最近生病，发点清淡的。

三次元版本已经有了所以我做了个对应的。动作比较固定，手指依然靠roll，训练分辨率为960*960，基于NAI。

由于我自己的lora目录比较混乱，命名相似难以分辨是谁做的，我以后的lora都会加个ID作为前缀（这只是方便我收纳，你不喜欢就自己改）

I've been sick recently, so I'll upload something lighter.

The real version already exists so I made an anime one. The movement is relatively fixed, the fingers still rely on luck, and the training resolution is 960*960, based on NAI.

Because my own lora directory is quite confusing, it is difficult to tell who made it because of the similar naming, so I will add an ID as a prefix to my future lora(This is just for my convenience, you can change the name if you don’t like it)

LoHa of Light Hello (umamusume). Please install extension on WebUI.

After installing extension, put LoHa file in your sd/models/lora folder and write LoRA notation in prompt to applying LoHa. <lora:your_loha_name:weight>

All of my works is free. If you'd like to support me, feel free to buy me a coffee. 👍

NOTICE

• After careful consideration, I decided not to produce LoRA of the characters below.

  • Otonashi Etsuko, Akasaka Misato, Hosoe Junko

• If anyone wants the characters above, I will consider making them.

• Except for the above, I will make the rest of the characters in Umamusume: Pretty Derby, unless there is a special notice.

Recommended options

• LoRA weight around 0.7

• Trigger words

  • Base

    • light hello \(umamusume\)

  • Casual :

    • casual, hair ornament, hairclip, necklace, jewelry, blue shirt, long sleeves, shirt tucked in, belt, white skirt, blue footwear, high heels

Use a model derived from or mixed with animefull model.

• Settings

  • Stable Diffusion WebUI by Automatic1111

  • DPM++ Series(SDE Karras, 2M Karras, etc.)

  • about 20 steps, CFG scale 3.5~6.5

  • I recommend kl-f8-anime2.vae

  • CLIP skip = 1

  • Use Hires.fix to get higher quality image

    • Upscaler

      • Latent, Latent (nearest-exact), Latent (bicubic antialiased) or other Latent series.

    • Denoising strength 0.50~0.65

LoRA training info

• trained on sd-scripts by kohya_ss.

  • LyCORIS by KohakuBlueleaf

  • Thank you all a lot!

• based on Animefull-pruned model

• 75 images consist of

  • 4x low quality 30 images

  • 8x medium quality 28 images

  • 16x high quality 17 images

  • => 1 epoch = 616 images

• replaced character feature tags with light hello \(umamusume\)

  • 1girl, brown hair, horse girl, short hair, red eyes, brown eyes, horse tail, solo, long hair, animal ears, alternate costume, purple eyes, bangs, horse ears, braid, simple background, medium hair, pink eyes, hair intakes, half updo, black hair

  • I think these tags are the essentials of Light Hello or

  • unnecessary for training because of duplication.

• resolution : 768x768 with aspect ratio bucketing.

• other training settings is included in training data or metadata of LoRA file.

All uploaded images are generated by AOM3A1

And, special thanks to my best friend. DH.

This is a LoCon model trained from 90+ images about lineart girls from 2 different sources. I use multires noise to generate brushwork&flatcolor of dataset.The tone is close to guava,and that's where the name comes from.

You can use the LoCon to design figures of flat color.

Trigger is guava

Weight 0.9-1.0 will perform better on scenery&realistic figures,also you can try new way to use,lol

A LyCORIS model of Deba Hekmat.

This is a style trained on (checks) Uhh... 1665 of a certain SHAFT anime (Here is a hint. anime aired in 2022). Before anyone asks, this is not a LyCORIS (LoCon) on Monogatari.

NSFW? Uh no. The model strong refuses NSFW content. I tried, and only bringing down the weight does the model swallows the fact you want to gen horny.

Try not to use it for commercial uses. No one is gonna stop you but I believe some people will be rather upset over it.

IMO this model is not good, I'm just posting because I could.

LoHa of Tracen Staff in Umamusume: Pretty Derby. Please install extension on WebUI.

After installing extension, put LoHa file in your sd/models/lora folder and write LoRA notation in prompt to applying LoHa. <lora:your_loha_name:weight>

All of my works is free. If you'd like to support me, feel free to buy me a coffee. 👍

NOTICE

• After careful consideration, I decided not to produce LoRA of the characters below.

  • Otonashi Etsuko, Akasaka Misato, Hosoe Junko

• If anyone wants the characters above, I will consider making them.

• Except for the above, I will make the rest of the characters in Umamusume: Pretty Derby, unless there is a special notice.

Recommended options

• LoRA weight around 0.7

• Trigger words

  • Base

    • akikawa yayoi

      • (optional) horse ear, horse tail

      • these tags make her Northen Taste

    • hayakawa tazuna

      • (optional) horse ear, horse tail

      • these tags make her Tokino Minoru

    • kiryuin aoi

    • anshinzawa sasami

    • kashimoto riko

  • Casual look of Akikawa Yayoi :

    • cat on head, sun hat, white headwear, hat flower, blue rose, vertical stripes, striped ascot, vertical-striped shirt, open clothes, open jacket, blue jacket, cropped jacket, puffy long sleeves, buttons, high-waist dress, white dress, socks, shoes, blue footwear

    • (optional) folded fan, holding fan, hand fan

  • Casual look of Hayakawa Tazuna :

    • green headwear, bowler hat, hair ribbon, yellow necktie, wristwatch, formal, long sleeves, collared shirt, green jacket, buttons, double-breasted, green skirt, skirt suit, pencil skirt, black pantyhose, shoes, green footwear

  • Casual look of Kiryuin Aoi :

    • dress shirt, white shirt, collared shirt, frilled shirt collar, long sleeves, black vest, buttons, blue pants, black footwear, shoes

  • Casual look of Anshinzawa Sasami :

    • sunglasses, white gloves, pearl necklace, long sleeves, wide sleeves, open clothes, open coat, white coat, red dress, pencil dress, short dress, needle, thigh strap, shoes, red footwear, high heels

    • (optional) holding needle,

  • Casual look of Kashimoto Riko :

    • formal, long sleeves, collared shirt, wing collar, white shirt, dress shirt, vertical-striped shirt, buttons, striped, pinstripe pattern, pinstripe suit, black suit, black jacket, open clothes, open jacket, frilled sleeves, black belt, belt buckle, pant suit, black pants, black footwear, high heels

Use a model derived from or mixed with animefull model.

• Settings

  • Stable Diffusion WebUI by Automatic1111

  • DPM++ Series(SDE Karras, 2M Karras, etc.)

  • about 20 steps, CFG scale 3.5~6.5

  • I recommend kl-f8-anime2.vae

  • CLIP skip = 1

  • Use Hires.fix to get higher quality image

    • Upscaler

      • Latent, Latent (nearest-exact), Latent (bicubic antialiased) or other Latent series.

    • Denoising strength 0.50~0.65

LoRA training info

• trained on sd-scripts by kohya_ss.

  • LyCORIS by KohakuBlueleaf

  • Thank you all a lot!

• based on Animefull-pruned model

• 282 images consist of

  • 4x low quality 119 images

  • 8x medium quality 78 images

  • 16x high quality 85 images

  • => 1 epoch = 2460 images

• replaced character feature tags with akikawa yayoi

  • 1girl, multicolored hair, brown hair, two-tone hair, animal on head, fangs, alternate costume, white hair, very long hair, fang, purple eyes, streaked hair, blunt bangs, orange hair, long hair, blue eyes, simple background, skin fang, blonde hair, solo, bangs, virtual youtuber

• replaced character feature tags with hayakawa tazuna

  • 1girl, multicolored hair, horse girl, black hair, purple hair, brown hair, green hair, two-tone hair, ponytail, red hair, split ponytail, green eyes, aqua eyes, very long hair, low ponytail, simple background, long hair, solo, short hair, bangs, low twintails, twintails

• replaced character feature tags with kiryuin aoi

  • 1girl, purple eyes, hair ornament, brown eyes, blunt bangs, blue hair, blue eyes, dark blue hair, black hair, hairclip, ponytail, grey hair, purple hair, asymmetrical bangs, solo, straight hair, short hair, medium hair, bangs

• replaced character feature tags with anshinzawa sasami

  • 1girl, ahoge, horse girl, pink hair, simple background, long hair, blue eyes, horse ears, blonde hair, solo, swept bangs, very long hair, bangs, hair between eyes, lipstick, food, mask, beads, tattoo, heart, candy, bead, weapon, poke ball, garter straps, between fingers

• replaced character feature tags with kashimoto riko

  • 1girl, virtual youtuber, hair between eyes, bangs, yellow eyes, purple hair, ahoge, green hair, short hair, green eyes, very long hair, brown eyes, alternate costume, red eyes, brown hair, hair behind ear, black eyes, solo, dark blue hair, blue hair, blue eyes, black hair, simple background, grey eyes, wavy hair, medium hair, purple eyes, long hair

  • I think these tags are the essentials of each of character or

  • unnecessary for training because of duplication.

• resolution : 768x768 with aspect ratio bucketing.

• other training settings is included in training data or metadata of LoRA file.

All uploaded images are generated by AOM3A1

And, special thanks to my best friend. DH.

Unlimited Blade Works | Weapon LoHA

This is a LoHA Lycoris and you need this Extension for this to work. I'm happy with the results.

Weight can go from 0.5 up to 1.0. I recommend 0.9. ControlNet will streamline full body shots much easier, or more complex poses. Works with almost every Style, every Character and Photorealistic Models.

This is only a BETA. I will add more Swords, Weapons and Guns to this. SD doesn't like to generate Weapons (just like Ranni with 4 Arms). Trying to improve it on every Update. I recommend using this LoCon for better Hands. If you know how to improve this, i'm happy to hear you out.

WARNING: This LoHA does have some problems and hits most of the time 3 out of 10 times. Hands are a big problem. Hand does not connect with the sword and sword does not appear in the right place sometimes. Unsheathing Sword is currently very unstable. Use pointy ears in Neg, if you are using Link Master Sword

Trigger Words:

Link Master Sword, Lightsaber, Dual Wielding Swords, One Handed Sword, Unsheathing Sword

I want to add these Concepts in the future as well:

Weapon Classes:

Axe, firearms, arc, Knife, hammer, rods, throwing stars, sickle, whips, guns, grenades, shields

Subcategories:

one-handed, Two handed, pistols, guns, machine guns, shotguns, grenade launcher, rocket launcher, sniper rifles, brass knuckles, throwing axes, spears, javelins, arrows, crossbows, Sling, sniper, bazooka, Foal

Variations:

warhammer, tomahawk, short sword, broadsword, rapiers, longsword, dagger, Kukri, katana, tanto, wakizashi, naginata, quarterstaff, war spear, double axe, hatchet, baton, Shield with integrated weapon, Dual wielding of other weapons

How to use LoHA's in auto1111:

• Download LoHA

• Copy the file to stable-diffusion-webui/models/lora

• Select your LoHA and use the triggerword

• Make sure to change the weight if needed (by default it's :1)

• Have Fun!

Spider Gwen

How to use it?

This is my first LyCORIS, it's far from perfect. Trained on SD 1.5 with 230 images.
Trained for weight 1.

I use it with revAnimated.

A MMD TDA model 3D style LyCORIS trained with 343 TDA models.

Trained on NAI model. Worked well on Any4.5, AOM2_NSFW and AOM3A1B.

This is a model of Turkish actress Ebru Şahin. It was made as a personal request for my friend Danes. :)

It should be pretty versatile, although it does need some poking and prodding for generating images of her from the side.

Please do not generate anything disrespectful to her with my model, thank you!

LyCORIS trained on Rocket from
Guardians of the Galaxy

Images given wern't cherry picked, so you can get MUCH better results

Can't wait to see your generations^^

Issues:

• Using "teal and orange" makes images oversaturated

• When using "prison uniform" you need to specify camera angle

Despite Derrian Distro having trained a LoRA of this character long ago when I gave my dataset of hers since I had some problems training her when I used his scripts, I still nonetheless went ahead and trained one myself as a LoCon since this time I was sure it would've trained her better with that architecture, although still didn't nail her kimono and socks pattern either since it's very complex for LoCon too sadly apart from her small dataset since she's yet another obscure character which shares my birthdate.

Main tags:

hatakaze, fox ears, animal ears, tail (I noticed that sometimes she doesn't have her fox ears and tail, so it's best to always have these tags in prompt for each generation)

Default costume:

default, low twintails, ginkgo leaf, leaf on head, hair ornament, glasses, round eyewear, black choker, japanese clothes, black kimono, long sleeves, sleeves past fingers, wide sleeves, yellow skirt, pleated skirt, hakama short skirt, black socks, obi, tabi, sandals

Train based on AnyLora model. Training materials are basically official arts and pixiv artworks of 萩オスさん.

Recommend weights from 0.6 to 0.8,and enable hires.fix.

Sometimes, to make the character appear more mature, you can consider adding words like "old" or "beard" to the prompts.

This LyCORIS has been trained after carefully selecting 58 images. All photos have been enhanced to make the most of them. I hope you like it! Any feedback would be highly appreciated!

From my experience it runs best with Edge Of Realism, pirsusEpicRealism, epicDiffusion ,RealisticVision, CyberRealistic, Deliberate checkpoints, among others, although I think it will be just fine in other SD 1.5 models as well.

CFG scale should be between 4-7.

LyCORIS weights should be between 1-1.4

Notice: This free work is intended for art and illustration purposes of emergent technology and not intended nor' trained for making illicit images, as the author I've taken care to not use any private data in the training process or designed it for such purposes. By downloading this model you are agreeing with the aforementioned intent and that you will not use my work to create / share images of a potentially immoral and / or illegal nature.

Please check out my Ko-Fi page if you are interested in supporting me.

LoRA commissions open! -> Fiverr

This is a LoHa LoRA! You need the extension! Why? They do a lot better with multi-concepts.

Trigger words for V5.9 - From 325 images to 904:
v4.14 needs "futanari" in prompt but v5.9 doesn't!

Try Clip Skip 1 and Clip Skip 2!

<lora:GirlyPenisv59:0.7>(penis, testicles)
You can now specify between "flaccid" OR "erection", "precum", "chastity, testicles" (don't prompt for "penis", "erection", "flaccid"), "bulge" (don't prompt for "penis", "testicles", "erection", "flaccid"), "penis out", "erection under clothes" (don't prompt for "erection","penis"), "foreskin", "tucked penis", "cock ring", "veins, veiny penis", "clothing aside" (can be reinforced with "leotard aside" or "panties aside" for example), "ejaculation, cum, projectile cum", "futanari masturbation", "femboy" (for more masculine face, probably doesn't work tho).

Its probably a good idea to complement the LoRA with more specific ones for things that its struggling with like "chastisty"... 70 out of 910 images had chastity in it but its still hard for it to draw that. I will make these specific ones if theres enough interest in it.
Most tags for camera angles and a few different sex positions should also work.

Follow the steps in the Model Version notes to reproduce my images.

Sample images generated from CAN or Nothing, check metadata!

All negative embeds should be easy to find. and Prompts should be included in the sample images! Click the little icon in the bottom right to see it.

It was trained on 904 512x512 images scraped from the internet and 9 epochs using animefull_final.

Danbooru tags generated by Waifu Diffusion 1.4 Tagger and fine tuned for specific tags.

TL;DR - Trigger words are pussy, clitoris, clitoral hood, and cleft of venus. An explanation of these tags are available on the danbooru website, as most anime models are trained on booru tags.

If you appreciate the effort, send me a couple bucks: https://ko-fi.com/aisavant

Full Description

AI is bad at things like hands because of the variety of contexts and shapes they are in.

This is true for female genitals as well. Not to mention, most of the anime models out right now have been trained on censored posts which I believe skews the ability to draw somewhat realistic vaginas in their current state.

This is the first version worth releasing, but I hope to push further improvements.

Current status:

Trained on about 400 hand picked images

Carefully categorized and tagged training images to teach the lora on different parts of vagina anatomy. Current learned words are: pussy, clitoris, clitoral hood, and cleft of venus. An explanation of these tags are available on the danbooru website. While things like "cleft of venus" are not their anatomical name, most anime models are trained on danbooru tags, so it's best to stick to using them when training and using lora.

Locan trained at 896,896 resolution, .00025 learning rate, 5000 steps.

Results seem to outperform popular anime models by themselves, weather that's standing, sitting, solo, or group activities.

Immediate to do list: re-run training with DyLora instead of Locon to test results

Long term trial and error to do list:

fine tuning

Introduce new training images to differentiate different sex acts

compare and contrast training with different experimental lora features

Wheeze meme template. I have trained it thrice and experimented, though it maybe requires more tests. For now, this can be V1, and maybe in the future, if I can get it a bit more reliable and compatible with other loras, I'll release a V2.

For recommended settings, I'd say:

• Keep it between 0.6 and 1, and experiment between that range. Going back to compatibility, the same weight range applies.

• Some loras work well with just their tokens, some with tokens and descriptions, and others simply fail. Basically, it is HIGHLY situational.

• Rectangular images can get pretty wild by default, and the same happens with this, despite the usual format of the meme. You should try to keep it to 512x512, but going slightly above that in WIDTH ONLY should be fine.

• I would not recommend high step counts. Maybe stay between 12-25 steps.

• I would also recommend using a model that favors 2d styles and nothing too realistic. Remarkably, it can still have a fairly decent effect on realistic models, but 2D-based checkpoints are more reliable imo.

WARNING: Upon rechecking, I must tell you this was trained as a LOHA. Loading issues may arise, but all I can say is check to make sure you have the Lycoris framework extension or whatever installed.