Hunyuan Lora(s)- Uses same tags, seems to work well. This is experimental to me and will take suggestions to help improve.
Using natural language while including the tags also helps a bit for consistency.
Enjoy!
Illustrious v2 - New model includes more data for base outfit for all characters as well as their alternative outfits (black, blue, and white, serafuku's + vampire outfit and Christmas outfits)
Use "3d" tag to better fit the game's style.
Mita from MiSide
Great freaking game.
This model contains her default outfit only. I may update it to include more.
This model includes:
Mita ("aka Crazy Mita without the evil)
Mila (Nerdy Mita)
Sleepy Mita
Kind Mita (Good Mita)
Tiny Mita (deformed Mita)
Cool Mita (Cappie)
Short Hair Mita
Yes, I know Tiny Mita has two arms in most generations, but honestly, I'm fine with that. I'll see if I can fix it in a newer version. I don't really have any other outfits other than her default ones, but I may add it later.
Description
FAQ
Comments (11)
So nice to see one with all the girls, coolio
Thx a lot. TinyMita concept working pretty fine <3 Good work!
While it still works with other checkpoints i guess WAI NSFW Illustrious is a must for the exact style.
Thats also not a bad thing^^
Love your Lora :)
Especially including small Mita (former cute Mita), time for some wholesome moments :D (maybe you can try ti recreate her before she got damaged)
I've found that it struggles on the noobAI models. I'm relatively new to Illustrious but from what I understand there is a bit of a fundamental difference between regular Illustrious based models which may explain that. Otherwise, I've found that adding a few generic positive prompts help. What models have you tried? I'm curious to see if I can reproduce.
@xbamaris i did try Oblivious Mix illustrious, and Ntr illustrious V4, XI, XII, XIII, while ntr did come somewhat close..... Wai NSFW produces the best results, and thats fine by me :)
Also i think not to exeed 30-35 steps is crucial, all my higher trys became dull and just not as pleasant.
I can also agree with @Hyokkuda that it becomes more difficult the more promts you use.
I downloaded WAI NSFW just for your lora bc i like accuracy.
And your Lora is accurate!
So keep up the good work :)
@xbamaris Your right! Just by adding "3D" it gets a whole lot more stable :-)
@xbamaris The reason this is they seemed to have trained on some checkpoint called 'AnyIllustrious-XL' While I'm more than happy the community wants to see this tech thrive and grow, there is no need for a checkpoint made just for 'lora training' as the anyillustrious model mentions and causes confusion making people think it may be better to train on said model.
This can cause issues with anyone trying to gen on later models cause not all merged models may have the dataset from the anyillustrious model.. where as Illustrious0.1 model, any model made after will have data from it thereby making pretty much any lora made from illustrious0.1 to be compatible with any newer merges. You can think of it like a tree diagram with Illustrious0.1 being at the very top. So any models that trickle downwards from the top will have some form of illustrious0.1 still in it.
That said you can train loras on other models, though make sure they're a base finetune and not a merge or any loras merged into it for best long term compatibility.
any chance you can make this for SDXL? or maybe the IL version will work
im lost I dont know how Illustrious works compare to basic SD1.5 and SDXL
NVM i think i figured it out
@kazemora419 me enseñas como?
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.