Style LoRA trained on images from Danbooru and screenshots
Tested on WAI-NSFW Illustrious (Linked in Resources Tab)
Recommended Resolutions: 832 x 1216 // 896 x 1152
LoRA to replicate the style of Overwatch characters in-game and in other fan content.
Trigger word to help with effectiveness: ow3d
Seems to work well with a weight of 1 during testing, but you may need to modify this value to find what works for you.
As always, feedback/suggestions are always welcome!
Vanitas Vanitatum et omnia Vanitas
Description
V2: Improved image dataset and tagging for better character representation and more accurate output
FAQ
Comments (7)
sadly doesn't seem to work at all, it's like the lora doesn't even register
Hey, thanks for the feedback. I just checked and it works fine on my end. Are you using a checkpoint based on the correct base model?
@Aeromoia sure am, the one suggested here
@nobody32d It could be due to LoRA weights possibly. Looking at your images you seem to use other character LoRAs that A. might be overpowering the style or B. the style may not be working for a character unfortunately.
Just to make sure I re-uploaded the file with one I’ve tested so a redownload might help
@Aeromoia when testing this one i only used it for loras
@nobody32d Definitely sounds like something funky is going on on your end. Just going by Civit's web-gen it seems to be working fine as the OW model was working as expected with the four different checkpoints I was using for test renders.
@anomedra maaan, this sucks