Spent several hours curating and selecting a set of 29 very old hand-painted photographs from China and Japan, which have been used as the basis for training these embeddings. This means that when you use these embeddings to generate your own images, you will be able to recreate the aesthetic and style of these ancient photographs with remarkable accuracy and detail. Try experimenting with futuristic prompts of characters.

My first attempt so please don't judge too harshly. Although any tips for improvement I would really appreciate.

I can’t say much about Maddy Cheary, just another bimbo model from Instagram, but I really like her facial features.

Version 2: Completely retrained model from scratch

I recommend reduced weight. For example, "(madd1chear2:0.9)" works fine in most cases

Also, for maximum similarity to the original, I recommend adding "busty" before the trigger word and "freckles" in your prompts.

made from a 1000 images of cam models

All of my embeddings are trained with v1-5-pruned-emaonly with the vae-ft-mse-840000-ema-pruned.vae.pt VAE

Stephany Gonzalez, fitness and Instagram model.

Recently, one of my favorite author @JernauGurgeh posted a TI with her. He inspired me to try my version. I tried very hard to convey not only her facial features, but also her amazing fitness shape. I hope I succeeded)

I hope someone likes it I will be glad to hear any feedback!

Trained on several hundred images of joints being smoked, rolled, held, stacked, weighed and blazed

A textual inversion of the Fox News reporter Emily Compagno.

Stormi Maya

This is the Textual Embedding counterpart to my Stormi Maya LoRA. Stormi is a bombshell American model.

Do you like my work? Buy me a beer!

[ SDv1.4, 5e-03, 4000, v/s .197 ]

Trained on SD 1.5 base model.
Multiple models used in examples. No hires fix or face restore used.

Sanna Marin

Ekaterina Schulmann - d4p

All of my embeddings are trained with v1-5-pruned-emaonly with the vae-ft-mse-840000-ema-pruned.vae.pt VAE

Images are a from both the consistentFactor_v3Light and the realisticVisionV20_v20 models.

Here's a test versus the celebrity already in most models:

Ashley Pink

Ashley is an up-and-coming American adult film actress

Do you like my work? Buy me a beer!

[ SDv1.4, 5e-04, 3200, v/s .204 ]

Anya Ivy

Anya is an awesome American adult film acress

Do you like my work? Buy me a beer!

[ SDv1.4, 5e-03, 3600, v/s .203 ]

Valenti Vitel whose real name is Valentina Grishko is a model from Russia.

I recommend reduced weight. For example, "(v1tell:0.9)" works fine in most cases.

I hope someone likes this TI. As always, I will be happy to receive any review or feedback

Generated from fitness model images. if it creates green clothing just put color infront of your clothing prompt and/or add green negative prompts.

Why three different versions? Lower step models looks more realistic but loves the green color more, higher step models give more diverse colors but look more stylized. If you can't choose go with step 2000.

A request of @Whitelionofchrace.

Stella Maxwell is a Belgian/British/Australian supermodel, a former Victoria's Secret Angel.

1850-step TI trained on a dataset of 20 images with my usual settings.

Curious about my work process? I have summarized it here.

Appreciate my work? My TIs are free, but you can always buy me a coffee. :)

Sacha Dhawan is a British actor of Indian descent. Currently on Hulu's 'The Great', and previously seen in Doctor Who and Marvel's Iron Fist. Also does some voicework, radio plays, and traditional theatre.

Trained for a little over 7.5k on SD1.5. Further attempts just led to his head looking like a fat peanut.

Lily McMenamy is an alt model with a unique look. Large lips and exaggerated eyes. This embedding is mostly an experiment to see if I could get SD to handle her features. It somewhat worked, but I'm still posting it because she's different than your usual instagram girl with perfect features, and can actually fight against the tendencies of some styles to make everyone look the same.

Trained on vanilla SD1.5, but please don't use it for generation.

A request of @Sonet and @zepekenh0.

Yanet García is a Mexican weatherwoman, model and actress.

1000-step TI trained on a dataset of 20 images with my usual settings.

Curious about my work process? I have summarized it here.

Appreciate my work? My TIs are free, but you can always buy me a coffee. :)

Alexis Tae

Alexis Tae is an American adult film actress

Do you like my work? Buy me a beer!

[ SDv1.4, 5e-03, 3600, v/s .200 ]

Aaliyah Hadid

Aaliyah Hadid is an American adult film actress

Do you like my work? Buy me a beer!

[ SDv1.4, 5e-03, 4000, v/s .203 ]

Audrey Fleurot French actress

Trigger Words "AudreyFleurotV1"
I accept all constructive criticism

Probably going to Hell for this.

Introducing, the Lord and Savior Jesus Christ.
Just in time for him to come back to life in all of your wildest AI generation fantasies, and bring you fancy eggs filled with goodies.

Trained on all of the classic images you've seen around over the years of the ol' guy, (but let's be honest, probably not accurate to what he actually looked like if real.)
Hell, the Bible doesn't even really talk about what he looked like aside from the fact that he was probably ugly.
This embedding however, generates the total smoke show you've all come to love.

I WOULD LOVE TO SEE WHAT YOU MAKE, I PRAY THAT YOU ALL WILL THROW SOME RESULTS UP HERE.

Happy Easter ya'll.

Lower CFG gives best results, I hang out around 4.5.

TIP - DON'T USE MANY (if any) DESCRIPTIVE WORDS ABOUT THE SUBJECT IN YOUR PROMPT WHEN USING A SUBJECT BASED TI.
Textual Inversions are trained on the appearance of the subject, therefore every time you add a descriptive element (hair color, body type, etc) to your prompt, you are fighting the embedding and results will be less accurate. Stuff like hairstyles (hair in a ponytail) usually will not fight an embedding and can help if they describe actual traits in the embedding, but if they aren't true to the character it will only fight it. And too many of them definitely will fight it.

When using textual inversions for people/characters, use a formula like:
<embedding> + scene + pose/outfit + environment/lighting/quality triggers

Lower CFG also gives more strength to the embedding vs your overall prompt. I’ve found a lower CFG works better for any embedding on this site.

TEST PROMPT
(I use a system the uses <> for TI triggers, remove those if your UI doesn't utilize that.)

<fenn_jesus>, hyper realistic photograph, portrait of a man, holy, mythological, mythology, highly detailed, photorealism, best quality, 4k, (solar eclipse, outside, dark sky), cinematic lighting, (looking at viewer), shot on Canon, detailed skin, sharp focus, rim lighting, two tone lighting, dimly lit, low key

far away, ugly, low-res, blurry, short hair, bad anatomy, anime, cartoon, 3d render, illustration, disfigured, poorly drawn face, (text, watermark, signature), mutation, mutated, extra limb, ugly, poorly drawn hands, missing limb, floating limbs, disconnected limbs, malformed hands, out of focus, long neck, long body, disgusting, poorly drawn, mutilated, mangled, old, surreal, far away shot, monochrome,

Alena Omovych is a ultra-bimbo Instagram/Onlyfans model from Ukraine.

I highly recommend reduced weight. For example, "(omov1ch:0.9)" works fine in most cases.

I hope someone likes this TI. As always, I will be happy to receive any review or feedback

Mai "Mary" Lee is an OC character: a mixed Chinese American young woman, with heavy influencer vibes.

Generated with an amalgamation of models and various references. Created for my own use in transformation captions and morphs, I'm making the embedding available for others to play with.

Faustine Bollaert French journalist and presenter
Trigger Words "FaustineBV1"
I accept all constructive criticism

Trigger word : lilawest

Dexter Series #4

Lila

Thank you for visiting my post <3 Please leave review images for feedback

Sterre Koning is a Dutch singer and actress

Used model: CyberRealistic V1.3

Stefania Liberakakis is a Greek-Dutch singer, voice actress and YouTuber.

Used model: CyberRealistic V1.3

Trigger word : sstark

GOT Series #3

Sansa Stark

Thank you for visiting my post <3 Please leave review images for feedback

By request! You could also just use "Caity Lotz" as a prompt in most models but this is a little more character specific.

All of my embeddings are trained with v1-5-pruned-emaonly with the vae-ft-mse-840000-ema-pruned.vae.pt VAE

Images use the consistentFactor_v3Light model.

This is an embedding of my character from the Korean game Lost Ark. She works best in anime and semi realistic models. A decent amount of her training images used rabbit ears so it might be necessary to put that in the negative if you don't like them. She also has a tendency of generating with white highlights in her hair, so white hair in the negative can also be helpful.

Triggers: gunslinger, red hair, red eyes, ponytail

negatives: white hair

This one is a bit tricky on the sample images. I'm using img2img with unrelated color images set to .89 Denoising strength in order to force color results. 90% of the training images are good quality screenshots from the show in black and white.

All of my embeddings are trained with v1-5-pruned-emaonly with the vae-ft-mse-840000-ema-pruned.vae.pt VAE

Images use the consistentFactor_v3Light model.

Scarlet is the first in my NBDY series for anyone who wants a consistent character for their renders.

Scarlet is a Nobody, her unique look is a reflection of their individuality and originality, rather than an attempt to mimic someone else. The character's physical traits combined to create a look that is entirely their own, with no clear inspiration from any one person.

Scarlet has pale skin, freckles, thick eyebrows, and red hair but will gladly try out other colors and styles.

To use this embedding with the Automatic1111 web UI, save the downloaded file to /embeddings/nbdy-scarlet-sd15.pt, and then include the trigger word nbdy-scarlet-sd15 in your prompt.

Inspired by Zovya's Nobody series. Be sure to check out other contributions to the Nobody tag.

A request of @Dingo00.

This model currently has 2 versions:

Version 1.1 is step 1000. At first I uploaded it as an "early access" version, but, after extensive testing, I've decided it's by far the best and the most accurate one.

Version 1.0 is step 1850. You can also obtain fantastic results with it, I think, but overall I feel her facial features are not as realistic as in 1.1.

Both versions are quite NSFW, I must warn you, though. So, if you don't want Tera flashing her boobs at the earlist opportunity, you might want to add some negative prompt to that effect. :)

You probably know I haven't done many TIs based on pornstars (actually, only one if I'm not mistaken), but Tera Patrick is a special beauty, so she's an exception. This is not an attempt to capture her late 90s looks, though, since there are not many quality pics of that era online (at least that I could find). The pictures I've used for her dataset are more of the mid 2000s to early 2010s era.

Curious about my work process? I have summarized it here.

Appreciate my work? My TIs are free, but you can always buy me a coffee. :)

A request of @TheEliteChief01.

Lisa Manoban is a Thai singer and dancer, member of the K-pop band Blackpink.

1000-step TI trained on a dataset of 20 images with my usual settings.

Curious about my work process? I have summarized it here.

Appreciate my work? My TIs are free, but you can always buy me a coffee. :)

Trained with v1-5-pruned-emaonly. 1500-step TI trained on a dataset of 23 images.

All of my embeddings are trained with v1-5-pruned-emaonly with the vae-ft-mse-840000-ema-pruned.vae.pt VAE

Images use the consistentFactor_v3Light model.

Trained on several hundred pictures of bongs, waterpipes, bubblers and etc, Bongz can draw a reasonably useful device for your smoking needs

Background:

This Textual Inversion embedding bears a striking resemblance to a famous JAV Actress with world-renowned assets.

Trained using base SD1.5 on 50 high quality 512x512 cropped images of the subject "modeling" with text removed from the images. The embedding is 16 tokens--that seems excessive, but it's the only way I got good results.

I'm very new to this, and I've found that training is, unfortunately, an art, not a science. So, this embedding isn't perfect and has some limitations. Please keep that in mind when you rate!

Note: Remove the ".zip" from the end of the file name and place in the embeddings folder. The default trigger is 1hit1.

Important Points:

• For whatever reason, the embedding works worse for realistic generations and on the base SD1.5 model. Other models and artistic styles produce better results.

• The results get much better with prompt engineering and, especially, lengthy prompts.

• You definitely have to adjust the weight of the trigger word relative to the rest of the prompt.

• Generations of the subject's assets can be prone to distortions, especially when they are exposed. This doesn't occur often, but does happen. I believe this was due to the fact that the extreme size of those assets made it impossible to fit both them and the subject's face in a close-up when cropping to a square image. This seems to be mitigated by using words in the prompt to describe those assets.

• You will need to include words like naked, nude, and topless in the negative prompt to avoid accidental wardrobe malfunctions (we wouldn't want that, now would we?).

• I've found that including open mouth and teeth in the negative prompt improves generations.

• Unless you include strong and specific descriptions of your intended background scenery or setting in your prompts, generations will tend to incorporate the following elements: palm trees, desert plants, brick walls, distant buildings, sunny weather, and the interior of houses. Prompting for background scenery and setting elements seems to mitigate this effect.

• For further tips & tricks, see the PNG info in the sample images. Yes, my prompting style is weird and complicated, but it works, right?

Samples:

These sample generations were done in Galaxy Time Machine Photo for You and Deliberate. The crappier looking ones were made using simple prompts; the better looking ones required extensive prompt engineering. I didn't use HiRes Fix, Face Restoration, ControlNet, Img2Img, Lycoris, or Negative Embeddings on any of these generations. I believe I used model LORAs on a few, but none for concepts or subjects. Your results will probably improve if you use any of those!

A request of @DCl.

Jessica Cediel is a Colombian tv host, journalist and actress.

1000-step TI trained on a dataset of 20 images with my usual settings.

Curious about my work process? I have summarized it here.

Appreciate my work? My TIs are free, but you can always buy me a coffee. :)

Textual inversion of pornstar Romi Rain.

This is a Style which brings out a morbid Widow aesthetic. Simply call Style-Widow_SD2 in your prompt and your female characters will be ready for mourning, Victorian widow's weeds style!

If you'd like the effect to come through stronger (some models need it), try bumping up the emphasis on Style-Widow_SD2 - or adding it into your prompt multiple times.

Please consider joining my Patreon! I have a comprehensive tutorial on how this, and my other popular TI embeddings are created! Happy to share my SD knowledge, through advanced tutorials. Also news, settings explanations, adult-art, from a female content creator (me!) patreon.com/theally

Eiza Gonzalez

Eiza is a Mexican actress and singer. She was Monica "Darling" Castello in Baby Driver and Nyssiana in Alita: Battle Angel

Do you like my work? Buy me a beer!

[ SDv1.4, 4e-03, 5200 steps, v/s .2003 ]

Trigger word : daetrg

GOT Series #2

Daenerys

Thank you for visiting my post <3 Please leave review images for feedback

Robert John Downey Jr. (born April 4, 1965) is an American actor and producer. In 2008, Downey was named by Time magazine among the 100 most influential people in the world, and from 2013 to 2015, he was listed by Forbes as Hollywood's highest-paid actor.

Looking for something... spicy? Here's my Patreon! It will support my work. Also, I recently started a blog where I will gather resources and tutorials about my work here!

Hey Civitai! What you see here is a textual inversion of one of my favorite actors: Robert Downey Jr! I used 40 public images of him with an 8900 steps process. To use it in your works:

• download the file and place it into stable-diffusion-webui>embeddings folder;

• add robdowney65 to your prompt;

• keep the sampling steps around 20/30, and CFG between 7 and 7.5 to work well. Great results with DPM++ SDE Karras & Edge Of Realism;

I created this model using 800+ images of the Gorgeous Ai Shinozaki

I was trained on v1-5-pruned for 5000 steps

In all the images I posted I was using Dynamic Thresholding at 3.5

It works well on dreamshaper_4BakedVae

Scarlitt Scandal

In my triumphant return to contributions you can't share with your kids, I present to you Scarlitt Scandal. Scarlitt is a magnificent American actress who has appeared in many adult films too racy to name here. You may be able to find her using your Google machine, however, so happy hunting.

Do you like my work? Buy me a beer!

[ SD v1.4, 4e-03, 4600, .202 ]

This is a model that creates shapes and images based on the artwork of Ellsworth Kelly. Very abstract. Simple shapes. Bold colors. Simple lines.

Trained on some fruit slice images

All of my embeddings are trained with v1-5-pruned-emaonly with the vae-ft-mse-840000-ema-pruned.vae.pt VAE

Images use the realisticVisionV20_v20 model

Second, and a bit better, attempt at a textual inversion of Youtube personality Baliey Sarien. SD doesn't seem to do too well with tattoos, so they will not be accurate.

这个ti+lora 的组合是我很久之前的工作了，现在看来依然好用。当然，你也可以只用我的ti，而使用别人的lora。

This ti+lora set is something I worked a long time ago, and it still works well.Of course, you could just use my ti and use someone else's lora.

Model

同时使用ti和lora你能在绝大部分模型中都得到不错的效果。

Using ti and lora together you can get good results in most models.

Why ti+lora?

ti和lora一起用能在照片基础的模型上工作的更好，而且不用写一大堆tag。

有时候你不能加太多人物lora的权重，就需要人物ti来补充人物特征。

Use both TI and Lora works very well with photrealistic models,and you don't neet to write a lot of tags.

Sometimes you can't add too much weight to the character lora, so you need character ti to complement the character characteristics.

example:

Realistic:

1girl,(weihua:1.0), <lora:Mika_Misono_Lora:0.65>, (ulzzang-6500:0.3)

normal:

1girl,(weihua:1.0), <lora:Mika_Misono_Lora:0.65>

Vanessa Morgan

Vanessa is a gorgeous Canadian actress known for her role as Toni Topaz in Riverdale.

Do you like my work? Buy me a beer!

[ SDv1.4, 4e-03, 5600, v/s .203 ]

Skai Jackson

Skai is an American actress who was included in Time's list of Most Influential Teens in 2016. She did a bunch of Marvel work and was even in Dancing with the Stars.

Do you like my work? Buy me a beer!

[ SDv1.4, 4e-03, 4800, v/s .202 ]

Riele Downs

Riele is a Canadian actress and dancer who starred in a few Nickelodeon specials many years ago.

Do you like my work? Buy me a beer!

[ SDv1.4, 4e-03, 5800, v/s .202 ]

Madison Pettis

Madison is a stunning American actress who starred in...(checks notes)...Beverly Hills Chihuahua 2. And 3. Oh well, she's very pretty.

Do you like my work? Buy me a beer!

[ SDv1.4, 4e-03, 3800, v/s .200 ]

Lauren McClain

Lauren is a supremely talented American actress and singer, in a group with her sisters. How sweet.

Do you like my work? Buy me a beer!

[ SDv1.4, 4e-03, 4000, v/s .202 ]

Daphne Blunt

Daphne is a beautiful American actress who happens to ahare a birthday with yours truly. She was in Penny Arcade!

Do you like my work? Buy me a beer!

[ SDv1.4, 4e-03, 3200, v/s .198 ]

Daniella Perkins

Daniella is a beautiful American actress and YouTube star. She currently stars in Grown-ish and appeared as a voice actress in SpongeBob SquarePants!

Do you like my work? Buy me a beer!

[ SDv1.4, 4e-03, 3800, v/s .199 ]

Ajiona Alexus

Ajiona is a gorgeous American actress and was amazing in Empire.

Do you like my work? Buy me a beer!

[ SDv1.4, 4e-03, 5400, v/s .202 ]

A very versatile negative embedding that helps generate high quality images. Higher chances of producing normal looking hands in realistic images.

Best Practices:

Keep your token usage to below 75 for best effect. Most of the time, you don't need to add other quality specifiers to your prompts (e.g. worst quality, bad quality).

Usage:

Negative prompt: file_name_of_embedding

If you notice any artifacts, you can try increasing or decreasing the weighting like this: (file_name_of_embedding:1.1) or [file_name_of_embedding:0.8].

Values between 0.8 and 1.2 seems to work the best (clip size 2):

Tested on:

• VersaMix

• ChilloutMix

• Grapefruit

Olivia Wilde

Olivia is a marvelous America actress, best known for her work in the American television series House. She is now directing films and doing quite well.

Do you like my work? Buy me a beer!

[ SDv1.4, 4e-03, 5600 steps, v/2 .201 ]

This person does not exist!

A TI created by merging two of my existing real celeb TIs together. The results were so consistent and her look so interesting I felt I had to share.

Even the name came from a random name generator!

Trigger word : rita00bennet

Dexter Series #3

Rita

Thank you for visiting my post <3 Please leave review images for feedback

It's a first try of a beautiful actress. Perfect model for Sci-Fi.

use sonqua-200 for trigger this embeddings on your favourite model.

Tested with many models, i found tha it makes better images with RealisticVisionV2.0.

Trigger word : mmax00

My First Style Embedding

Thank you for visiting my post <3 Please leave review images for feedback

bad hand

Trigger word : dextermorgan

Dexter Series #2

Dexter

Thank you for visiting my post <3 Please leave review images for feedback

Trained with v1-5-pruned-emaonly. Checkpoint for sample images was consistentFactor_v3Light using ControlNet for posing.

Elizabeth Debicki is an actress known for Man from U.N.C.L.E., Tenet, The Crown, and Guardians of the Galaxy.

Note: I accidentally pulled the previous version of this embedding. This is a new version (in fact probably the 4th version) I have been working on while testing different training methods. I also completely redid the source images!

介绍（中文）

基本信息

该页面为推荐用于 AnimeIllustDiffusion [1] 模型的所有文本嵌入（Embedding）。您可以从版本描述查看该文本嵌入的信息。

使用方法

您应该将下载到的负面文本嵌入文件置入您 stable diffusion 目录下的 embeddings 文件夹内。之后，您只需要在填写负面提示词处输入 badv4 即可。

评价参数

每个文本嵌入将从以下参数描述其特性：

正/负面文本嵌入：如果一个文本嵌入是正面文本嵌入，则您应该将它填写在正面提示词内发挥作用；反之，如果一个文本嵌入是负面文本嵌入，则您应该将它填写在负面提示词内发挥作用。

作用范围：一个文本嵌入主要影响的画面内容。

向量数：代表该文本嵌入占多少个词元。通常来说，您输入提示词中超过 75 个词元的部分会被忽略。因此，一个文本嵌入的向量数越少，您填入它之后剩余输入提示词的空间越多。

向量强度：一个词元向量值的大小。向量强度越高，效果越强。普通词元的向量强度为[-0.05,0.05]。不可将用括号增强提示词等效于增强向量强度。

Introduction (English)

Basic Information

This page contains all the text embeddings recommended for use with the AnimeIllustDiffusion model [1]. You can view information about the text embeddings from the version description.

Usage

You should place the downloaded negative text embedding file in the embeddings folder of your Stable Diffusion directory. After that, you only need to enter "badv4" in the field for negative prompts.

Evaluation Parameters

Each text embedding is described by the following parameters:

Positive/Negative Text Embedding: If a text embedding is a positive text embedding, it should be used in positive prompts. Conversely, if a text embedding is a negative text embedding, it should be used in negative prompts.

Scope: The main content of the image that a text embedding will primarily affect.

Vector Count: The number of token that the text embedding represents. Generally, the part of a prompt containing more than 75 tokens will be ignored. Therefore, the fewer the number of vectors in a text embedding, the more space you will have left to input the prompt.

Vector Strength: The magnitude of a token's vector. The higher the vector strength, the stronger the effect. The vector strength of token is usually between [-0.05, 0.05]. Using parentheses to enhance prompts is not equivalent to increasing vector strength.

引用 / References

[1] AnimeIllustDiffusion Model Webpage: https://civitai.com/models/16828/animeillustdiffusion

Trained on the Topps Mars Attacks artwork

A Dark fantasy style embedding i made using 24 images to train. Turned out pretty good, and you get results just by using the embedding. No trigger words. Also adding things like: scenery, forest, nature, snow, outdoors, glowing, tree, water, ruins, dark, gives some pretty good results.

The embedding is good for just making sceneries, but works really well with characters. You will kinda get a blend of dark fantasy + your prompt. Have tested it on different models and it works really well on most of them.

Works great with:

CFG Scale: 7-9

Sampling Steps: 20-30

A textual inversion of the Fox News reporter Katie Pavlich.

Trained on the artwork of the children's book "You will go to the moon".

Appreciate my work? Can you buy me a coffe please :)

Marilyn Monroe: born as Norma Jeane Mortenson; Los Angeles, June 1, 1926-Los Angeles, August 4, 1962) was an American actress, singer, and model. Famous for playing "blonde bombshell" comedic personas, she became one of the most popular sex symbols of the 1950s and early 1960s, as well as an emblem of the sexual revolution of the time. one of the most sought-after actresses for a decade, and her films grossed $200 million (equivalent to $2 billion in 2021) by the time of her death in 1962.6 Long after her passing, Monroe remains a major cultural icon pop.7 In 1999, the American Film Institute ranked her sixth on its list of the greatest female film legends of Hollywood's Golden Age.8 Various film critics and the media have cited Monroe as one of the best actors who never received an Oscar nomination.

Used model: Realistic Vision v1,3

recommended parameters:

Prompt:

RAW photo, a close up portrait photo of 26 y.o woman in wastelander clothes, long haircut, pale skin, slim body, background is city ruins, (high detailed skin:1.2), 8k uhd, dslr, soft lighting, high quality, film grain, Fujifilm XT3

Negative Prompt:
(deformed iris, deformed pupils, semi-realistic, cgi, 3d, render, sketch, cartoon, drawing, anime:1.4), text, close up, cropped, out of frame, worst quality, low quality, jpeg artifacts, ugly, duplicate, morbid, mutilated, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, blurry, dehydrated, bad anatomy, bad proportions, extra limbs, cloned face, disfigured, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, fused fingers, too many fingers, long neck

DPM++ 2S a Karras with 25 steps
CFG Scale 7

QR to my page buymeacoffee

A negative prompt embedding for Deliberate V2. Use in your negative prompt to make the image look better. Do not use as a regular prompt.

Trained on the bad image dataset from here:

https://civitai.com/models/17083?modelVersionId=20387

A tribute to our regional grocery chain's mascot - Schnucks makes little dudes in the style of the Schnucks' Soldier mascot.

Trigger word : debramorgan

Dexter Series #1

Debra

Thank you for visiting my post <3 Please leave review images for feedback

trained on AI art of reddit user onche_ondulay

A request of @TheEliteChief01.

Lexi Rivera is an American Youtuber and social media personality.

1850-step TI trained on a dataset of 20 images (most of them provided by @TheEliteChief01) with my usual settings (reducing the learning rate to 0.002 after step 1000).

Curious about my work process? I have summarized it here.

Appreciate my work? My TIs are free, but you can always buy me a coffee. :)

Stable Diffusion is finally old enough to learn about weed. This embedding might be a bit over-baked but it's still up for a good time

This is a Style which brings out a beautiful bridal aesthetic. Simply call Style-Bridal_SD2 in your prompt and your female characters will be ready for their wedding day!

If you'd like the effect to come through stronger (some models need it), try bumping up the emphasis on Style-Bridal_SD2 - or adding it into your prompt multiple times.

Please consider joining my Patreon! Advanced SD tutorials, settings explanations, adult-art, from a female content creator (me!) patreon.com/theally

This model was trained off a few variations of Robert Motherwell's Elegy to the Spanish Republic Series. It can create variations of the Elegy to the Spanish Republic style in img2img. It mostly does a great job of creating abstract art when using it in txt2img, along with ControlNet. You really need to give the model a base on which to work. The model was created for Stable Diffusion 2.1.

Scarlett Ingrid Johansson (born November 22, 1984) is an American actress. The world's highest-paid actress in 2018 and 2019, she has been featured multiple times on the Forbes Celebrity 100 list. Time named her one of the 100 most influential people in the world in 2021. Her films have grossed over $14.3 billion worldwide, making Johansson the highest-grossing box office star of all time.

Looking for something... spicy? Here's my Patreon! It will support my work. Also, I recently started a blog where I will gather resources and tutorials about my work here!

Hey, nice people! So happy to present this one! Today I have for you a textual inversion of Scarlett Johansson! I used 40 public images of her with a 9000 steps process. To use it in your works:

• download the file and place it into stable-diffusion-webui>embeddings folder;

• add scajohansson84 to your prompt;

• keep the sampling steps around 20/30, and CFG between 7 and 8 to work well. Great results with DPM++ SDE Karras & Edge Of Realism;

As a big "Thank You" to @Clarq for creating Pixie AKA Nadia, I thought I'd create a TI of her friend and very frequent 'sparring partner' Natasha Vale.

For those unfamiliar with her, she was a major British glamour model and occasional pornstar of the late 90s / early 00s. Also worked as a lapdancer, TV presenter on Granada Men & Motors (usually when the assignment required the presenter to remove clothing!) and personal trainer.

Emb can replicate every detail of the object, including its body, teeth, cup size and everything else, something that Lora is incapable of doing.

Compared to Emb, Lora can only be considered a 'sophisticated wax figure.'

Don't assume that Lora, with its hundreds of MB, is necessarily better than Emb just because Emb only has a few KB (I used to think so too).

In fact, it's the opposite: Lora is not suitable for replicating objects, as the technical documentation states that Lora has a 'valley of terror' effect when it comes to facial reconstruction, meaning that the result looks more like a wax figure than a real person.

Lora is only suitable for low-end computer training and clickbait production shearing the sheep, where high traffic does not necessarily translate into high quality. It's like the difference between a shallow celebrity and a respected seasoned performer;

True high-quality models require Emb. Early computers were the size of houses, but now chips are much smaller - size doesn't always equate to quality.

The preview images are all REAL, single-generated images, and there is no 'monkey play' image that has been inpainting a second time.

The full body facial image is a bit troublesome (generated at once, not partially redrawn). I don't know if it is a defect of sd1.4. My intuition is that sd1.4 is superior to sd1.5 in restoring faces at close and medium distances, but at long distances, 1.5 is better than 1.4. Next, I will make a version based on 1.5 training to verify it.

emb可以把对象的一切细节都复刻下来，身体、牙齿、罩杯，一切，这是lora无法做到的。目标左边门牙比右边门牙略短这个细节emb都能捕捉到。

lora相比emb最多只能算是【精致的蜡像娃娃】。

不要因为emb只有几kb,而Lora有上百Mb就先入为主的觉得lora一定比emb牛逼（我早期也这么想），事实上恰恰相反，lora根本不适合复刻还原物品（原技术文档写明了lora在人脸还原上有一种“似像非像”的“恐怖谷效应”，大白话就是：如同人物的蜡像）。

lora只适合低配置电脑体验训练和营销号快速出图割韭菜罢了，流量大不代表品质高，只代表门槛低，如同流量明星和老戏骨的差距，真正的高上限精品模型还得是emb。

早期的计算机有房子那么大，现在的芯片才多大？尺寸大小不等于品质的大小

预览图没有“二次重绘”的“猴戏图”，都是真实的一次生成图，好的坏的都看的到。

Anna Nyström is a Swedish fitness model and blogger. She has gotten a massive following on Instagram where she has around 8.3 million Followers

I highly recommend reduced weight. For example, "(annystr:0.9)" works fine in most cases.

I hope someone likes this TI. As always, I will be happy to receive any review or feedback

Karen Sheila Gillan (born 28 November 1987) is a Scottish actress. She gained recognition for her work in British film and television, particularly for playing Amy Pond, a primary companion to the Eleventh Doctor in the science fiction series Doctor Who (2010–2013), for which she received several awards and nominations.

If you like my work, a review is more than welcome! Also, follow me on Twitter or Instagram!

Looking for something... spicy? Here's my Patreon!

Also, I recently started a blog where I will gather resources and tutorials about my work here!

Hey, people of Civitai! I was off for a some days but here I am! Today I have a textual inversion of the lovely Karen Gillan! Hope you'll like her ;) I used 30 public images of her with a 6700 steps process. To use it in your works:

• download the file and place it into stable-diffusion-webui>embeddings folder;

• add kargillan87 to your prompt;

• keep the sampling steps around 20/25, and CFG between 7 and 8 to work well. Great results with DPM++ SDE Karras & Realistic Vision;

Hi guys!

It's the same history of the SD 1.5 embedding , but ported to SD 2.1 512x512.

If you wanna test it, the magic word is: trInkdBer or art by trInkdBer .

Greetings!!!! :8)

New version is now available and visible: https://civitai.com/models/28913/lorraine-baines

edit: incredible. they deleted every single image in this post.

Laura Vandervoort

Laura is a Canadian actress, eh, who played Supergirl in Smallville.

Do you like my work? Buy me a beer!

[ SDv1.4, 4e-04, 6600, v/s .1996 ]