1)The larger character style loras (100mb+) were trained on very large datasets (200-2k images) and they will fight a style heavy base model for style at strength1.0, If YOU WANT TO USE THEM FOR JUST THE CHARACTER, TRY TO USE THEM AT LOWER STRENGTH! It will not always work if the base style is too heavy but it will improve compatibility significantly!

2)While my goal with style loras is to make images beyond the original material's... "area of expertise", I mark my style loras as nsfw if they were trained on a decent chunk of R+ images so people are aware there is a higher risk it makes "accidental" nudity (despite all the naughty stuff being tagged).

3)While I am planning to revisit and make new illustrious versions for loras, it will take time as my gpu is having to do a lot of other stuff, so message or comment if you want a specific one updated and I will bump it up in the queue.

4)I'm using comfy with a messy workflow for image gen (aside from lora demonstration posts), so my metadata is often broken/incorrect, if you want a prompt or data message me!