Use this style to turn everything into neon outlines, or simply to add colorful lighting to your images. Activate with "neon_outlines".
Weight, CFG, etc. heavily depend on the subject, prompt, negative embeddings, model, and so on.
Start at a weight of about 0.8 to find out how your it affects your prompts.
If your image is made of ugly squiggly neon outlines and added blur, you probably need to lower the lora weight by quite a lot and/or add some realism/quality negative embeddings. If your image looks totally normal, you need to raise the weight.
*
Recommendations for starting and/or generating images online:
Example images can be seen in this post:
https://civarchive.com/images/4013328
Dreamshaper, Realcartoon3D and CyberSushi models seemed to produce good results consistently.
Under additional resources, I choose lora weight of 1 to 1.2, and add the negative_hand and fast_negative negative embeddings
Add "in a dark room" etc. in the prompt.
Add a color theme to the prompt (like "pink and teal color theme").
Add "close up" and/or "portrait" to the prompt, especially if the subject is a human or human-like character.
Prompt Example:
Positive:
neon_outlines, award winning photograph of a woman standing in a dark room, close up, blonde, long wavy hair, cute dress, relaxed atmosphere, pink and teal color theme, 8k resolution
Negative:
negative_hand FastNegativeV2
Under "Advanced" settings, CFG scale "Precise" (10), Sampler "Popular" (DPM++ 2M Karras) and steps "High" (46) seem to improve the results
Any VAE is better than none, choose one that fits your model. I chose VAE-FT-MSE-840000-EMA
Description
FAQ
Comments (12)
Looking great! Good job :)
Thank you!
How did you train Lora to be so small? Only 9.1M
to be perfectly honest, I have no idea. I use kohya_ss and leave most settings on default, except for switching to the adamw optimizer. Still don't know enough about loras to understand what's happening
is there a "how to" to create the second video?
Which video are you referring to? I usually only do simple text-to-video stuff in deforum or animatediff.
the image of the girl and the background is constantly changing
@grjmmr284 Assuming you mean https://civitai.com/images/4043982, it's done using Deforum, a simple text to video animation with mostly the default values. 3D animation mode, 16 FPS, FILM interpolation, interp x5.
DPM++ 2M SDE Karras, 20 steps
The prompt was:
neon_outlines, blonde woman with long wave hair, flowery dress, dark room, relaxed atmosphere, pink and teal color theme <lora:neolin3:0.7> <lora:add_detail:0.25> <lora:ClothingAdjuster3:-0.5> --neg lowres, cropped, (worst quality:2), (low quality:2), (normal quality:2), jpeg artifacts, blurry, bad eyes, negative_hand-neg
Not really good settings, but even bad settings look good with this lora because it's only colorful outlines, so it kind of hides defects and everything looks cool instead
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.
















