UPDATE! a new/alternate civitai-trained version (which seems more niji and realism friendly) can be found over here: https://**********/models/694707009028659577
RECOMMENDED WEIGHT is 0.9-1.1!
——————
A style lora aimed at assisting selective color prompts, since sdxl can be very hit and miss with that as a concept.
Helpful words and tags to add with the trigger:
Indicate what color(s) you want to appear
1colorpop or 2colorpop (depending on how many colors you want to emphasize)
Selective color _____ (blue/pink/red/yellow/etc)
A genre “mood” such as “sci-fi mood” or “fantasy mood” (also dystopian, dark fairytale, horror, etc)
Other important notes!
Realism vs stylized prompts: Please note I have found this lora to be less effective with photorealistic styles and it may need to be weighted higher in some cases—(The lora was mostly trained with images in ink/charcoal/pencil drawing styles and graphic novel-esque styles, but I tried to tag in ways to keep that more flexible.) I have personally found that a weight near 1 is best, regardless, if you want a significant difference between a color “pop” and surrounding grayscale.
Some recommended settings:
Weight: 0.9-1.4
Clip Skip: 1 for more realistic prompts, 2 for anime prompts
VAE: I have mostly been using sdxl or sdxl fp16 fix
(To be updated with more recommendations…)
The clip skip recommendation is based off the fact that this was trained with SDXL 1.0. I have found changint the clip skip doesn’t have a profound effect on most prompts, but that it can help complex prompts or prompts combining many loras together.