Your dedicated onlyfans altar, summon your goddess here.

Going forward I'm updating all my character LoRa to Lycoris as my tests shows promising result with lycoris to reduce memory and better mixing. Instead of multiple models, I will consolidate to single model with multiple versions.

===

Open to suggestion or request

NOTE : I will keep updating more actresses in versions update

---

As usual, all models used in this lycoris are all 21+, I'm doing this for fun only, and please do not use for commercial purpose. Apparently, the guidelines stated that explicit nudity is prohibited for real person, even though that they have posted nudity online as well, so please be mindful with nsfw posts.

===

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.

This is a LoHa model trained from 100+ images about galaxy&forest background from 3 different sources. I use multires noise to generate light&shadow of dataset.The tone is very close to green and purple vines,and that's where the name vines comes from.

You can use the LoHa to design light particle,sky,cloud,ancient trees and other scenery.

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

Trained on the Flurks NFT collection by Stonetoss.

Supports most existing traits.

For better variety but also consistency, like in my examples, do this:

• Set lora weight at like 0.5

• Enable controlnet segmentation model, with a simple real flurk, weight 0.1-0.2. This will give a bit better shape consistency

You can tip me at:

• Eth / L2s / Sidechains any tokens:

  • 0x3345908DE481886DCd33c3c73585010FA2DFBbc1 or peeledkot.eth

• Monero:

  • 82j8HrA34JEHBLF71pjgB6Q7csbgBuMpSENeFjznjmqUFxDChYReA8GCPscoba4RoWXsaGx1SgpQeWuxfnp4WfyPQznHvZJ

This Lycoris project aims to train more photographic data to improve the quality of this concept.

A takt op destiny Artstyle LoHA trained on 3100 images from the anime, best weight 0.3~0.8

Note: reason why it's similar to ufotable style because some ufotable staff had worked on the anime alongside MadHouse & MAPPA

english↓

This is lora by sophia valentine, a member of the virtual youtuber group nijisanji.

The strength of lora is recommended to be around 0.8~1.

Trigger words:sfia,1 girl

It also works well if you put the words that occur frequently by putting lora itself through analysis.

It is difficult to get the three moles on her face exactly right. If you are particular, it may be better to draw them later on the generated picture.

This may not be a lora with satisfactory accuracy since it was made as a trial.

Please enjoy it if you like.

https://www.anycolor.co.jp/guidelines/

japanese↓

これは、バーチャルyoutuberグループのにじさんじに所属しているソフィア・ヴァレンタインのloraです。

loraの強度は0.8~1くらいがおすすめです。

トリガーワード:sfia,1 girl

彼女の顔にある3か所のほくろを正確に出すのは難しいです。こだわる方は生成した絵に後から描いた方がよいかもしれません。

また、lora自体を解析にかけて頻出するワードを入れてみると効果があります。

お試しで作ったので、満足のいく精度のloraではないかもしれません。

よければお楽しみください。

Since I don't own a GPU and the only way to continue creating LoRAs is to rent GPUs (free Colab, prohibited Stable Diffusion, and remote UIs), I have chosen to do so with Paperspace. Please check out my Ko-Fi page if you are interested in supporting me.

This is a LoHa LoRA! You need the extension!
Why? LoHa have virtually no effect in output art style and they do a lot better with multi-concepts.

Trigger words for V1.20:

For original clothes: <lora:Fulgorev120:0.8>(fulgore, arm blade)

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

Sample images generated from Prototofustrawshi.

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 53 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.

Manga style looks like Anime, but it is mainly Lineart. Anime has no line-hatching for shading, Manga does. Also manga book often focuses in 1 story at a certain timeline, not like comic book with many related stories binding together in many ways.
This model is strong. Suggesting weight 0.4 to 0.8 depending on based model. Higher weight might have panels and speech bubbles.

Dataset is from manga books in bookstore.

注意：我的模型经过 locon 训练，所以需要安装LoCon 支持。https://github.com/KohakuBlueleaf/a1111-sd-webui-locon

有没有什么好点子？我准备一点点做山海经里面的奇珍异兽。比如，我想要什么风格或者模型，想看什么什么，留言评论，让我看见。有精力我就试着做一做。

（已出列表：穷奇；火光兽；帝江（浑沌）...）

小科普时间：

帝江，浑沌，这太熟悉了吧？

如果我用直接的方法来翻译这个怪物的名字，它应该是<混沌的化身>。当有人为混沌创造鼻子和眼睛，使其看起来像 "秩序 "的时候，混沌就会死亡。

它是中国古代神话中最凶恶的四大凶兽之一。它出现在哪里，皇帝就会变得昏庸，人民就会变得呆滞，庄稼就会不遵循节气。

有人说他是上古时代的神灵死后的化身。也有人说它是当地社会秩序混乱时人们看到的思想集合。

它有头但没有脸，有脚但不能向前或向后移动，有翅膀但不能飞。它出现在哪里，哪里就会有极大的灾祸。

Little Class time:

If I translate the name of this monster in direct method, it should be <Chaos Incarnate>. When God creates noses and eyes for Chaos to make it look like in "Order", Chaos will die.

It is one of the four most vicious monsters in ancient Chinese mythology. Wherever it appeared, the emperor would become faint, the people would become dull, and the crops would not observe the seasons.

Someone say he is the incarnation of a god from ancient times after his death. It is also said to be a collection of thoughts seen by people in times of local social disorder.

It has a head but no face, a foot but cannot move forward or backward, and wings but cannot fly.

我的Ko-fi网址，里面有一些我做的图片，欢迎来看看。

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: 支持性捐助仅适用于经济独立的人群。Supportive donations are only available to those who are financially independent. PS：感谢每一个帮我买可乐的朋友】

Hey, i would appreciate if you leave a review <3

V4.2 Update! Compatibility with most anime(and even some non-anime) models!(NAI-trained). Dataset expansion to 720 manually tagged images! Better style control!

I worked hard on it, and here it is! Overall, improved everywhere, except NSFW, dataset is still 100% SFW. Also no more latex-y clothes everywhere, that's been fixed. Hats shouldn't appear nearly as much, though, i didn't have that issue personally. General quality is quite a bit higher, since training was completely switched to Adan with optimized parameters special to ranks im using. Features like separate weight decay for unet and tenc were used, tenc weight clipping too.
v4 features expanded dataset, which led this lora to better generalization, and now it follow composition of original model better, which means you're not losing much while using it. Stability at weight 1 improved too.

Well, don't just believe my words, since they might just be lies for your model xD I have not released my own mix im training on yet, and i don't think it'll happen in near future, so, please, test yourself, and don't just believe my words. v4 should generalized styles better, so if it worked in v3, it should work better in v4.

Hey, it's my first time actually posting something that i trained, though, this is not my first attempt or anything, im doing that for quite a while, just finally deciding to share.

Im not most experienced person in this, and i can't experiment more than i already do, since my hardware is not allowing me to even do Dreambooth.
But you can be certain that i really worked hard on getting to what i train today.

LoRA you'll see and read about today is not my only project, i have multiple others, including NSFW, though, im not sure i'll ever release that. But it is likely that i will release more SFW in future, so i really would love to hear your feedback on using this one.

You can contact me about this LoRA through discord, Anzhc#5269, please, leave message, since im not gonna add someone to friends without knowing who they are.
If you'd like to support my endeavours and experiments with money - there will be a Patreon link at the bottom of text, since i will feel bad posting it at the top, just blatantly asking for money without you even reading about my stuff xD

If you're interested in detail about my training processes, you could join D8ahazard Dreambooth discord channel. im active, and share my findings and parameters there. At the bottom part of text i am providing general description of what i was doing.

Well, to actual description down there!

This is a personal training of style, or, averaged representation of an amalgamation of style concepts, if you will.
It's nothing special(wrong, now it's special, at least to me :D -- v4 update), since i can't do much on my GPU, but it works and i personally love it.
---Couple comparisons before we start wall of text---

P.S. v4 1.0 released, so, please, look at it's images for better representation of capabilities of last version, also - no more latex clothes everywhere.

---Text wall starts here---
S12 does not require trigger words, as it was trained as a fine-tune, but ohwx is there if you want to make it more powerful(mostly not needed).
Main features of that lora is abstraction and abstract power of subjects, which im very bad at describing, but i think you'll understand what i mean with example. Base models are usually already capable of understanding that concept, but my goal was to enchance it and make more tangible.

Style itself in base is anime-based, but with huge representation of 3D-like imagery.

Prompt for this: masterpiece, best quality, 1girl, upper body, powerful, abstract <lora:S12 ohwx Adan_18000:0.75>
negative:(worst quality, low quality:1.4)
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 3914799726, Size: 512x768, Denoising strength: 0.7, Clip skip: 2, ENSD: 4537, Hires upscale: 1.7, Hires upscaler: Latent (nearest-exact)

It is very simple to use. Preferrable weight of S12 LoRA is ~0.5-0.75, since 1 give too much of an effect, and will make image too dark for my liking, since it was trained with offset noise. In combo with other LoRAs it can be put down to 0.25 even. (And it does combine with some of them quite easily, without introducing visible errors, but i suspect it will not be the case, if other LoRAs are trained with offset noise too, since that'll increase effect)
Note: image below are made with the use of S12, but it is not a solo LoRA in generation and upscale.

(I don't remember prompt, and it was lost in upscaling one, sorry. I was deleting thousands of images cuz no space ;-;)
This image uses LoRA that i called "Effect"(i know im so f'ing original), it is meant to enchance composition a bit. S12 here is at 0.25

Dataset is fairly small, but enough for style purposes, 360 images as of v3.3.2(Will be expanded to ~580 for v4.0, and further for itterations). At the time of version 3.3.2 they were generated on Midjourney v4(one specific sub-style of their, but i don't remember which, probably 4b) with special style prompt, which is called very original, "style12".

I wanted to separate more nice knowledge, like "split theme" for example, since it's not very known by models i use, but dataset representation was not enough to overwrite that token knowledge, so it worse quite poorly and mostly with high weight, if ever.

---Capabilities---
Generally anything, since it's style LoRA. +Special concepts that i already described, + expasion of them in future.

NSFW?
Yes, but it was not trained on NSFW imagery. Dataset is 100% safe for work, so don't expect anything new in that regard and rely on your model knowledge. It will apply style to NSFW, and i quite like it sometimes.

---Example Images---
Since im using custom deep mix that i never posted on the internet, i will be including some images from one of base models of it - ACertainModel, which results you can apply to generally any other anime model, like Anything3. You'll be able to discern those generations by more 2d-ish look, or you won't be able, in that case consider me doing a good job and this LoRA is applicable to most anime models without much concern.

---Training quirks and techniques---



Lower info is valid for versions below v4 0.5

Dataset consists of 360 images generated in MJ, but with special style prompt that does not represent specific style. You can expect overall look to be quite MJ-ish, but... Not that MJ-ish, i guess.

It is trained with Adan D-Adaptation optimizer, which is decently new and experimental, so, expect quite a few rough spots.

Train type is fine-tune, so it will affect whole model for the most part.

Some images were tagged with custom quality tags, different from standard quality tags like "masterpiece" and so on, so they are not affected. They are not represented enough to matter in v3.3.2, so they will not be disclosed.

Techniques like offset noise were used, so it also improves contrast. Preferred weight of lora in that matter is 0.75, as it is too strong at 1.0. Noise offset was at probably ~0.12-0.15.
As it was trained with D-Adaptation, LR is not relevant much, as well as it's not available in main/dev D8 at the time of writing, so i don't know(rip).

Lora ranks i prefer 96/384, but for Adan training i had to lower it, since D-Adapt optimizers are heavy on VRAM. I think it's in range of 48-64/256.

Scheduler used was DEIS im pretty sure.

Since there were bugs that were making training be worse if you generated sample images in middle of training(should be fixed already), i would also specify that it is one-shot, and had no sample generation or saving checkpoints in process.

Note: It is trained on custom deep mix, which is a combo of ACM+AR in step 1, and around 1/3 of AOM2 in step 2, with selective high usage of certain blocks, which turned out to work awesome, and in lots of cases better for my usage. It will work with any other anime model, because of similarities, but best would be described above, specifically AOM2, since style is closer to it.

Images at that point are almost completely manually tagged(pain...), so, lora should be quite responsive to wide arrangement of them, not bleed from concept to concept, and be highly editable.

As i said at the start, here's link to my patreon - https://www.patreon.com/anzhc
It is not centered around AI, but i don't mind sharing my experiments there in WIP state, if that ever would interest patrons.
Just for genuine support, if you want.

A lora to make breasts bigger keeping the stature

General instruction: try some bullshit, random words, something dramatic, use wildcards. It will react differently. This time style is very strong by default, because of 1345 image dataset, so you might want to lower weight.

But it's not overtrained or anything, just very strong, and you will have lots of leaky hair ornament. Other than that it should be decently editable.

Fifth Lazy Style Lora.

I should clarify things, even though it's not really relevant, but for those who are interested in knowing what my process is. I've seen people post tutorials on lazy trainings, though, my lazy trainings are not taking their steps, so don't really compare those. I named them lazy, because i utilize heavy auto-tagging, which i then lazily process with tools available to me, instead of thoroughly going through them for hours and hours to make it most editable.

I do not just go on MJ galleries and gather other peoples generations by keywords and go with that, i use only my own, and you will not find those in public MJ gallery, as they are private.(Yes, i splurge on 60$ MJ sub ;-;)

Based on 1345 image dataset of specific style.

Im not quite sure what that style is, but prompt on which they were generated suggests that it's Cosmic Anime-Oil Fusion, and i guess i see it being true, yeah.

Generally, this is not supposed to be used for anythin complex, because it was not tagged in a way to be, but you still can try and show me your results :3

GPTS Lazy Styles will have a lot of bleed, lower than average editability(but better than it was in GPTS1)

If you want it to be editable and still have some style effect, try weight from 0.5 to 0.75. That should make it editable and still be looking good :D
Maybe in this case you do want weight to be lower, as this particular one is usually abstract in backgrounds, and you will have hard time fighting leaks, because of dataset size.

Generally, this dataset processing used my custom tag filtering experiment, testing of which on GPTS1 led to increased editability, so, it might not be as hard to edit this one, but still, main purpose is just to get that bit of "wow" effect style.

Lighter version might be available later.

LoCon of a style that I don't know what to call

RECOMMENDATIONS:

The model is not particularly fastidious and seems to work well with character locon models. It doesn't need any special tags. This model is more aimed at character images on a detailed background. Looks like I was able to achieve the nice lighting I wanted (but it doesn't always work).

Good luck!

V2 has been added. Now this changes everything, as long as you understand how the prompting works.

This locon can help to create multiple types of poses and body distorsions related to the act of swallowing a person whole. Some poses work very consistently, some less than others. I would advise giving a look to the dataset to understanding the logic behind the prompts.

Put the strenght at 0.7 or 0.8. It's the best compromise between recognizing the poses and rendering quality. 0.8 helps with head swallowing.

The main prompt is "swallowing whole". I would suggest adding precisions concerning the angle too. The vore belly can be triggered with "big belly", and i believe that adding "2girls" helps producing the swallowed body parts.
Compared to v1, this version is very versatile, and pretty much anything is possible. It's like the AI managed to understand what swallowing whole is. I would advise to be as precise as you can with the prompting.

For the head swallowing part, it's a lot more consistent with the right angle (namely, "mawshot" and "close up", or things like this.

I think things are more consistent when you precise what is being swallowed (visible, not inside the body). For example : "swallowing lower body", "ass and legs inside of mouth".

Hires is very important, otherwise the picture will have troubles rendering correctly the details of the body parts swallowed. Not any upscaler will do the trick, too. Latent destroys the picture, as do a high denoising strength. So I suggest R-ESRGAN 4x+ for the upscaler, and 0.2 for the denoising strength.

For negative prompts, I would of course suggest "easynegative" and "bad hand" (both are textual inversions downloadable on this site), but also "penis" (because she meat being swallowed is sometimes interpreted as a penis). "Letterbox" so that you don't have cropped pictures, "english text" and "speech bubble" and of course "nipple on belly" (not guaranteed). For security, I sometimes put "scales" too, because they sometimes appear on the bodies (except if you want them of course).

You can put "big belly" in the negative tags, or "narrow waist" in positive, if that's what you want. I did not test it a lot with flat bellies though.

Now, for the limitations. First off, some poses or concepts are more consistent than others. Second, you probably won't get good results on your first generation. Then, I believe that that version, although widely superior to the v1, does not appreciate deriving from a 512x512 resolution for most concepts (hires is advised, of course). Lastly, I did not test it with other loras much and can't say how this fares, and it seems that the lora forces it's own artstyle, and it's hard to derive from it.

I did not test it much with models other than AnyLora. AnyLora is great though, I highly recommend it.

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License

the regular creativeml-openrail-m license isn't applicable for this work

Feel free to tag me if you use this model, I'd enjoy seeing what you do with it twitter(@ihtke)

FIRST 4/25/23

Lemme take u in deep short backstory of my inspiration

Blue Moon🥰 from Cyberpunk 2077 the breathtaking character as game as well.

When had i first heard and learnt about US CRACKS..... oy boy(long sigh). I instantly wanted to meet them in the game (ngl i love every single quest that i completed). And eventually i have been taken a chance to meet them that was abrupt+awesome (as now remembering *incoming call Kerry*

Kerry: drop whatever u doin' meet me at the marina, don't make me wait )more spoilers ahead so u better play it by yourself)

Well...(these warm memories still chasing me) lemme tell u of model a little bit.

Fist thing first all the data that i could scrap and i did were game engine screenshots(intensively waving to the nsfw enjoyers) made by me+from net.

(do not frustrating now there's always is a solution see about that version and +nsfw capabilities also depends of models)

Second as per of data lack this training was a challenge one (as any other train in kohya-ss). I've spent literally a whole(~26h.) day only on trainings (in the name of science i suppose...yeah)

To start🤔:

(masterpiece, best quality, detailed:1.4), Blue Moon cyberpunk, 1girl, eyes, lips, spiked collar, blue hair, long hair, black jacket, yellow collar, solo focus, pink baseball cap, leather jacket, denim, looking at viewer, realistic, cyberpunk, head tilt, outdoors, grey eyes, upper body

(masterpiece, best quality, detailed:1.4), Blue Moon cyberpunk, 1girl, \\m/, bangs, simple background, blue hair, blunt bangs, bracelet, breasts, yellow choker, crop top, double bun, double v, double w, jewelry, lips, looking at viewer, see-through skirt, sport bra, small breasts, solo, standing, tank top, v, v over eye, w, wristband, grey eyes

Important note:

issue with

Kiroshi optics

(needs further investigation)

TL;DR SD sometimes can be a real bitch.

When u trying to make pupils of characters more natural and live then SD will always blur them out and contrary when u gonna make them double on each eye he's mostly just giving up. I recommend whether INPAINT(sketch) or Photoshop to make it properly).