I've been going through dozens of models these past few months, but none of them seem to be able to do official art looking or artist-style pics without looking off in some way. I've just been using this lazy merge I made a while back, which produces stuff I like more consistently, so I decided to toss it here for fun in case someone else likes it. It does SFW and NSFW pretty damn well, IMO.
Unfortunately, its dataset isn't that recent since the merge is around 6 months old, but it's pretty lora friendly as long as they're made well.
Use your usual quality tags and all that jazz. Using anime coloring alongside official art and the name of the anime will most times produce a close to identical look of the show. I also recommend using this lora by LittleJelly.
It adds a bit more of that "official" look and clarity if you want it.
Description
FAQ
Comments (9)
let me know when i can generate with this
I am unfortunately buzzless, so I can't even bid on my own model to get it up there, lol.
@megor96 what buzz im low on yellow but i have lots of blue
@megagundam777932 Yellow, which seems to be the only one you can use to bid.
@megor96 thats Lowsy
I cannot recommend it for users of high-speed LORA as it is not a good match.
This is an amazing checkpoint, well done
Do you have recommended scheduler, steps, cfg, positive, negative, etc? What was it mixed from?
I can't remember the exact models since it was months ago, and I have no idea how to check or if it's even possible. I personally gen with 30 steps, cfg around 4-6 (doesn't change much for me), Eular a, and automatic schedule type according to forge. The positive and negative prompts work fine with the basic quality tags and your prompt, no special or annoying pony-style prompts.