why keywords are needed for line sliders; it seems to work without them.
They aren't needed. They just sometimes help, if the slider doesn't work initially.
As for other ideas for sliders, I've been trying to think of other things that could work.
One is for a glow/directional lighting slider. To increase/decrease the glow of things like fire and flashlights and stuff.
Also maybe a color temperature slider.
I'm open to suggestions. :)
@Bridgewalker Thank you for being open to suggestions.
I can't speak for others and what interests them, but I'll express my personal opinion on what would be cool.
Adjusting the background detail and filling. A negative weight creates a solid background fill, minimal textures, and simplified shapes. The focus is primarily on the character, who remains at normal detail. Conversely, a positive weight adds complexity and detail to the background beyond the norm.
Or a slider that adjusts the era in which the anime was drawn. From, say, the 1950s to modern anime.
@yuduz367 Ooh, a background detail slider would be pretty sweet. I might try to make that next. Could be a while though.
@Bridgewalker I also thought it would be nice to have a skin slider. Typically, each generation gives the character's skin a perfect appearance. That is, the skin is smooth, free of imperfections such as moles, wrinkles, stretch marks, and generally other imperfections that I can't immediately recall.
@Bridgewalker Oh my, I just found this, and it's a great model! I have no clue if sliders are easier/harder than regular loras, but for your second try at one it's impressive ^_^
Re ideas for other sliders, I'm really into the glow/directional lighting idea you mention. But ofc, at your leisure (can't rush artistry, amirite?).
Hands down the best lora of it's kind, been looking for this for a looong time..THANK YOU
Thank you so much :)
A must have. Thank you for Releasing
What an EXCELLENT idea for a lora. I love it. Thank you!
I approve.
Thank you so much :)
God
nice
Z-Image Version when? 😁
I can't run z image locally. Hard to test things :( I might pay for a month of a cloud computer to train a bunch of my things on z image in January though
@Bridgewalker
I'd love to see a LoRA for this! But it might be better to wait for the Base model to drop before starting the heavy lifting. Hopefully, we won't have to wait too much longer.
@Bridgewalker I'd be happy to help, your loras are terrific
I may not understand how it works, but it doesn't work with illustrious checkpoints and loras.
Yes it does. All of my examples were made with illustrious checkpoints and loras.
Are you using the right values? You have to use negatives for thicker lines, and positives to remove lines.
@Bridgewalker The issue is when using an illustrious model and equipping this lora, it changes to noobai and cannot change back.
@gisa ah, that's a bug with civitai, not the model. If you search for it in the additional resources from the generation page, it won't set your checkpoint to Noobai.
@Bridgewalker Noted, thank you.
How did you train a concept like this and manage to keep it so isolated?
I manually inked over images for the training data, so they were the exact same base image, just with lineart applied.
@Bridgewalker Did you use target flow?
I'd use the hell out of this and your toon loras if you made ZIB versions. Stellar work
Wow, works like a charm! Very helpful, exactly what I was looking for. Thanks!
Would you mind sharing datasets of your line weight and/or shading loras? I want to create such slider(s) for Anima but have no skill to make good dataset.
P.S. Thank you for your work!
P.S.2 tried to add lineart from AUX preprocessor to image and then train, but no luck there - quality of result is very poor and not like your model.
We need something like this for anima.
Details
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.