I am the Reindeer Czar, so I created my second LORA, of Red-Nosed Raider from Fortnite. This depicts her really well I think.

Strength for LORA is say, 0.4 to 0.65 to get the good result, if you want a background.

It was trained on 5850 iterations, so quite much. Add the tags blue gradient background and simple background in the negative tag to get rid of the background better.

If there is some tip to do this in training to ignore the background, I can retrain for Version 2.

Any feedback to improve the model is helpful, and please share your cool images of Red-Nosed Raider!

The tags I use the most - 1girl, artoria pendragon \(lancer alter\) \(fate\), large (or huge) breasts, wide hips, blond hair, yellow eyes.

For normal outfit - large breasts, underboob, ribbed_leotard, black clothes,

For Royal Icing (Or something that looks like royal Icing) - white negligee, lingerie, black_stockings.

First attempt at making a Lora. Trained on 666 images of - pay close attention! Artoria Pendragon Lancer Alter PLUS everything that had her tag. That means most of the images with Royal Icing outfit. It means that hypothetically you can make something close to Royal Icing with other characters at slightly lower weights.

Using this lora on high weights will eventually lead to features unique to Artoria (The headwear and maybe the face thing). And, honestly, I still don't know which one of the epochs to upload because it was trained on 3300 steps with 1 repeat and 2 batch sizes, which I was told is overtrained. However, I ran a few tests with epoch 10 (out of 10, the most trained one), and the results were mostly fine. Images are done with MeinaHentai.

Also, higher weight for the most part - I believe- makes skimpier outfits.

Again, that's first LORA, I'm not sure how to properly run tests to determine if it's fine or not, because so far epochs got me everything I needed.

Welcome to feedback, review & give me more stars and red hearts!

Chinese actress & ballerina.

Be kind to these lovely.

Legal & Risk

It is prohibited to use this model for commercial purposes and any scenarios of illegal acts and purposes.

lora

I am the Reindeer Czar, so I created my first LORA, of Dolph from Fortnite. This depicts him really well I think.

Strength for LORA is say, 0.4 to 0.65 to get the good result, if you want a background.

It was trained on 5600 iterations, so quite much. Add the tags blue gradient background and simple background in the negative tag to get rid of the background better.

If there is some tip to do this in training to ignore the background, I can retrain for Version 2.

Any feedback to improve the model is helpful, and please share your cool images of Dolph!

1st ascension and 2nd ascension seems good enough, but i still find it difficult to generated her 3rd ascend form

• Recommended weight : 0-7-1

• 1st ascension:1stascend, yellow eyes, armored dress, blue armor

• 2nd ascension:2ndascend, yellow eyes, blue dress, dress, gloves,

• 3rd ascension:3rdascend,navel, bare shoulders,black panties, revealing clothes , wings, forked eyebrows, black clothes, black wings, small breast, dragon wings,(the result is still not very good atm)

This model was trained on SD1.5. I suggest using CFG 5-7 and anything above 15 sampling steps.

Model used: realisticVisionV13_v13

A style LoRA sporting vivid colors and solid support to make all kinds of characters.

How to use
Use trigger tag "navigavistyle" and then just prompt away.
This LoRA came out surprisingly stable and can produce high quality results even on plain old NAI so all sample images are from it. If you're out of inspiration you can even prompt just "1girl" or "1boy" and generally get something decent out of it.
That said due to the large dataset no manual cleaning has been done so you may, rarely, see stray text or more characters than you prompted for.

Don't forget to review and upload your own images, it's always fun to see what you're creating!
Like my LoRAs and want more? Please consider a small tip towards my electricity bill and further training >.<

Misc
Trained on NAI. Dataset of 200 images.
All images are untouched raw txt2imgs, no inpainting done, for quality transparency.

This model was trained on SD1.5.

I suggest using CFG 3-8, strength 1.0, and anything above 15 sampling steps.

Model used: realisticVisionV20_v20

【Reminder】

Although this is a set of MahjongSoul Characters' LoRA, it's not guaranteed that you will have a result that can reappear ingame characters (Due to the limitations of training set). This set of LoRA is intended for the non-commercial use.

Although this is the LoRA of the Sparrow Soul character, but because there are really too few pictures available for training, it is not guaranteed that the clothing and image of the characters in the game can be fully reproduced. You should use this set of LoRA for non-commercial use only.

Recommend Lora weight is around 0.6,making it too high may cause blurred image or bad results. For pictures that don't have prompts resource, see "About this version".

It is recommended that the LoRA weight is around 0.6, too high may cause blurred or ghosted pictures. For the prompts that are not written in the renderings, please see the prompts in "About this version" on the right.

Next I'll slowly train & upload the LoRAs of the rest of characters. I need more images to enhance the performance of these LoRAs so if you made some awesome pictures please share it.

Next will slowly (really slowly!) upload the rest of LoRA. If you use these LoRA to generate good pictures, I hope you can share them, because these pictures can be used to improve the quality of LoRA (refinement).

Slowly uploading new demonstrate images.

【Currently Uploaded LoRAs】

Ayane --> Kisaragi Ayane

Yuzu -->Yuzu

Qingluan -->Qingluan Qingluan/ヤイラソ、

Qingku -->Qingkuniang Qingku Niang,

Hana -->Ninomiya Hana

Nanami -->Nanami Reina

Chihori --> Terasaki Chihori

Yui --> Yagi Yui

Mikami -->Chiori Mikami

Xenia -->Xenia Zenia/ジニヤ、

Ichihime --> Ichihime Ichihime,

Haruna -->Igarashi Haruna

Wanjirou --> Wanjirou

Mai --> Aihara Mai

Lily --> Kitahara Lily

Fuzi -->Fuzi Nadeshiko,

Hibiki -->Himekawa Hibiki

Nanaha -->Onodera Nanaha

Kazuha --> Saionji Kazuha Nishienji one bird,

Nana --> Shiraishi Nana

Joseph -->Joseph Joseph/ジョセフ、

Fuyumi -->Shinomiya Fuyumi

Fujita Kana

Hina Momo Young Peach / Young Peach,

Kujou Riu Nine glass rain,

Ayako -->Morikawa Ayako

Kitami -->Kitami Sawako

Anju -->Suzumiya Anju Suzumiya Anju/Suzumiya Anju,

Hinata -->Takanashi Hinata

Qixi Tanabata/チーシー、

Fuji Fukuhime/Fuji,

Kaavi

KaguyaHime Princess Kaguya

Mirai future

Miki -->Nikaidou Miki

Sara Sarah/サラ

Eliisa

Nian (NPC of 2019 Sparrow Soul Spring Festival Event / NPC of 2019 Chinese New Year Event)

【Characters LoRAs that still underway】

(Female Character) Fujimoto Kirara and rest of male characters.

Remember to use the trigger word choiyewon. Use LORA weight 1 and Hires fix with latent bicubic antialiased, 20 steps, 0.4-0.6 denoising strength, and 1.25x upscale. I've used the DDetails extension for pretty much every example picture.

Variable styles of trees, amazing effects apply to the game concept

discord channel：https://discord.gg/8WtKG3Rc9q Q群 692574838

This lora will make your character more white and bright. So I name it Pale Silver.

Will make your character white and bright, so call it Pale Silver.

Resource used:https://civitai.com/models/22530

The tall woman from Made in Abyss.

Weight: 0.8

Trigger + Default Outfit: ozen, black jacket, cape, black pants, black gloves

This does not have her limbs full of thousand-men pins.

A girl who has white twin braids and wears a mint green dress

Helpful positives: twin braids, mint leaves, white hair, leaf dress, leaf, plant

Helpful negatives: cat girl, cat ears

Occasionally it gets confused and replaces the leaves on her head with cat ears.

This is my Breast Squeeze LoRA, with the variation that the breasts are squeezed by the elbows. This LoRA is trained on NovelAI, so it will not do particularly well with realistic models.

The strength is good from 0.6 to 0.8, but I use 0.75 strength. All images are generated are generated with Hires Fix and the Orangemix VAE. All pictures were made with the blends available on my profile.

The keywords are 'elbow squeeze', and 'breasts squeezed together'. This LoRA has two types of generations. If you add in 'arms up', then the arms will be folded upwards. If you add 'arms down', then the arms will be folded downwards.

On a sidenote, this is one of my more proud LoRA creations since it just... works without much hassle unlike other concepts. I might also be working on the more general Breast Squeeze concept rather than the specific elbow placement here. It was a pain in the ass to get the dataset, since no booru or site has tags for the elbow squeeze.

训练素材使用动漫截图80张，生成后能高度还原动漫画风，但风格太过于固定，使用分层融合方法能有效去风格化。但冰狼爪很难训练，最终效果非常差，于是我在训练集中去除了该部分内容，现在就是尽可能还原特征并去风格化的模型，效果可以参考我的测试图。触发词ANIMAL FOOT,FOX GIRL,white hair,

Setsuna has two types of hairstyles, using key words respectively:"twin tails" 和 "long hair"如果不生效，请提高权重

The training material consists of 80 screenshots from anime, which can highly restore the anime style after generation. However, the style is too fixed, and using a layered fusion method can effectively de-style the images. However, it's difficult to train the ice wolf claw, and the final effect is very poor, so I removed that part from the training set. Now, the model aims to restore features as much as possible while removing the style, and the effect can be seen in my test images. The trigger words are "ANIMAL FOOT", "FOX GIRL", "white hair".

Setsuna has two hairstyles, one with twin tails and the other with long hair, respectively identified by the keywords "twin tails" and "long hair".If it doesn't work, please increase the weight.

Trained on NAI.
(Probably dosnt work well on realistic models)

Recommended weight: 0.6 - 0.8

feel free to share your result on the comments.

"You're all wasting your life!"

——

pugna's lora

A lora of Pugna which is a dota2 character

——

This lora is trained on animefull-latest, the effect should be the best

This lora is trained on"animefull-latest".That ckpt should fit perfectly.

-

Available in two styles cartoon and poster

There 2 styles:chibi and "offical wallpaper"

The trigger word for the cartoon is chibi, usually use 512*512 (white background can be used)

chibi is more cartoon style,I use 512*512 a lot.(and white background.)

The trigger word for the poster is official wallpaper, which can use 1024*512 (but the output also depends on the training of the base model)

"official wallpaper" is trained in 1024*512,but the output also depends on the ckpt.

-

Due to the relationship between the national server model, Skeleton can be added to the prompt so that Pugna only has a skeleton

Remember add "skeleton" into the prompt.（model issues）

Of course, you can also use it in reverse to remove the skeleton to make the sissy Pugna sexier lol

Of course,you can remove it to make 1"pugna" girl more sexy.lol

-

When using it, remember to add visor and base to the negative prompt words (there are masks in my dota2 model)

Remember add “visor” and "base" to the negetive.(I have a visor on my dota model which is ridiculous.）

——

Originally, I wanted to add guard and blood-sucking special effects, and at the same time take into account the three styles of "game", "poster" and "cartoon".

But I tried all the dynamic parameters and couldn't find a direction that can be realized

I even dimmed to 512 (512 will report an error lol)

But the picture looks more and more strange, and different concept levels will even pollute each other

In the end, I decided to cut some materials and have the current version.

I feel that the face is still more difficult to deal with in the case of a large picture.

And my previous needs may not be able to be realized with lora, maybe it will be solved in the future

Resource used:https://civitai.com/models/22530

There is a LoRA of her already.

Weight: 0.8

Trigger + Default Outfit: lupusregina beta, hat, frills, dress

Her long red hair with twin braids and yellow eyes are pruned, but feel free to add them anyway.

This only a test lora for trainning test.

Trigger world: kimminji

DO NOT USE THIS LORA FOR BUSSINESS USAGE OR GENRATE ANY NFSW CONTENT!!!

Tested on checkpoints:

Chilloutmix

Level4

VisionGen - Realism Reborn

If works not well, change lora weight between 1-0.6.

If you want to try this lora on more checkpoints, test yourself.

【Cthulhu theme】

Do you like a lot of tentacles, the feeling of the world collapsing, I like it, I hope you like it too

Use keywords:

cthulhu,monster,giant,feeler,multiple feeler,blood,scenery,outdoor,no humans,monster girl,light

cthulhu is the main keyword, other calls depend on the situation

Note: There are surprises in the colorful pictures

【Cthulhu theme】

Do you like the feeling of a broken world with many tentacles? I do, and I hope you also like it

Using keywords:

cthulhu,light,feeler,multiple feeler,blood,scenery,outdoor,no humans,monster,monster girl

“cthulhu” as the main keyword, other calls depending on the situation.

There's a surprise when drawing sfw pictures.

Trained on kohya-ss with 105 images on Stable Diffusion 1.5

Trigger Words: atalanta

Other Words: armor, underboob

V1 Recommended Weight: 1 - 1.2

V1_AMP Recommended Weight 0.65 - 0.8 (performs differently from other LoRA's)

Vaporunny - the most fluffable LoRA

This LoRA generates the vaporeon/lopunny fusion know as Vaporunny. Both fusions, vaporeon_ears and lopunny_ears, are packed into this LoRA. Works at 0.7 to 1 scale, but will vary per model.

Triggers and Example Prompt:

• vaporoonny

• lopunny_ears or vaporeon_ears

• white_pupils (black sclera recommended), blue_eyes (white sclera recommended), purple_eyes, and red_eyes

• white_collar, white_fluffy_collar, or leave out for no neck features

• fins

• tail (as always with tails, this produces mixed results)

Example prompt:

vaporunny, vaporeon_ears, white_pupils, white_fluffy_collar, fins, 1girl, blue_skin, :3, upper body, light_blush, sleeveless_shirt,

Training details

The LoRA was trained on AnyLoRA for 20 epochs using 54 images with 10 repeats. Clip skip = 2

No regularization images were used because I am not sure what dataset would work best for the Vaporunny character. Idea suggestions for a regularization dataset are welcome. A future version with regularization may preserve the base model's style better.

YOGISYA 's art style is very unique and full of mystery.

It's a experimental model

Maybe i update it in the future.

Tags:

• extra arms, closed eyes, hat, apron, gloves, buttons, and blood on face(you can remove it).

• Nsfw: should be work but needs some tags like nude, navel, nipples, pussy, collarbone, bare arms.

• Also you can add greyscale or monochrome tags.

Use the version that you like it.

All example images are made with text2img+SDultimateupscale

Buy me a coffee.

Made for a request by another user.

v1.1

Use strength 0.7-1.0, work in progress, has a problem with hair braid not vary accurate.

v1.2

Use strength 0.6-1.0, work in progress, use braided ponytail, for better hair results

V1.3 use strength 0.55-0.8 .... those 9 users got it before u, deal with it

V1.4 made. same as v1.3 but with leotard images testing....seems like the leotard is over powered. posting it tomorrow got some gamming to do

Trained with some NSFW images. first attempt was with a lot of images this is the 3rd version with only a few images. trying to figure out how to balance them, like my other lora no Triger words just try to describe her

This model is suitable for the conceptual design of the scene.

The base model can be Counterfeit-V2.5 or other related style models

Can be used in combination with other LORA

If you like my model, please share it with others or follow me.

discord channel：https://discord.gg/8WtKG3Rc9q Q群 692574838

This is my most satisfied concept scene model so far, I will not continue to update other scene models, will be more focused on improving the level of detail and structure, if you have any suggestions can be left in the comments section below, I will check

The model is not for commercial use, only for mutual learning

Continuing to LORA my interests when they're missing!

Trained with Megaman Roll images where she wears a variation of red dresses, not accurate at all but still pops good results. Use the metadata of the images as an example, it will get a lot of variations and poses.

Resource used:https://civitai.com/models/22530

Weight: 0.8

Trigger: bachira meguru

His short, multicolored hair and yellow eyes are pruned, but feel free to add them anyway.

Blue Lock Uniform: The uniform will very likely not appear when generating a wider shot, but throw him some sportswear, I'm sure he'll wear it.

Note: My model is trained with locon, so LoCon support needs to be installed.https://github.com/KohakuBlueleaf/a1111-sd-webui-locon

PS: If you have good ideas, you can leave a message. If you have downloaded Lora, I hope you can leave a message to tell me your experience and confusion, so that I can learn.（求5星好评，让更多人能够用到。）

PS: Leave a comment if you have any good ideas. If you have downloaded Lora, I hope you can leave a comment telling me how you feel about using it and how confused you are so that I can learn.

My Ko-fi website, there are some pictures I made, welcome to have a look.

My Ko-fi site, which has some pictures that I made, feel free to come and have a look.

https://ko-fi.com/asseer

【Note: 支持性捐助For financial independence only的人群，任何人都可以下载图片但没有那么高清。Supportive donations are only available to those who are financially independent. All can download the images but not in as much high resolution.】

【PS: Thank you to every friend who helped me buy Coke. Feeding 1 euro = 7 yuan, you can download all the 2K+ high-definition pictures I made within a month. PS: Thanks to everyone who helped me buy a coke by throwing in 1 euro = 7 dollars to download all the 2K+ HD images I made for a month.】

My pictures are all generated from the following tags/my pictures are all from prompts below.

Positive word: Prompts

{best quality}, long(v1.0)/LONG(v2.0), chinese ink painting, super huge dragon, solo, , , , very detail, mountain, clouds and mist, greyscale,

Negative words: Negative Prompts

NSFW, bad anatomy,bad hands, text, error, missing fingers,extra digit, fewer digits, cropped, worstquality, low quality, normal quality,jpegartifacts,signature, watermark, username,blurry,bad feet, blue eyeball, stamps,（自己调试/Debug by yourself）

Euler a sampling method=Euler a

Steps=30 Sampling times=30

CFG scale=7 correlation=7

Model= Anything4.5/Stable Diffusion v1.5

A Lora in the style of artist ddd.

It doesn't really work with clothes despite using reference images with clothes so just use for nudes.

Use a Lora weight of 1

The second edition of "Brilliant" has been updated, and the effect is better with the moxin master's mod.

As the title implies, the rest of the painting will produce a blurred effect when used, and the faces of the figures will be distorted and deformed.

If “Cthulhu” appeared then at your own peril.

Loading intensity between 0.2-0.8, the higher you go the messier the picture becomes.

Ink Lora by moxin needs to be used in conjunction with this, between 0.6 and 0.8.

You can use accurate face, lips, eyes , nose, to let the AI to correct automatically, or you have some sort of high weighted character gauging module of your own.

for example:

Loaded main mod: Anything 4.5.

Positive prompts:

(1boy/girl), solo, (style combination),

{best quality}, crazy detail, beautiful face, accutrate and clear face, lips, eyes, nose, Illustrations,

[thin muscular],

hanfu, shuimobysim,

((detail background)), landscape, mountain, forest, chinese architecture,

Reverse cue words:

pussy, penis, lowres,bad anatomy, bad hands, text, error, missing fingers,extra digit, fewer digits, cropped, worstquality, low quality, normal quality,jpegartifacts,signature, watermark, stamp, username,blurry,bad feet, matte, cross eyes, sketch, sots art, extra toes and fingers, open mouth, frame, Split Screen

Sampling method: Euler a 35steps

Facial repair check

Width 768, height 512.

Relevance 7

random seed -1

Opened Lora:

moxin ink heart: 0.6

GuoFeng3 Lora: 0.75

My Lora: 0.65

Lora that loads a character model such as Hipoly3D: 0.3

If it still can't run out, it may be a problem with the computer graphics card, and I can't do anything about it.

Does not require a dataset to train

Trained on:

#:3624|~striking eyes, large color contrast between hair and background, focal points converging to aesthetic and colorful image

using https://github.com/ntc-ai/conceptmod

Use higher lambda values and the keywords for more extreme pictures. Example gif sets Lora starting at 0 to 10 in 0.5 increments.

Lora extracted with:

https://github.com/kohya-ss/sd-scripts/blob/main/networks/extract_lora_from_models.py

Free for use anyone can use it for anything. It doesn't have a dataset so there is nothing to gatekeep with.

This model is suitable for the conceptual design of the scene.

This model just needs a simple tage or a description to produce a great scene

The base model can be Counterfeit-V2.5 or other related style models

If you like my model, please share it with others or follow me. I will update better models for environmental concept artists

The model is not for commercial use, only for mutual learningdiscord channel：https://discord.gg/8WtKG3Rc9q Q群 692574838

XSarchitectural_32ModernChinesestyle

Suggestions for use:

• If you want to get the parameters, download the image and use the stable diffusion feature.

• I used realistic large models and emaVAE to help my images get good quality, you can download them from the following URL, and thank the authors for developing these files:

  • https://civitai.com/models/4201/realistic-vision-v20

  • https://civitai.com/models/34192/ultra-sharp-high-contrast-tutorial-vaeandupscaler

If you think it's good, please remember to give me a five-star review

If you have any questions or suggestions, please contact us atxsarchitectrualstudio@gmail.com

Enigma from the anime series Musaigen no Phantom World. Now you can give her as much screen time as you want, side character be damned.

How to use
I kept all tags for flexibility so you have to use some prompts to recreate her such as "red eyes", "pointy ears", "long black hair", "pink lipstick", "makeup".
To change up her clothing more easily you probably have to lower the LoRA strength a bit otherwise she'll stick to black like the edgelord she is, 0.7-0.8 usually does it.
(Should be less of an issue as of v1.1)

Don't forget to review and upload your own images, it's always fun to see what you're creating!
Like my LoRAs and want more? Please consider a small tip towards my electricity bill and further training >.<

Misc
Dataset was a handful of shitty screencaps upscaled and cleaned. I took a gamble with 768 resolution training and it came out better than expected imho.
Trained on NAI. Sample images from AOM2_NSFW.
All images are untouched raw txt2imgs, no inpainting done, for quality transparency.

millie parfait lora train with 100 images, use weight strength of 1

( ͡° ͜ʖ ͡°)

it's 2 AM here so i'll keep it short for now:

• trigger word is overwatch 3d porn style

• suggested strength: 1

• blender, 3d are relevant extra prompts

• trained on about 620 overwatch porn renders from a handful of top artists, lazily grabbed using grabber with: "[artist name] overwatch -animated"

• some character LoRAs work best at around 0.8 strength, other at 0.5 (especially 2D anime girls)

• rule of thumb: the more clothes your character wears, the less 3D it will look

• defaults to an absolute fertility goddess body

• also tends to turn blonde characters into Mercy

• trained on nai, works pretty good on Anything V4.5, where my example images come from

• check the example images for decent prompts

As usual with my stuff, lots of room for improvement but i like what it does already and should really just go to bed

First attempt to make a LoRA and it was difficult, didn't like it as much as TI and found it harder to wield, this is actually version 2 but after so much messing about with it i'm moving on for now. I may return to update this one when I have made a few more LoRAs. It is Jessie Buckley, one of my faves, she's a awesome actress and super charming.

Seems to like 0.6-0.8

ALICEALICEALICEALICEALICEALICEALICEALICE......

Where is my Alice>>>>:?

LoRA of Agumon (Digimon), i made it for fun. It's my first attempt for an anime character.
I only tried to use it on AbbysOrangeMix3 for the moment.
List of prompt : agumon

The model is in the original Ghibli style and can be used with other Loras in the generative animation style which can create fantastic scenes and buildings。

Download counterfeit-v2.5 model before use

v2.0 The light and shadow effect of the structure is optimized to be more friendly to the architectural expression. Please adjust Tag to avoid unreasonable building structure

v3.0I improved the light, shadows and scenes, added a lot of detail, and tried to get away from the original style, which is more animation oriented. If you don't like this style, please continue to use v2.0, of course, if you want to try new style and details, you can try it

discord channel：https://discord.gg/8WtKG3Rc9q Q群 692574838

This model is not for commercial use, only for exchange learning

Resource used:https://civitai.com/models/22530

Weight: 0.8

Trigger + Appearance: tsurugi \(blue archive\), halo, choker, black wings

Her long black hair with blunt bangs and red eyes are pruned, but feel free to add them anyway.

Default Outfit: school uniform, black serafuku, sailor collar, black pleated skirt

Bikini: black bikini

I recommend giving her the crazy smile for her more signature look: https://civitai.com/models/5891/crazy-expressions

Introduction

Suzume is a LoRA that has been trained on a dataset of screenshots extracted from available trailers. Explore Suzume to boost your Suzume style AIGC productivity.

Installation

Download it to folder models/lora/

Usage

• trigger word: ssmrin

• preferred weight: 0.65

Credit

Credit to Makoto Shinkai, Lightwing_Lightwing

We will comply with the 774inc. secondary creation guidelines.

This is my first Lora and I like shaggy.

The model is very strong, 0.6 and lower is recommended

If you'd like to support me and do more:
https://ko-fi.com/edgonkofi - commissions open!
If you're looking for a >>LoRA Making Tutorial<<

Sirius - Azur Lane

Booba!

Quick tip: put X_Background in negative prompt and add some weight to it, just in case

"Guess Who?"

My 2nd LoRA, this time being the Universal mascot Woody Woodpecker! Contains both his look from the 2017 Live-Action/Animated movie and the 2018 animated web series!

To use Live Action model, use keywords such as "Realistic computer animation, cgi, 3d"

To use Animated model, use keywords like "Cartoon drawing, 2d, cartoon"

By Kisaku Ghost Work KK77

https://www.pixiv.net/users/86166786

Published
https://www.pixiv.net/artworks/107156893

R18+ Hentai Pixiv
https://www.pixiv.net/artworks/107157571

Training Sample
https://www.pixiv.net/artworks/107156376

Haruno Chika | Haruno Chika | Source of All Waifu [Source of All Waifu] Koikatu

地点：Night, Hong Kong View, High-rise_hotel,
动作：Standing at attention,
服饰：blazer, bow, bowtie, jacket,pleated_skirt,red_bowtie, school_uniform,
角色：,HarunoChika,
头发：brown_eyes, brown_hair, cowboy_shot,short_hair,
角色鎖定：
1 girl,Solo, Manga,20yo,mature female,Beautiful Finger,Beautiful long legs,Beautiful body,Beautiful Nose,Beautiful character design, perfect eyes, perfect face, Japanese,cosplay,
眼神表情：
looking at viewer, (innocent_big_eyes:1.0),alluring,light smile, seductive smile,embarrassed, shy,
畫質:
official art,extremely detailed CG unity 8k wallpaper, perfect lighting,Colorful, Bright_Front_face_Lighting,
(masterpiece:1.0),(best_quality:1.0), ultra high res,4K,ultra-detailed,
photography, 8K, HDR, highres, absurdres:1.2, Kodak portra 400, film grain, blurry background, bokeh:1.2, lens flare, (vibrant_color:1.2)
身材:
(Beautiful,large_Breasts:1.4), (beautiful_face:1.5),(narrow_waist),
Negative prompt: EasyNegative, lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, jpeg artifacts, signature, watermark, username, blurry, out of focus, censorship, Missing vagina, Blurry faces, Blank faces, bad face, Ugly, extra ear, amputee, missing hands, missing arms, missing legs, Extra fingers, 6 fingers, Extra feet, Missing nipples, ghost, futanari, Extra legs, Extra hands, panties,pants, (painting by bad-artist-anime:0.9), (painting by bad-artist:0.9), text, error, blurry, jpeg artifacts, cropped, normal quality, artist name, (worst quality, low quality:1.4),
Steps: 27, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 4144468641, Size: 512x768, Model hash: 3e9211917c, Model: yesmix_v16Original, Denoising strength: 0.5, Hires upscale: 2, Hires steps: 18, Hires upscaler: R-ESRGAN 4x+ Anime6B

Used embeddings: easynegative [119b]

Time taken: 52.51s

Torch active/reserved: 10029/10908 MiB, Sys VRAM: 12288/12288 MiB (100.0%)

A LoRA for creating images with women having their panties pulled to the side.

If you feel like supporting my work, please consider visiting my Patreon to get access to more detailed instructions on how to get the most out of my models: Link.

Update: I re-trained the model with more steps and changed the tagging slightly.

It does quite a bit better at reproducing the concept, especially from behind.

I changed the trigger word to 'pnts' so it no longer contains the word panties, which should help as well.

Recommended LoRA Strength: 0.6-0.8 from front, 0.7-0.9 from behind

Based on UberRealisticPornMerge

If you like this you might also like my other LoRAs.

A Lora in the style of KlimSpree.

Use a Lora weight of 1

This is a LORA for Girls in Boxes

It is very hard to get a specific poses or good hands and feet

I used 30 Pictures for training

I would recommend LORA weight 0.6-0.8 for

Trained with NAI

If you enjoy this LORA, I genuinely love seeing your creations with it

An outfit of Shadaloo Dolls from Street Fighter series.

Recommended weights: 0.5 - 0.7

Trained and tested on AnythingV4.5

Besides main tag you can also use: salute, hand on hip, empty eyes

If you like this Lora please leave an images in comments, I would love to see them!

Viktor Tsoi (Russian: Виктор Робертович Цой; 21 June 1962 – 15 August 1990) was a Soviet singer and songwriter who co-founded "Kino", one of the most popular and musically influential bands in the history of Russian music.

Model trained on Deliberate

Also fine working on other SD 1.5 based checkpoints
Recommended settings to generate:
LORA weight: 0.7 - 0.95
CFG: 5-7

Resource used:https://civitai.com/models/22530

Weight: 0.7 (other weights not tested)

Prompt his main appearance with dazai osamu. Should he look a bit off, he's got short brown hair and brown eyes.

He wears a brown jacket and white pants. The vest, collar, and necktie come included with the jacket. You may add bandages if you like.

A depressed guy that can nullify others' abilities. Care to jump off with join him?

This is my first LoRA and I messed up by not removing some images from the dataset. Decided to upload this straight after a few images though because I thought he looked good already, so just beware. May or may not redo this. I don't know, I'm just as lazy as he is.

Fun fact: Was planning to do the protagonist, so I named the project folder after Atsushi's name but the images on Gelbooru were limited.

All the content I make is and will be completely free on CivitAI
But if you want to support me with a coin, then you can do it here.

Sample images was generated with 🌕Luma

Recommended embeddings:
1. easynegative
2. bad-hands-5
3. badhandv4
Strength 0.6-0.8is optimal.

PLEASE MAKE SURE THAT YOUR LORA FILENAME MATCHES LORA NAME IN PROMT.

Classic outfit: hairband, bare shoulders, dress, puffy short sleeves, corset, purple skirt
Also you may improve it withprompt:athletic, glowing eyes, red lips
Negative prompt: loli, kid, child, old

Enjoy and share your feedback.

SSRHM,pink jaket,black skirt,black shirt,pink hair,pink eyes,side braid, hair bow

Recommended tags:

SSRHM,pink jaket,black skirt,black shirt,pink hair,pink eyes,side braid, hair bow

Behold the Intoners song.

My first publicly posted lora.

Girls bathing in a bowl of ramen

Helpful positives: food, in food, noodles, long hair, blonde hair, ramen, partially submerged, bathing, steam, bowl

Helpful negatives:eating

Adding "long hair, blonde hair" will generally turn the hair into noodles.

Not the strongest anatomically, it likes to add extra limbs especially extra legs.

Ok, this one was requested a lot.

Trigger word is sithoutfit.

For me worked best at wieght 0.7.

Some aliens were used on the training images, so i rccomend using "(face paint:1.6),(mask:1.6)" in the negative prompt if you want to supress it.

There will be a lot of light sabres here, sometimes when not wanted, if you want to get rid of a light sabre from a specific image put it in the nagetive prompt with high wieght.

You will see that there is no Anakins outfit in this LORA. I tired to use it in one of the first versions, but this is what i got:

For me those are jedi, not sith, that is why i got rid of this outfit in the data set.

Kisaku 鬼作 KK77
https://www.pixiv.net/users/86166786

Model: yesmix_v16Original

a man in a blue jacket and yellow scarf, shirt, ,black_pants,

Prompt:
地点：Night, Hong Kong View, High-rise_hotel,

动作：standing in a hallway with a wall in the background and a light on,

服饰：a man in a blue jacket and yellow scarf, shirt,black_pants,

Horn color: , Kisaku,

头发：belt, black_eyes, black_hair, collarbone, male_focus, solo,

Role lock:

1 man,Solo, Manga,50yo,male,Beautiful Finger,Beautiful long legs,Beautiful body,Beautiful Nose,Beautiful character design, perfect eyes, perfect face,

Eye expression:

looking at viewer,

Quality:

official art,extremely detailed CG unity 8k wallpaper, perfect lighting,Colorful, Bright_Front_face_Lighting,

(masterpiece:1.0),(best_quality:1.0), ultra high res,4K,ultra-detailed,

photography, 8K, HDR, highres, absurdres:1.2, Kodak portra 400, film grain, blurry background, bokeh:1.2, lens flare, (vibrant_color:1.2)

figure:

(beautiful_face:1.5),

This is my first ever LoRA. Hope you enjoy it!

LoRA model for the character Wattson from Apex Legends. Trained for 20 epochs on the Hassaku checkpoint model.

LoRA weight can be anywhere from 0.1-1. It should work fine at 1.

CFG scale 6-8, although you can try experimenting.

Trigger word (optional): "Wattson"

This LoRA is very NSFW friendly, but it can do SFW images just fine as well.

Sometimes multiple subjects/fractals may appear but it is rare for this to happen, and will get ironed out over the next couple updates.

How to use LoRA's in automatic1111's web-ui:

• Update webui (use git pullor redownload it)

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

• Select your LoRA like in this video

• Make sure to select your desired weight, or leave it at the default of 1. (example: is 1 weight, is 0.5 weight)

If you like this LoRA, I would appreciate a heart as thanks :)

LoRA to imitate the drawing style of 4th Ending of Chainsaw Man anime. Chainsaw Man - Ending 4 by TOOBOE

weight 0.6~0.8 is best

chika1, fujiwara chika, school uniform, black bow, hair_bow, black dress, red ribbon, long sleeves, blunt bangs, neck ribbon, collarbone, collared dress, large breasts

chika1, fujiwara chika, school uniform, black bow, hair_bow black dress, red ribbon, short sleeves, white shirt, blunt bangs, neck ribbon, collarbone, collared dress, large breasts


If you want to add her hat, add the headwear, hat bow prompt.

It's still a little crude. Has a few quirks that I'm patching.
Results vary wildly depending on the underlying checkpoint.

Success rate is about 20% currently, the other 80% are cronenburg abominations.

set the lora to 0.5 to start and increase until you are satisfied with the result. Some checkpoints need a higher value than others.
Works the best with kotosmix_v10 and HenmixRealV10

Try the different trigger words, it can be a little stubborn but sometimes they really change it up.

Some bugs that I still need to fix

1. Sometimes adds mosaic censoring

2. It sometimes makes too many nipples

3. can turn limbs into boobs
   
   
   

• We used one of Meina's model to generate the images.

• We don't provide exact prompts on the images, because we do post-processing on them so we lose the information, but we include tags how to summon her, which should be enough for you to use the model.

Tips to use:

• Use LoRA weight between 0.6-0.8

• If you want anime style make sure to use the tag "drawing"

• To summon her use the following tags: tyrande whisperwind, night elf, 1girl, pointy ears, long hair, green hair, colored skin, jewelry, medium breasts or large breasts, looking at viewer, earring

• You can also add purple skin, but it rarely works, usually she is white or really dark purple

• The prompts above only describe the character, make sure to put clothes on her too

If you have any questions feel free to ask!

Hope you guys like it! ❤️

You can ask for character LoRAs on our patreon page. Consider supporting us. 😘

Aile from Mega Man ZX! No armor included but with long hairprompt seems very close to the ZXA version.

Hope you like it!

Recommended weight 0.7

Recommended prompting

short hair, brown hair, short sleeves, bodysuit, robot ears, green eyes, short over long sleeves

Laura S. Arseid from Sen no Kiseki/Trails of Cold Steel. Trained on NovelAI.

Laura apparently has very little fanart, so my dataset was again very limited. Hopefully, it works well enough.

Tested at 0.8-1.0 strength. Contains her Sen I and Sen III outfits. Trigger phrase is "Laura Arseid" for both.

• "Long ponytail" helps generate her ponytail.

• Add "blue necktie, red jacket, white shirt, hair bow, green skirt" for her Sen I look.

• Add "white coat, ascot, blue shirt, spandex shorts, black thighhighs" for her Sen III look.

Experimental LORA based on the "You Like Kissing Boys Don't You" meme.
For now the model doesn't respond very well to details inside prompts, but I'll fix it in version 1.1! So if you have any cool generations, share them in the comments so I can use them to train new version :)

While the model can be little broken so you need to experiment a bit! In my experience, it's best to set the weight at ~0.7-0.8.

LoRA to enlarge the areola

Trigger:large areolae

A LoRA Model made to assist with the Generation of Shino Sosaki/Mandalay from My Hero Academia

I recomend setting the LoRA to 0.6, atleast thats the setting i used to get the best results.
Images sometimes generate the Weird Cat Ear-Thingy from her Hero costume, even if its not specified in anyway.
My Hero Academia belongs to Kôhei Horikoshi

Canadian singer and songwriter.

Rapunzel is a well-known Disney princess. To create the Lora, a meticulously selected and refined DataSet was used to achieve a versatile and flexible model. So that it adapts perfectly to styles and models. It also works in 3D. Use the keyword “(RapunzelWaifu:1)” for greater precision, and I also recommend using the keywords “(dress)” to invoke her characteristic outfit. If you want to change her outfit, I recommend lowering the weight of the Lora between “0.8 - 0.5” for greater flexibility, try it out and have fun. Remember to leave a Review with an image if you liked it and comment!! HAVE FUN :D

As a reminder, commissions and buying coffees are open here, or for if you're interested in taking a look at my gallery!
Also check out my twitter~

I piggy-backed on Lykon's Ahri Model for some of these generations to potentially salvage some of the outfits. I started to fuss with auto-tagging and it got a bit overzealous with the prompts. I wanted to fix this but for now it's pretty difficult to undo without some massive overhaul. In the meantime, just expect that sometimes the colors and outfits won't be quite right. For many of us, that won't be an issue (think when Mona does a scry on Venti's future)

A Lora of Erika from pokemon.

Use Lora weight of 0.7

My 4th try at it and i wont probably try this anymore, this was done on Hollowstrawberry's lora trainer: https://colab.research.google.com/github/hollowstrawberry/kohya-colab/blob/main/Lora_Trainer.ipynb

Trained with 61 images at 768 resolution for 5 repeats and 10 epochs, used flip_aug and min_snr_ganma, 2e-4 unet/1e-4 text econder lr, batch size 2.

Getting waai fu right is a nightmare and i think part of the problem is the models not being good at it, i only tested with 7thanimev3c: https://huggingface.co/syaimu/7th_Layer/blob/main/7th_anime_v3/7th_anime_v3_C.safetensors but if someone has a merge with a furry model and it looks better on those please do tell me.

More ammount of time went into pruning this time around, i decided to swap 1girl for furry female since leaving it before was detrimental, other than that the minimum tags to prompt her are "waai fu, furry female, tiger ears, body fur", if you want her glasses add "orange-tinted eyewear, round eyewear" and for her default outfir use "default china dress, capelet, pelvic curtain, elbow gloves, orange belts", add or remove "animal nose" to positive/negative prompts depending on what you want. Additionally you can use tiger girl, tiger tail to help prompt the tail but expect random tails around and tail limbs, i honestly dunno how to fix those.

Other than that here is the tag frequency for anyone interested: https://files.catbox.moe/0a76lr.txt

Merida is a well-known Disney princess. To create the Lora, a meticulously selected and refined DataSet was used to achieve a versatile and flexible model. So that it adapts perfectly to styles and models. It also works in 3D. Use the keyword “(MeridaWaifu:1)” for greater precision, and I also recommend using the keywords “(dress)” to invoke her characteristic outfit. If you want to change her outfit, I recommend lowering the weight of the Lora between “0.8 - 0.5” for greater flexibility, try it out and have fun. Remember to leave a Review with an image if you liked it and comment!! HAVE FUN :D

Snow White is a well-known Disney princess. To create the Lora, a meticulously selected and refined DataSet was used to achieve a versatile and flexible model. So that it adapts perfectly to styles and models. It also works in 3D. Use the keyword “(SnowPrincessw:1)” for greater precision, and I also recommend using the keywords “(dress, puffy sleeves, skirt)” to invoke her characteristic outfit. If you want to change her outfit, I recommend lowering the weight of the Lora between “0.8 - 0.5” for greater flexibility, try it out and have fun. Remember to leave a Review with an image if you liked it and comment!! HAVE FUN :D

https://www.artstation.com/alexchow

if you look at the example prompts provided with the pictures, you will understand how to control this thing right/the style.

This was trained using Realistic Vision 1.3

It is my first lora, so don't expect perfect results.

I'll keep improving to release something more accurate

weight 0.6 is best.

1girl, solo, blue eyes, orange skin, gloves, tentacle hair, colored skin, facial mark, facial tattoo, belt, forehead mark, lips, armor

If you don't want her to be holding a weapon, add holding weapon to the negative prompt.

A LORA model trained on Nathan Weir.

To use the model place the words "nathanweir person" in the prompt.

Trained on 5520 steps.

Follow me to make sure you see new styles, poses and Nobodys when I post them. Things move fast on this site, it's easy to miss.

Images were generated using the legendary A to Zovya RPG Artist's Toolsv2model.

Great tool to get the wet hair look that the standard Stable Diffusion doesn't really do too well. Works with photoreal and animation styles. Best results between strength 0.5 and 0.9. Be sure to prompt "wet hair".

Do you have requests? I've been putting in many more hours lately with this. That's my problem, not yours. But if you'd like to tip me, buy me a beer. Beer encourages me to ignore work and make AI models instead. Tip and make a request. I'll give it a shot if I can. Here at Ko-Fi

Female ace trainer from pokemon black and white

Use acebw as trigger word.

You may want to add "Orange vest", "long sleeves", "blue skirt", "pantyhose", "drill_hair" to your Prompt
use between 0.7 - 1.0 weight

Trained on Nai

All images were generated using yabalMix (https://civitai.com/models/28648/yabalmix) with xformers on and using kl-f8-anime2 vae.

images were generated at 512x768 and upscaled using highres fix by 1.95 and using 4x_foolhardy_Remacri upscaler

Female ace trainer from pokemon x and pokemon y

Use acexy as trigger word.

You may want to add "purple dress", "white bowtie" and "hairclip" to your Prompt

Trained on Nai

All images were generated using yabalMix (https://civitai.com/models/28648/yabalmix) with xformers on and using kl-f8-anime2 vae.

images were generated at 512x768 and upscaled using highres fix by 1.95 and using 4x_foolhardy_Remacri upscaler

v2.1 & v2.0

Other versions

A LoRA (Hadamard Product) based on 966 Shiny Colors cards, recognizing the 26 characters to varying degrees.

Resumed from deresute-v1.3-epoch16 by accident instead of training from scratch, retaining surprisingly much of the knowledge of Cinderella Girls characters. This suggests that it's likely the model's capacity is nowhere close to being reached.

This also serves as a demo of an extremely inefficient style transfer method: train the characters, and then train on lots of unrelated styled images.

Usage

This Hadamard Product LoRA requires installing an extra extension to be used in the Web UI.

The style itself doesn't have a trigger word. For the characters, below are the character tags and respective top 20 related tags.

Sakuragi Mano,
brown eyes, smile, 1girl, open mouth, brown hair, looking at viewer, solo, bangs, light brown hair, outdoors, skirt, holding, medium hair, blush, shirt, dutch angle, breasts, hair ornament, multiple girls, blonde hair

Kazano Hiori,
black hair, mole, mole under mouth, long hair, looking at viewer, smile, 1girl, solo, bangs, open mouth, ponytail, purple eyes, multiple girls, blonde hair, skirt, dress, hair ornament, breasts, outdoors, holding

Hachimiya Meguru,
blonde hair, blue eyes, smile, long hair, looking at viewer, hair ornament, 1girl, open mouth, solo, breasts, ahoge, hairclip, twintails, multiple girls, shirt, outdoors, bangs, black hair, skirt, large breasts

Osaki Tenka,
long hair, brown hair, yellow eyes, bangs, looking at viewer, 1girl, hair between eyes, smile, solo, open mouth, holding, blush, long sleeves, dress, multiple girls, frills, brown eyes, braid, sisters, skirt

Morino Rinze,
red eyes, black hair, bangs, short hair, 1girl, looking at viewer, solo, smile, blue hair, long sleeves, hair ornament, holding, hair bun, outdoors, multiple girls, skirt, shirt, open mouth, flower, long hair

Tsukioka Kogane,
long hair, brown hair, bangs, looking at viewer, breasts, brown eyes, large breasts, 1girl, smile, solo, open mouth, blush, skirt, jewelry, bow, shirt, ribbon, dress, holding, outdoors

Arisugawa Natsuha,
long hair, smile, bangs, orange hair, looking at viewer, breasts, jewelry, 1girl, solo, earrings, black eyes, open mouth, red hair, shirt, multiple girls, skirt, brown hair, jacket, ahoge, outdoors

Kuwayama Chiyuki,
brown hair, long hair, smile, looking at viewer, brown eyes, bangs, breasts, 1girl, ahoge, solo, braid, holding, open mouth, jewelry, blush, dress, skirt, multiple girls, large breasts, shirt

Osaki Amana,
brown hair, long hair, yellow eyes, smile, bangs, looking at viewer, open mouth, 1girl, solo, jewelry, breasts, multiple girls, blush, frills, dress, holding, collarbone, sitting, skirt, bare shoulders

Saijo Juri,
blonde hair, short hair, purple eyes, 1girl, looking at viewer, bangs, solo, smile, jacket, jewelry, outdoors, shirt, shorts, holding, multiple girls, open mouth, dutch angle, skirt, breasts, brown hair

Komiya Kaho,
red hair, long hair, red eyes, smile, open mouth, bangs, looking at viewer, 1girl, solo, ahoge, jacket, skirt, multiple girls, shirt, holding, shorts, brown hair, outdoors, hair between eyes, gloves

Tanaka Mamimi,
purple hair, purple eyes, bangs, diagonal bangs, looking at viewer, jewelry, 1girl, twintails, solo, earrings, long hair, smile, nail polish, holding, shirt, skirt, choker, open mouth, breasts, jacket

Yukoku Kiriko,
long hair, grey hair, purple eyes, bangs, looking at viewer, twintails, dress, smile, 1girl, solo, multiple girls, open mouth, long sleeves, holding, frills, brown hair, breasts, shirt, black hair, bandaid

Mitsumine Yuika,
black hair, long hair, smile, looking at viewer, 1girl, blue eyes, solo, twintails, glasses, open mouth, hat, jacket, brown hair, skirt, bangs, jewelry, holding, outdoors, multiple girls, shirt

Sonoda Chiyoko,
brown hair, looking at viewer, twintails, 1girl, solo, red eyes, smile, bangs, hair bun, double bun, holding, skirt, food, long hair, open mouth, multiple girls, jewelry, shirt, breasts, one eye closed

Serizawa Asahi,
blue eyes, short hair, 1girl, bangs, looking at viewer, grey hair, solo, open mouth, outdoors, smile, jacket, shirt, skirt, shorts, long sleeves, jewelry, shoes, sky, holding, brown hair

Shirase Sakuya,
black hair, long hair, yellow eyes, smile, bangs, breasts, jewelry, ponytail, hair between eyes, large breasts, looking at viewer, 1girl, solo, high ponytail, cleavage, earrings, multiple girls, jacket, open mouth, necklace

Izumi Mei,
long hair, blonde hair, brown hair, multicolored hair, gradient hair, gyaru, black eyes, breasts, looking at viewer, jewelry, 1girl, earrings, large breasts, solo, smile, tan, cleavage, jacket, dark skin, multiple girls

Mayuzumi Fuyuko,
black hair, long hair, bangs, looking at viewer, brown eyes, blunt bangs, smile, 1girl, skirt, solo, two side up, open mouth, shirt, jewelry, breasts, long sleeves, multiple girls, multicolored hair, holding, sitting

Asakura Toru,
short hair, looking at viewer, 1girl, solo, jewelry, multicolored hair, gradient hair, bangs, blue hair, earrings, ahoge, brown hair, outdoors, jacket, hair ornament, dutch angle, blue eyes, dress, multiple girls, smile

Higuchi Madoka,
short hair, mole under eye, mole, brown hair, purple eyes, hair ornament, hairclip, 1girl, solo, looking at viewer, bangs, skirt, jacket, long sleeves, outdoors, multiple girls, black hair, shirt, nail polish, dutch angle

Ichikawa Hinana,
brown hair, looking at viewer, brown eyes, long hair, bangs, 1girl, solo, skirt, open mouth, hair bun, smile, ribbon, dutch angle, parted bangs, hair ribbon, dress, jewelry, hair ornament, single hair bun, shirt

Fukumaru Koito,
black hair, purple eyes, twintails, 1girl, looking at viewer, solo, fangs, open mouth, skin fangs, skirt, hair ornament, ribbon, shirt, indoors, short hair, dutch angle, hair ribbon, long sleeves, multiple girls, brown hair

Nanakusa Nichika,
looking at viewer, 1girl, brown hair, green hair, solo, short hair, smile, bangs, brown eyes, multiple girls, long hair, jewelry, open mouth, jacket, shirt, outdoors, holding, earrings, gloves, dress

Aketa Mikoto,
multicolored hair, brown hair, gradient hair, jewelry, looking at viewer, jacket, 1girl, blonde hair, smile, earrings, brown eyes, solo, multiple girls, 2girls, gloves, nail polish, green hair, shirt, dress, breasts

Ikaruga Luca,
1girl, multicolored hair, black hair, yellow eyes, looking at viewer, solo, blonde hair, bangs, skirt, streaked hair, midriff, crop top, two-tone hair, medium hair, jacket, gloves, shirt, fishnets, breasts, off shoulder

As a bonus, Cinderella Girls prompts mostly work as well.

Training info

Training info Dataset:

• 996 Shiny Colors cards collected from Shanimas DB

Tags generated with SmilingWolf/wd-v1-4-convnextv2-tagger-v2 with a threshold of 0.35. No balancing and tag cleaning. No distinction between Produce and Support cards.

Inspired by a 2022 paper, a custom LR scheduler was used:

SequentialLR(optimizer, [
    LinearLR(optimizer, 0.1, total_iters=steps_per_epoch),
    CosineAnnealingWarmRestarts(optimizer, steps_per_epoch*2, T_mult=2)
], [steps_per_epoch])

Training cost: ~12 T4-hours (in addition to deresute costs)

Data distribution:

character cards
1 Sakuragi Mano 52
2 Kazano Hiori 49
3 Hachimiya Meguru 47
4 Osaki Tenka 46
5 Morino Rinze 45
6 Tsukioka Kogane 44
7 Arisugawa Natsuha 44
8 Kuwayama Chiyuki 44
9 Osaki Amana 43
10 Saijo Juri 43
11 Komiya Kaho 43
12 Tanaka Mamimi 42
13 Yukoku Kiriko 42
14 Mitsumine Yuika 40
15 Sonoda Chiyoko 40
16 Serizawa Asahi 39
17 Shirase Sakuya 38
18 Izumi Mei 35
19 Mayuzumi Fuyuko 34
20 Asakura Toru 28
21 Higuchi Madoka 28
22 Ichikawa Hinana 26
23 Fukumaru Koito 25
24 Nanakusa Nichika 23
25 Aketa Mikoto 22
26 Ikaruga Luca 4

Training config (v2.0):

[model_arguments]
v2 = false
v_parameterization = false
pretrained_model_name_or_path = "animefull-final-pruned.ckpt"

[additional_network_arguments]
no_metadata = false
network_module = "lycoris.kohya"
network_dim = 16
network_alpha = 1
network_args = [ "conv_dim=0", "algo=loha",]
network_train_unet_only = false
network_train_text_encoder_only = false
cache_latents = false

[optimizer_arguments]
optimizer_type = "AdamW8bit"
learning_rate = 0.001
max_grad_norm = 1.0
lr_scheduler = "custom"

[dataset_arguments]
debug_dataset = false
# batch size: 4 x 2
# resolution: 640

[training_arguments]
output_name = "shinymas-v2.0"
save_precision = "fp16"
save_every_n_epochs = 1
max_token_length = 225
mem_eff_attn = false
xformers = true
max_train_epochs = 31
max_data_loader_n_workers = 0
persistent_data_loader_workers = false
gradient_checkpointing = false
gradient_accumulation_steps = 1
mixed_precision = "fp16"
clip_skip = 2
log_prefix = "shinymas-v2.0"
lowram = true

[sample_prompt_arguments]

[saving_arguments]
save_model_as = "safetensors"

v2.1 is tuned from v2.0 with resolution raised to 768, batch size and learning rate halved.

This is a LoRa with the style ofgravemud.art

The base model is Suzumehachi.

If you are the artist whose style is conveyed in part or fully by this LoRa, I suggest that you introduce this tool into your workflow. That way you can please your fans with even more art.

Hi all. It's my first style Lora to emulate the style of a Korean NSFW artist Roborobocap(definitely check him out on pixiv, the art is lit). The lora was trained on all the images available online with Lurielmodel as a base.

BE WEARE, Lora has a number of huge problems:

1) I found it to be verystrong: 0.5-0.7 is my range, if you go above it becomes risky.

2) I used images with text, so a negative prompt with (English text, text) is necessary.

3) Overall stability of outputs and quality of lora is kinda shit (hopefully v2-v3 will solve these problem)

I daily drive ComfyUI so you might not see prompts in civit interface (but If you also use ComfyUI just drop my pngs into the web interface and you'll get my workflow).

It is an nsfw lora and it is tailored for mature content, but you can get sfw images with it too.

Hope you all like it. If you have insights or your prompts are better then mine (which is likely) feel free to shit in the comments :)