little cocon (umamusume) Lora (LYCORIS - loha) ver.1

※Please do not post NSFW content in places where it can be seen by many people, such as on social media.

I promise! 😁👍

※NSFWはSNSなど、多くの人目につく場所に公開しないようお願いします。

約束だよ😁👍

I'd be happy if you could follow and support me to increase my motivation.

モチベーション向上のためにフォローや支援してくれると嬉しいぞ。

👇😎👇

https://twitter.com/oLAcle1220

https://patreon.com/olacle

Prompts

• Trigger word

  • little_cocon

• base

  • horse girl,horse ears,horse tail,green eyes,short hair,blonde hair,blue hair,multicolored hair,hair bow,blue nails,

• race clothes

  • long sleeves,black choker,hooded jacket,blue jacket,hoodie,hood,hood down,o-ring,frills,belt,asymmetrical legwear,boots,blue footwear,socks,striped,blue belt

  Recommended settings（おすすめ設定）

  ・race clothes（勝負服）

  weight 0.75~

  If you want to emphasize your upper body, it may be better not to use words related to socks and shoes.

  上半身を強調したい場合は靴下、靴系のワードは使用しないようにするといいかも

  ・other clothes（その他の服）

  weight ~0.75

  Please adjust the weight down if you want to make the model wear other clothes.

  その他の服を着させたい場合はweight を下げて調整してください。
  

To generate her star-shape pupil, please inpant with prompt 'rubypupil, eye close-up' and select 'Only masked'.

v2 has better compatibility with other models.

You might get perfect eyes sometimes, it's rare tho. Like gacha.

LoHa of Happy Meek (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.

• Something weird happened when training. If you interested in training, read comments on LoRA training info section.

Recommended options

• LoRA weight 0.6 ~ 0.7

• Trigger words

  • Base

    • happy meek \(umamusume\)

  • Original Art :

    • hair flower, hairclip, striped bowtie, blue bowtie, brooch, jewelry, collared shirt, dress shirt, white shirt, blue vest, white jacket, open clothes, open jacket, puffy short sleeves, frilled sleeves, white gloves, pleated skirt, blue skirt, boots, white footwear

  • Tracen summer uniform

    • black-framed eyewear, glasses, tracen school uniform, summer uniform, serafuku, puffy short sleeves, purple bowtie, horseshoe ornament, sailor collar, sailor shirt, purple shirt, white skirt, pleated skirt, frilled skirt, miniskirt, zettai ryouiki, white thighhighs, brown footwear, loafers

  • Tracen winter uniform

    • black-framed eyewear, glasses, tracen school uniform, ear ornament, hairclip, winter uniform, puffy long sleeves, sailor collar, purple shirt, white bowtie, horseshoe ornament, purple skirt, pleated skirt, purple thighhighs, brown footwear, loafers

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

• 129 images consist of

  • 4x low quality 47 images

  • 8x medium quality 46 images

  • 16x high quality 36 images

  • => 1 epoch = 1132 images

• replaced character feature tags with happy meek \(umamusume\)

  • 1girl, red eyes, pink hair, simple background, solo, tail, braid, white hair, official alternate costume, long hair, alternate hairstyle, brown eyes, horse ears, purple eyes, horse girl, brown hair, blonde hair, alternate hair length, bangs, horse tail, blunt bangs, animal ears, pink eyes, yellow eyes, short hair, grey hair, light brown hair, ponytail, medium hair, seiun sky \(umamusume\), bob cut

  • I think these tags are the essentials of Happy Meek 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.

• Yukino Bijin, Neo Universe, Zenno Rob Roy, Light Hello, Tracen Staff, Ines Fujin, Inari One and Tanino Gimlet were trained with learning rate of text encoder(TE) which has five times higher than Unet learning rate.

  • This is very interesting. Most of people uses TE lr as smaller than Unet lr. I did mistakes but got a good results. I don't know why these LoRA works. It may a clue to help to set proper learning rate.

  • I found this phenomenon in training of happy meek. During training, loss goes to nan. Happy meek has two tokens, so I think that higher TE learning rate with small tokens of trigger word make model boom.

All uploaded images are generated by AOM3A1

And, special thanks to my best friend. DH.

My first LyCORIS model.

Working range, 0.7-0.8.

Turkish actress model. From the popular series 'Knock on My Door'.

Very beautiful girl.

Let us embark on a journey to embrace some thot. Sometimes they are simply breath-takingly beautiful.

If you love my work and wish to buy me a coffe: ko-fi.com/tauron it helps me stays awake and get more content up

My model is out: Tauron Hybrid Mix - TauHybrid_Anime_Final | Stable Diffusion Checkpoint | Civitai

DISCLAIMER:
I'm doing this for fun and not for commercial purpose, I do not support selling imageries generated with all subjects inside this model. I do not support indecent generation images of all subjects used in this model. DO NOT post online explicit nudity, sexual or suggestive images using this model. All subjects are 21+.

===

Open to suggestion or request

NOTE : I will keep updating more actresses in versions update

===

Usage Guide

--

A. Lora Weight

• 0.1 ~ 0.4 -- is to combine multiple faces or features, or if you experience overbaking

• 0.5 ~ 0.7 -- is a balanced weight to get still accurate feature but with more flexiblity

• 0.8 ~ 1.0 -- is to create accurate representation of the features.

B. CFG Scale

• Lower CFG Scale means AI gets more creative and less overbaking problems.

• Higher CFG Scale means results get very accurate with training sets and more prone to overbaking problems.

• Dynamic CFG Threshold, using this method allows you to bump up quality and achieve photorealistic easily, but with lycoris i found out recently that this method actually lower success rate, so only use when you are composing with a Lora / model that works fine with this method.

C. Sampling Steps, High Res Fix & Upscaling

• I like to use sampling steps at 50, with high res step at 20, or 1/2 of sampling steps.

• I find that 2M Karras works better and faster than SDE Karras. with ERSGAN 4x+ method + ERSGAN 4x for Upscaling. This method actually gives quite good result without being too sharp or too plasticky

• Upscaling - GPFGAN @0.25, Codeformer @0.3 & Codeformer Weight @0.15 seems to yield superior results to me.

--

Combining with Attribute LoRa / LyCoris

You may need to do your own testing, but using character weight 0.2 below attribute lora weight yields superior results, if for some reason one features persists not showing or overbake, you may want to lower the weight of the overbaked feature Lora, and then use emphasis () to boost the specific attribute you want.

DDetailer Script

If subject face is small / far away use: DDetailer Script with bbox anime face with default settings except denoiser at 0.2-0.3. You can combine this LoRa with a lot of other Style / Attire / Pose / Acts LoRa, Also works well with character LyCoris.

How to install this script? I kinda forget how because its a bit complicated beacuse you need to install stuff from venv. but a quick google search will land you somewhere in tutorial page.

Please leave a review if you have any critiques or if you find this LoRa good.

Sticky slime trained on 260 sticky pictures at NAI model.

Trained trigger words:

pink slime | white slime | green slime | yellow slime | transparent slime

The model training is not very good because the tags I provided lack differentiation. I have attached the training set and hope that someone can continue optimizing this concept.

V4:

Finally, we've reached this version. It incorporates various action elements and reduces the issue of limb distortion in multi-person scenarios. You can combine this Lora with various poses to achieve the desired posture (based on testing, the results with 3D Lora are better). However, the trade-off is a slight decrease in facial similarity at longer distances. I recommend using a face editor to improve it significantly.

Unfortunately, I cannot view or process images. As for the model recommendation, Basil_mix_fixed is suggested for this version. Although Perfect World hasn't been tested, it should theoretically work well.

终于到这个版本了，加入了很多动作要素，减少了多人情况下的肢体崩坏情况。可以结合各种pose lora做出你想要的姿态（测试来看三次元的lora效果会好一些）。

但是代价是在远距离脸的相似度略微下降了，建议使用face editor修复，可以大幅度改善。

图片不好发，你们懂的。

该版本推荐使用Basil_mix_fixed,至于Perfect world没有测试，理论上应该可以。

V3:

If the facial effects are not satisfactory, you can increase the weight to improve them.

I recommend using Basil_mix_fixed, followed by Perfect world.Currently, the following models can be used:

提升了面部效果，难度比想的大了很多，鬼知道我做了多少个版本才有现在的效果。感谢朋友的帮助。

该版本推荐使用Basil_mix_fixed,其次是Perfect world。

效果不够的话可以提升权重。目前可以用的：

• crying

• grin

• tongue out

• naughty face

• tongue out

• angry

• pain

• fellatio, open mouth, tongue out,

• (gloom:1.4)

NEW VER:

This version has been retrained and it is recommended to use Perfect World with a weight of 0.6-0.8. I tested some clothes and poses that were not in the data set, and they turned out decently. The next step will be to consider adding more expressions and actions.

该版本重新训练，推荐使用Perfect world，权重0.6-0.8.

测试了一下不在训练集内的衣服和动作，姑且还行。下一步考虑加入更多的表情和动作。

****

Why is it called Almighty, because I hope the Lora can be used with pose and clothing.

The current version: more than 300 pictures, a dozen Concept. recommended with realistic. Can be combined with pose lora, using Basil_mix_fixed can reduce limb collapse. Clothing lora need to increase some weight.

I recommend using Basil_mix_fixed, followed by Perfect world. other models are also available, but I still prefer Basil_mix_fixed, because the most restored. Weighting 0.6-0.9.

为什么叫万能，因为我希望该Lora可以搭配各种动作和衣服使用。

目前版本：300多张图，十几个Concept。推荐搭配realistic使用。可以结合动作lora，使用Basil_mix_fixed可以减少肢体崩坏。衣服lora需要加大一些权重。

推荐使用Basil_mix_fixed,其次是Perfect world。其他模型也可以，但是我还是更喜欢Basil_mix_fixed，因为最还原。权重0.6-0.9。

内置的衣服： Included clothing:

1.Advent_Children_costume,midriff, crop top,skirt, swept bangs, bare shoulders, sleeveless, black skirt, gloves, bangs, collarbone, belt,

2.chinese_dress_purple,thighhighs, double bun, hair bun, chinese clothes, dress, china dress, jewelry,

3.chinese_dress_dragon,double bun, dress, jewelry, hair bun, cleavage cutout, chinese clothes, earrings, gloves, bracelet, cleavage,

4.clothes_lift,nipples, breasts, shirt lift, large breasts,

5.sliver dress,

6.golden dress,

7.Honoka_costume,swimsuit, breasts, bikini, torn clothes, shorts, torn thighhighs,navel, front-tie top, belt, short shorts, front-tie bikini top, bikini top only, jacket, cleavage, open clothes, open jacket,

8.Maid,breasts,maid headdress, cleavage,alternate costume, apron, enmaided, ,jewelry, earrings, leotard, wrist cuffs

9.Standard_clothes,crop top,gloves, suspenders, elbow pads, midriff,  fingerless gloves, red footwear, black thighhighs, black skirt, elbow gloves, suspender skirt, boots,  tank top, navel,sports bra, abs, bare shoulders, zettai ryouiki, pleated skirt, toned, miniskirt

10.nsfw,bondage, bound, arms behind back, nude, nipple rings, rope, pubic hair

weight 0.8 is best.

This LyCORIS has been trained after carefully selecting 57 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.

英语(English): This model does not come with clothing information, but it has correctly memorized the body data through regularization, so the character proportions are perfect. You can dress her in any clothes. A weight of around 0.6 is recommended.

中文(Chinese): 本模型不带有服装信息，但是通过正则化正确地记忆了身体数据，因此人物比例完美。你可以给她穿上任何衣服。推荐权重0.6左右。

日语(Japanese): このモデルは服の情報を持っていませんが、正則化を通じて体のデータを正しく記憶しているため、キャラクターの比例は完璧です。彼女にはどんな服でも着せることができます。推奨される重みは約0.6です。

韩语(Korean): 이 모델은 의류 정보를 포함하고 있지 않지만, 정규화를 통해 몸 데이터를 올바르게 기억하고 있으므로 캐릭터의 비율은 완벽합니다. 그녀에게 어떤 옷이든 입힐 수 있습니다. 권장 가중치는 약 0.6입니다.

General instruction: General prompting.
Supports complex/simple tag pair!(new bonus processing)

All generated images use this model without any additional embeddings or loras - https://civitai.com/models/67130/acmar-g-mix-or-amateur-deep-merge-model

Eleventh Lazy Style Lora.

Based on 586 image dataset of specific style.

100% SFW.

Generally, this is not supposed to be used for anythin complex(prompt wise), because it was not tagged in a way to be.

GPTS Lazy Styles will have a lot of bleed and lower than average editability. Their purpose is to generally capture feel/style of specific dataset.

This time i've added a bonus processing to dataset, which tried to determine if image is simple, normal, or complex, based on heatmap of HED processing.(Bless ChatGPT4)

And yes, i skipped over GPTS9 and 10. Their datasets were of insufficient size, and are too weak to be reliably used for the styles they want to be, so, not today.

Salt. Just Salt.

LyCORIS trained on Drax the Destroyer 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^^

Kazokukeikaku is an old Japanese PC game.

It is a comedy in which a group of people with no relatives get together and live in a pseudo family.

I created this LoRA at the request of a friend, and I am publishing it here.

inital tags:

"kazoku keikaku",

takayashiki aoba:

"takayashiki aoba, long hair, black dress, turtleneck, belt"

takayashiki jun:

"takayashiki jun, short hair, glasses, shorts, choker, vest, dress shirt"

takayashiki haruka:

"takayashiki haruka, short hair, shorts, hairclip, miniskirt, denim"

takayashiki matsuri:

"takayashiki matsuri, long hair, pleated skirt, hairband, red skirt"

takayashiki hiroshi:

"takayashiki hiroshi, glasses, collared shirt, blue necktie, wristwatch, grey pants, old man"

takayashiki tsukasa:

"takayashiki tsukasa, 1boy, hair over eyes, faceless male"

takayashiki akari:

"takayashiki akari, long hair,jacket, midriff, crop top, denim, green thighhighs"

Your dedicated onlyfans altar, summon your goddess here.

If you love my work, you can buy me a coffee to help me stay awake and cope with life to create more lora and stuffs : ko-fi.com/tauron

My Mix is out : Tauron Hybrid Mix - TauHybrid_Anime_Final | Stable Diffusion Checkpoint | Civitai

DISCLAIMER:

I'm doing this for fun and not for commercial motive, My main purpose is to compose a fictional character using AI with the beauty of all persons above, please support them if you love their works. i do not support any indecent images generated of all persons in this model, I do not support selling generated images of persons in this model. please refrain to what they have been posting online.

All persons in this models are 21+ and not a minor, please DO NOT post explicit nudity, suggestive content, or pornographic content online using this model.

===

Open to suggestion or request

NOTE : I will keep updating more actresses in versions update

===

Usage Guide

--

A. Lora Weight

• 0.1 ~ 0.4 -- is to combine multiple faces or features, or if you experience overbaking

• 0.5 ~ 0.7 -- is a balanced weight to get still accurate feature but with more flexiblity

• 0.8 ~ 1.0 -- is to create accurate representation of the features.

B. CFG Scale

• Lower CFG Scale means AI gets more creative and less overbaking problems.

• Higher CFG Scale means results get very accurate with training sets and more prone to overbaking problems.

• Dynamic CFG Threshold, using this method allows you to bump up quality and achieve photorealistic easily, but with lycoris i found out recently that this method actually lower success rate, so only use when you are composing with a Lora / model that works fine with this method.

C. Sampling Steps, High Res Fix & Upscaling

• I like to use sampling steps at 50, with high res step at 20, or 1/2 of sampling steps.

• I find that 2M Karras works better and faster than SDE Karras. with ERSGAN 4x+ method + ERSGAN 4x for Upscaling. This method actually gives quite good result without being too sharp or too plasticky

• Upscaling - GPFGAN @0.25, Codeformer @0.3 & Codeformer Weight @0.15 seems to yield superior results to me.

--

Combining with Attribute LoRa / LyCoris

You may need to do your own testing, but using character weight 0.2 below attribute lora weight yields superior results, if for some reason one features persists not showing or overbake, you may want to lower the weight of the overbaked feature Lora, and then use emphasis () to boost the specific attribute you want.

DDetailer Script

If subject face is small / far away use: DDetailer Script with bbox anime face with default settings except denoiser at 0.2-0.3. You can combine this LoRa with a lot of other Style / Attire / Pose / Acts LoRa, Also works well with character LyCoris.

How to install this script? I kinda forget how because its a bit complicated beacuse you need to install stuff from venv. but a quick google search will land you somewhere in tutorial page.

Please leave a review if you have any critiques or if you find this LoRa good.

[English]

Recommended weight: 0.8 - 1.0

Trigger words

• Character: kujo karen

• School uniform(full body):

  full body, union jack jacket, striped bow, white shirt, long sleeves, pleated skirt, blue skirt, kneehighs, sneakers

Negative Prompt

• If the character's hairstyle is not accurate, please enter the following words in the Negative Prompt:

  double bun
  

===========================================================

[日本語]

推奨ウェイト: 0.8 - 1.0

トリガー

• キャラクター: kujo karen

• 制服(全身):full body, union jack jacket, striped bow, white shirt, long sleeves, pleated skirt, blue skirt, kneehighs, sneakers

Negative Prompt

• 髪型の再限度が低い場合は、次のワードをNegative Promptに入力してください:

  double bun

英语(English): Due to the lack of high-definition training sets, this model sometimes does not perform perfectly, and because it is a LOHA file, I currently do not know how to handle it in layers. In short, this is a test version, suitable for generating real images rather than 2D. The weight takes effect at 0.4 and it is best not to exceed 0.6.

中文(Chinese): 因为缺少高清训练集的缘故，这个模型有时候表现不是很完美，而且因为是LOHA文件，所以我目前不知道如何分层处理。总之，这是一个测试版，适合生成真实图像而不是2D。权重在0.4的时候起效，最好不要超过0.6。

日语(Japanese): 高解像度のトレーニングセットが不足しているため、このモデルは時々完璧に動作しないことがあります。また、LOHAファイルであるため、現在のところレイヤーでの処理方法はわかりません。要するに、これはテスト版で、実際の画像を生成するのに適しており、2Dではありません。重みは0.4で効果を発揮し、0.6を超えない方が良いです。

韩语(Korean): 고해상도 훈련 세트가 부족하기 때문에 이 모델은 때때로 완벽하게 작동하지 않습니다. 또한 LOHA 파일이므로 현재 계층별 처리 방법을 모릅니다. 요컨대, 이것은 테스트 버전으로, 실제 이미지를 생성하는 데 적합하며 2D가 아닙니다. 가중치는 0.4에서 효과를 발휘하며, 0.6을 초과하지 않는 것이 좋습니다.

英语(English): This model does not come with clothing information, but it has correctly memorized the body data through regularization, so the character proportions are perfect. You can dress her in any clothes. A weight of around 0.6 is recommended.

中文(Chinese): 本模型不带有服装信息，但是通过正则化正确地记忆了身体数据，因此人物比例完美。你可以给她穿上任何衣服。推荐权重0.6左右。

日语(Japanese): このモデルは服の情報を持っていませんが、正則化を通じて体のデータを正しく記憶しているため、キャラクターの比例は完璧です。彼女にはどんな服でも着せることができます。推奨される重みは約0.6です。

韩语(Korean): 이 모델은 의류 정보를 포함하고 있지 않지만, 정규화를 통해 몸 데이터를 올바르게 기억하고 있으므로 캐릭터의 비율은 완벽합니다. 그녀에게 어떤 옷이든 입힐 수 있습니다. 권장 가중치는 약 0.6입니다.

Apocalypse Ravi from Epic Seven, trained as a LyCORIS network.

Requires extension for use:

https://github.com/KohakuBlueleaf/a1111-sd-webui-lycoris

Recommended tags are:

dsaravi, black hair, yellow eyes, pointy ears, very long hair, twintails, ahoge, horns, forehead mark, hair ribbon, red tabard, chest belt, elbow gloves, thighhigh boots, black shorts

Check out my HuggingFace repository:

https://huggingface.co/Doctor-Shotgun/ds-LoRA

Check out my Pixiv:

https://www.pixiv.net/en/users/90447503

Would appreciate images, feeedback, requests, reviews, and of course comments in the comments. Please check out my other models if you like this one.

LoCon and Lycoris extensions required

Mei-Fang from Arcana Heart.

wieghts 0.6 - 1

recommended tags: bun_cover, hair_bun, double_bun, leotard, shrug_\(clothing\), highleg, thighhighs

Lone Crescent Bellona from Epic Seven, trained as a LyCORIS network.

Requires extension for use:

https://github.com/KohakuBlueleaf/a1111-sd-webui-lycoris

Recommended tags are:

dsmellona, orange hair, very long hair, side ponytail, hair bun, hair ornament, yellow eyes, black dress, detached sleeves, cleavage, bare shoulders, fishnet thighhighs, asymmetrical legwear

Check out my HuggingFace repository:

https://huggingface.co/Doctor-Shotgun/ds-LoRA

Check out my Pixiv:

https://www.pixiv.net/en/users/90447503

Achates Larunda from Epic Seven, trained as a LyCORIS network.

Requires extension for use:

https://github.com/KohakuBlueleaf/a1111-sd-webui-lycoris

Recommended tags are:

Base Outfit: dsachates, blonde hair, very long hair, twintails, blunt bangs, red eyes, hair ribbon, hat, skirt, gloves, white thighighs, garter straps, capelet

Swimwear: alswim, blonde hair, very long hair, twintails, blunt bangs, red eyes, hair ribbon, bikini, sunglasses, eyewear on head, necklace, midriff, small breasts

Check out my HuggingFace repository:

https://huggingface.co/Doctor-Shotgun/ds-LoRA

Check out my Pixiv:

https://www.pixiv.net/en/users/90447503

The LoCon/LyCORIS system is designed for the SD2.1 series, specifically targeting WD1.5beta3. Its aim is to generate outputs similar to anime.

Please install the LoCon extension. For larger outputs, optimal results may not be guaranteed.

For negative inputs, we suggest "(mayng:1.4)".

Please refer to the prompts provided in the sample images.

mayng(SDv2 Negative Embedding)

https://huggingface.co/852wa/hakoMay/tree/main

关于此版本

没有绑定tag关键词（其实是我不会）随便抄一串tag就能跑，

别问tag是啥，当你提问的那一刻，我的1060高端显卡已经开始跑图了

后续可能会更新一个疯牛版，{超过拟合}

请勿用本模型的产图，打扰此画风的原作者，仅供交流

底模自己另行尝试

DISCLAIMER: I am NOT affiliated with Majalis in any way. I just liked the game, scraped a bunch of images from the internet and created this model.

This is a style/characters from the game Tales of Androgyny by Majalis duo.

Please share your images here, as I would love to see what people can create with this.

This is my second attempt at creating a model for this game style, it turned out quite well and can it definitely generate Hiro in various context, because he had the most images, but as you can see from examples you can also get decent examples on other characters. I can already see the reason why characters like Trudy or Alma are harder to generate, but it will take a lot more work on the dataset preparation and proper tagging. So, there will probably be a V2, if find the time for it.

The model was trained mostly on NSFW images (it's an erotic VN, duh!), but it's capable of generating SFW images of characters as well.

It was trained on AnyLora, but it turned out to be pretty portable to other models, so far I've successfully tried FantasticChix-HR and Perfect World.

You should use CLIP Skip 2 (that's what AnyLora uses). You should start the prompt with "majalis style" followed by the character you want like "kylira". You can also add "toa" to your prompt if you want to add strength to the model, because this tag was present in every training image. For optimal results, use "1girl" + "futanari" for all the futa characters (Alma (warlock), Urka (orc), goblin, Merrybelle, Quetzal, demon king, dark knight, Dullahan, and much more...) and "1boy" + "otoko no ko" for the boys (Hiro, Kylira, Trudy)

You can find all of the tags used (and their frequencies) in the metadata of the model.

It's capable of generating penis instead of vagina for futas and traps in sexual positions even without external LoRAs.

trigger word : kirdunst

recomm. strength : >0.9 (0.96 isgood balance I have found for likeness),

works well at full strength as well...

Note : model is trained on only images from Spider-Man 1 , so the red hair is also baked in.

if you like my stuff pls consider supporting me on Kofi

LoCon trained on CyanCapsule's Style, no keyword needed and it works great at 1 weight.

All sample images were generated using Aurora.

Some sample images were done using V1, but V2 is just an improvement.

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 V1.20:

<lora:footjobPOVv120:1>(standing footjob, penis, pov) and IF you want the dude legs "male legs" on top of those 3 and "from below" for a lower angle

Be sure to use tags for the female character appearance!

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

Sample images generated from Nothing model.

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 37 512x512 images scraped from the internet and 20 epochs using animefull_final.

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

The LoCon/LyCORIS system is designed for SD2.1 series, targeting WD1.5beta3. It aims to produce outputs reminiscent of black and white anime line drawings.

Please install the extension feature of LoCon. Better results can be achieved by using Hires.fix with a 1.5 scale-up at 512x768 resolution.

For positive inputs, we recommend "monochrome, lineart, <lyco:sd2ly_line:0.65>".

For negative inputs, we suggest "(mayng:1.4), fullcolor, photo, shadow".

Please refer to the prompts from the sample images.

mayng(SDv2 Negative Embedding)

https://huggingface.co/852wa/hakoMay/tree/main

... and the god of Ai said: Let there be waifus! and there were Ai waifus.

There are anime loras out there what is the difference?

My version focused heavily on cosplay -- this means that it has a bulk of cosplayers data sets in instead of arts / illustrations - the ratio in general are 60-40. Secondly, with Lycoris we only need 11MB, more space for more waifus! Third, there are NSFW datasets mixed in, so have fun and enjoy!

===

If you love my work, and want to treat me a coffee : ko-fi.com/tauron you dont have to, but if you do, it helps me stay awake until midnight (which is the only time im able to work on these stuff lol) and cope with life..

My model is out, check out: Tauron Hybrid Mix - TauHybrid_Anime_Final | Stable Diffusion Checkpoint | Civitai

--

DISCLAIMER:
All of my models are to be used with 21+ subjects only, I'm doing this for fun only, DO NOT use for commercial purpose. Nudity generated with this model DOES NOT DEPICTS REAL PERSON.

===

NOTE : I will keep updating more actresses in versions update

Open to suggestion or request

Request Rules:

1. No requesting minors or anything that has underage as their datasets! please don't make civitai or ai in general into trouble.

2. If requesting a person, that person should have a public images (ex. no private instagrams), they themselves has to consented sharing their images online for public

3. Be a decent human being, if you requested, please at least give a review, since I'm doing this for free. and dont be a dh and review bomb if you don't get what you want, not all lyco is perfect.

4. Do not start discussion that is pointing towards anything illegal.

5. The requested subjects has to have adequate datasets available online

6. Put a link to the requested subject datasets.

===

Usage Guide

--

A. Lora Weight

• 0.1 ~ 0.4 -- is to combine multiple faces or features, or if you experience overbaking

• 0.5 ~ 0.7 -- is a balanced weight to get still accurate feature but with more flexiblity

• 0.8 ~ 1.0 -- is to create accurate representation of the features.

B. CFG Scale

• Lower CFG Scale means AI gets more creative and less overbaking problems.

• Higher CFG Scale means results get very accurate with training sets and more prone to overbaking problems.

• Dynamic CFG Threshold, using this method allows you to bump up quality and achieve photorealistic easily, but with lycoris i found out recently that this method actually lower success rate, so only use when you are composing with a Lora / model that works fine with this method.

C. Sampling Steps, High Res Fix & Upscaling

• I like to use sampling steps at 50, with high res step at 20, or 1/2 of sampling steps.

• I find that 2M Karras works better and faster than SDE Karras. with ERSGAN 4x+ method + ERSGAN 4x for Upscaling. This method actually gives quite good result without being too sharp or too plasticky

• Upscaling - GPFGAN @0.25, Codeformer @0.3 & Codeformer Weight @0.15 seems to yield superior results to me.

--

Combining with Attribute LoRa / LyCoris

You may need to do your own testing, but using character weight 0.2 below attribute lora weight yields superior results, if for some reason one features persists not showing or overbake, you may want to lower the weight of the overbaked feature Lora, and then use emphasis () to boost the specific attribute you want.

DDetailer Script

If subject face is small / far away use: DDetailer Script with bbox anime face with default settings except denoiser at 0.2-0.3. You can combine this LoRa with a lot of other Style / Attire / Pose / Acts LoRa, Also works well with character LyCoris.

How to install this script? I kinda forget how because its a bit complicated beacuse you need to install stuff from venv. but a quick google search will land you somewhere in tutorial page.

Please leave a review if you have any critiques or if you find this LoRa good.

A lora for creating ass-to-ass scenes

Trained on 116 images

Image info

• Model used abyssorange2hard

• Model used kotosmix

• Model used anything v4.5

• bad hands embedding

• bad artist embedding

• bad prompt embedding

• deep negative embedding

Notes

• The bigger the ass the more "connection" between the asses

• Use a landscape resolution for the best generation (768x512),(640x640)

Training settings?

The most important part of training is getting a good dataset!

Training settings can be found by installing the additional networks extension in auto1111

Extensions -> available -> load from -> additional networks

Restart webui

place lora into stable-diffusion-webui\extensions\sd-webui-additional-networks\models\lora

To view the training settings

Go to additional network tab -> choose model -> training paramters

手指依旧需要roll，完全盖住眼睛有些困难，我在例图中给出了一些示例用法，它既可以表现害羞尴尬（比如sexing，为了维护我少数的lora和谐环境，在这里不演示了），也可以表现哭泣、病娇之类的

Good fingers still need luck, and it is difficult to completely cover the eyes. I have given some example usages in the illustrations. It can express shyness and embarrassment (such as sexing.In order to maintain my rare purity, I will not demonstrate it here), and it can also express crying, sickness and the like.

如果你想盖住眼睛，加入covering own eyes如果还是不行就再加入closed eyes之类的。如果不想就不加额外词，也可以加入peeking through fingers来强化

If you want to cover the eyes, add covering own eyes if it still doesn't work then add closed eyes or something.If you don’t want to add extra words, you can also add peeking through fingers to strengthen

如果有问题请给我反馈

About this Model

This LyCORIS:Locon model, eyedirectionV1.6, is for orienting the direction of the eye without using the popular keywords "looking away" in the positive prompt. The model changes eye direction while minimizing large head turns. ("looking away" is not bad if we want significant head turns, but for subtle turns it is too much for some of our setups.

How to Apply

Trigger: eyedirection

Settings:

Weight: 0.3 - 0.5

Methods: DPM++ 2M/2Sa/SDE Karras

Steps: 25 - 30

CFG: 7-14

Positive prompt keyword options:

eyes up, eyes down, eyes up left, eyes up right, eyes down left, eyes down right, eyes left, eyes right, looking above, looking below, looking sideways, eyes left looking sideways, eyes right looking sideways, eyes up looking above, eyes down looking below, eyes up right looking above, eyes down right looking below, eyes up left looking above, eyes down left looking below

ex. <lyco:eyedirectionV1.6:0.3> eyedirection, eyes left looking sideways, (solo, portrait:1.4)

For random eye direction, at least the word "eyes" must be present in the prompt:

ex. <lyco:eyedirectionV1.6:0.3> eyedirection, brown eyes, (solo, portrait:1.4)

Negative prompt:

(eye contact, eye to eye, looking at viewer, looks at viewer:1.5)

Limitations

It will work in models with realistic or photorealism in them. I have not directly tried it with anime models, and since this is my very Very VERY first model, there are still room for improvements and upgrades, so expect for next versions. Also, use this model responsibly and within legalities.