Yet another Shexyo LORA.
Trying a way to train LORA's without having to manually paint out 50 million watermarks just through tagging. It worked.
Use 0.8-1.2 depending on Checkpoint.
Drop a like and post your gens!
Adetailer for faces
Img2img upscale
4x foolhardy Remacr
Description
FAQ
Comments (14)
Actual prompt examples would be useful...
complain to civit, i absolutely cannot be fucked typing out every prompt for every image because they wont import. Im sure u can figure out some text on your own its not that complicated
@bigmantingzz what's the point of those sample images then ? It's not like people can replicate them, so they prove nothing.
@BigBossXXX its an artists style lora, your not supposed to be replicating whatever i showcase. You just use the triggerword over whatever you want to generate and it applies the persons artstyle to your prompt 😭. The images show the artists style therefore it works = proves it. Man i cant believe some of yall are too lazy to even type 💔🥀
@BMTZZ No, my point is that you could as well put images from the artist at this point, since we have no proof that those images are actually generated using your lora. Also, it's not simply a matter of prompt, there's also how many stesps, which scheduler, which sampler, which cfg, which model ? Replicating the exact feel of 1 image may require hundreds of trials. One could for example use img2img on an original shexyo work and pretend the lora is really great and then, when other people can't replicate that pretend they're just bad.
@BigBossXXX holy yapper
Good work but has issues with resolution.
Examples are 1072x1608 (536x804 before upscale) and generating outside that specific scale creates anatomy issues with the muscle creases specifically, was the training dataset made up of images of that specific resolution? Generating at standard 832x1216 leads to pretty bad deformities :/
idk twin seems to work fine on my end, some other guy posted aswell seems to work fine for him too. Dont gen on 1.1kx1.6k or 500x800 🤷
Maybe the best Shexyo Lora on the site! But my generations look a little worse then yours, it would be nice to have the complete medatata of at least one image.
Could you post the complete metadata of at least one image, please? I can't seem to be able to get the same quality.
I hope you can make a LORA in the style of the artist jk_art07
i told you nah already on the other model twin 😭