V3
Totally unnecessary but saw some nice new art and wanted to expand the dataset. I am a dumbass. V2 is perfectly nice but well we now have a V3. The hakke no fuuin is as iffy as always if not more. Had to compensate for the expanded dataset by literally pasting the svg seal in a bunch of images to pad it.
Anyway...
Works fine with: <lora:uzumaki_kushinaV15:0.7>, uzumaki_kushina,
Available "outfits" are:
green_sleeveless_dress_white_shirt_turtleneck: her typical dress.
tattoo_HakkeNoFuuinShiki_spiral: Just pick something that shows her belly, It is iffy as SD1.5 hates how complex it is
high_ponytail: Her ponytail version works fine.
As always as it is mostly fanart and doujins i will add the full dataset with the tag summary.
V2:
Well fuck, I revisited the dataset of V1 and it was disaster. Then again this one was my second or third LORA ever. So Pretty much had to clean up the dataset from scratch. I had left a bunch of text and watermarks. I also re-tagged and isolated her green dress and the Hakke no Fuuin Shikki in her belly.
Works fine with: <lora:uzumaki_kushinaV2:0.7>, uzumaki_kushina,
Available "outfits" are:
Kushina_Green_Dress: her typical dress.
tattoo_HakkeNoFuuinShiki: Just pick something that shows her belly, I guess as it is tattoish it might appear in her arm or something if she is fully clothed but I haven't fully tested it(i saw it appear on her arm once.)
Ponytail: Her ponytail version works fine.
As always I will add the tag summary inside the dataset for people to experiment.
V1:
This one turned rather sensitive and somewhat unstable, at high values it seems to take a cartoonish approach to Naruhodou style. I had best results with a weight between .3 to .6
Normally it works fine with high-ish CFG(11) but if you are getting poor results it might be a good idea to lower it.
Got relatively ok-ish resuts with the following:
highres, 1girl Uzumaki_Kushina, huge_breasts, hair_ornament, blush, <lora:Uzumaki_Kushina:.4>
Note: Using hairpin together with hair_ornament might fix Kushina's usual hairpin or make the model go bonkers.
Note2: To get the seal use the following together with appropriate clothing(like bikini): (tattoo), Hakke_no_Fuuin_Shiki,
Not completely happy with this LORA, but i guess it is good enough. As always I leave the training data if someone wants to try their luck.
Description
FAQ
Comments (17)
Nice, very nice, other kushina models never include the fuuin shiki seal. Now time to make futa preggo kushina banging boruto.
Thanks for the hard work dude.
bad result bad result bad result
Can you be a bit more specific?
@knxo i try this lora on yesmix1.6 original the result is not good, can you please training this lora on yesmix1.6 original checkpoint?
@MILF__hunter Checked the model if it is this one https://civitai.com/models/9139?modelVersionId=25571 well it looks rather generic. Most people train anime LORAs in the NAI/Any families of models as most mixes use them as a base and thus are the most compatible with stuff downstream. Ideally i should be using either NAI itself or AnthithingV3 but i like Anything V4.5 style better and has better support for NSFW than V3. TLDR It is unlikely there's enough people to make the retrain worth it and since i didn't like the model any way I can't justify to myself the effort it would take. Either way, most of the time I add the datasets so you could also just grab it and fire it yourself in a collab. It is already tagged and everything.
@knxo okay thank you very much, i using lycoris file instead of lora file and the result is good now. thank you. btw may i request lora/lycoris chara to you?
@MILF__hunter I normally don't do requests, as i handpick my dataset and clean them myself and normally use at the very least a 100 images. The main issue is one of willpower as if i don't like the character a lot, it becomes a pain going over it. It is just a lot easier to do light photoshop on a nice pair of tits than on an old man's dong(<-The reason i don't do men LORAs). I would say to drop the name but honestly i don't want to give you false hope.
@knxo honestly, i want a lora for mikoto uchiha from naruto anime, but if you dont want to make it then it's okay, no problem. dont push yourself too hard, thank you
@MILF__hunter Ah that one. Not enough images. That one was in my maybe bucket list. But the only images there are have her as generic black haired lady. To me a good dataset of her would need to first inpaint her images to give her the sharingan so she can be at least recognizable. And well to do that you first need a sharingan lora first. As far as i know there only exist like half a dozen images from the anime and from doujin authors i can only remember a couple by Naruhou-dou. So yea that would have to wait for more images to exist preferably with her having the sharingan.
whats Vae you used ? my v2 result is underwhelming
Normally use clearvae. https://civitai.com/models/22354/clearvae I kind of doubt that is your issue though as clearvae is rather simplistic with low saturation.
@knxo I'll try to use the aything pruned, the one you linked were not the same as you used. different hash by the image data as I noticed
okay, fantastic Lora. I've tried using the exact checkpoint and the result is absolutely superb. Its so much better. If anyone wondering, just search "anything4.5 pruned" on huggingface. the size should be 4GB and using that, the result is 10times better than the one on civitai.
@yogaarmando7792 I am happy you like it. As for the model, I actually do see it in civitai, it is the last of the list of 4 attachments in https://civitai.com/models/4855?modelVersionId=5581, it says Pruned SafeTensor with the checksum SHA256=6E430EB51421CE5BF18F04E2DBE90B2CAD437311948BE4EF8C33658A73C86B2A
Despite the somewhat shady reputation of that mix, I like it because the style is very neutral and veers away form the more cute-ish ones that seem only made to produce loli or the ones that looks like mass produced genshin impact rip-offs. A bit more classic, almost 2.5D but not too much.
could you make a pony?very thanks!
Sorry, my videocard is to crappy to train SDXL so that would be a no.
Loved it! Best model so far, Need more kushina D!ck penetration and BJ models with blush + embarrassment
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.