Here we have our sluttiest exam proctor and Boruto's victim.

Ignore the following rant:

I have always found Kishimoto to be very weird with his decisions. How shippuden spiraled out of control DBZ style, how they nerfed Hinata's tits in Boruto. How boruto dares to exist while having the worst parts of Naruto's personality while being an entitled whinny brat. How they made Anko fat. For what? Progressiveness brownie points? 

In my mind Naruto's story would have been much best served by Naruto maturing and having to assassinate Sasuke to finally become hokage.

But nope we have the poor orphan rising in the world while hated by everyone, a rather interesting story. Turning into the reincarnation of a shinobi deity, son of a former hokage and what amounts to a princess and last known member of one of the most powerful clans in history that had their own ninja village with hookers and blackjack...

Back to the training and changes. The old version works ok. If anything it was a victim of a time when I used the most basic prompts in the sample images so people were able to see what they did when used by a complete noob. Obviously that approach didn't work as most creators started using over complex fine tuned prompts or even use photoshop(WTF?) to fix sample images. So I caved a bit and instead begun using higher res images and changed my simple negative prompt for a kilometer long shlong made with negative TIs In the hope they would be as simple as the first ones but give much better result out of the box. Which seemed to work as I got a lot more traffic in the long run.

Anyway... The point being that if my older V1 lora's sample images seem to look like crap today, is mostly because they were made with super basic prompts. So don't fear about trying them if you like the V2 or V3. V2 and V3 versions were mostly about adding outfit triggers.

V2:

Anyway. as Always the objective of this retrain was getting Anko's outfit a bit more accessible.

Works fine with: <lora:Mitarashi_AnkoV2:0.7> mitarashi_anko,

For her complete default outfit you should use:

• <lora:Mitarashi_AnkoV2:0.7> mitarashi_anko, forehead_protector, Anko_fishnet_bodysuit_bare_legs_Brown_microskirt_Brown_open_trench_coat

Her available outfits can be accessed in parts as the images in the dataset were a bit of a mix and match:

• forehead_protector: I didn't explicitly trained konoha's logo into so who knows what it will show. I mostly trained it so you could remove it and even use as a negative if needed.

• Anko_fishnet_bodysuit_bare_legs_Brown_microskirt_Brown_open_trench_coat: her default complete outfit.

• Anko_fishnet_bodysuit_bare_legs: Her outfit sans her skirt or trenchcoat.

• Brown_microskirt: her skirt separate from the rest of her outfit.

• brown_open_trench_coat: her trench coat separate from the rest of her outfit.

As Always I have added the dataset together with the tag summary so you can check available tags and experiment.

V1:

I saw a disturbing lack of Anko. So this is to remedy that. A good part of it is from Naruhodou. So support him if you can he is a great artist.

The LORA came out rather stable crapping out at .95. You may want to set it as .55 or .6 for a more balanced style otherwise it looks too much like naruhodou's art. Seems to have some propensity to add more than one Anko, probably some contamination from the model. It that an error? is that a feature? Anyway the solo tag seems to fix that

I experimented with her dango(the candy she likes to eat) but YMMV with that, the LORA seems to wants to use them as jewellery some times.

Had good results with: <lora:AnkoMitarashi:0.55>, AnkoMitarashi, solo, big_breasts, AnkoCoat, AnkoFishnet, microskirt, covered_nipples

Covered nipples seems to work better than simple nipples otherwise they may poke out of her mesh.

I tried to englobe her attire as following(so you can try to mix and match):

1. AnkoCoat

2. AnkoFishnet

3. microskirt

Tried to train her cursed seal but could find almost no images plus it is on the back of her neck, I tagged it as Ten_no_Juin.

Anyway I have added the tag summary inside the dataset so you can experiment.