A commission of @lemke. (Sorry I kept calling her "Ana", mate. xD)

Eva-Maria Lemke is a German journalist and tv anchorwoman.

1500-step TI trained on a dataset of 15 images with my usual settings. (Reduced 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. :)

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

use sonqua-200 (v0.1)or sonqua-2000 (v0.1) for trigger this embeddings on your favourite model.

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

Textual inversion model based on 20.000 photos of gay bears, muscle bears, daddies and chubbies. NSFW

This is a simple dragon embedding to change the Stable Diffusion 1.5's interpretation of the word dragon. Trained on v1-5-pruned-emaonly.safetensors
Trained with 21 dragon images for 700-ish epoch's.

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

As this is an animated character my reference material was 2D and 3D art. Results are model specific.

y'all just want the pretty ladies I guess? :D

Please leave comments! And show off your work!

Who doesn't love a girl in leggings?

This embedding is the first iteration (more to come) of models geared towards producing those perfect, photorealistic, yoga pants/leggings that we all love!

follow me on Instagram, Youtubeand my website.

Suggested weight 0.7-0.9. Adding "intricate detail" tends to make him look shrink-wrapped and overcooked. Open to suggestions on refining the training process.

This is still a WIP, but have to start somewhere. This was trained on ~14 images with the following settings:

Learning Rate: 0.002

Batch Size: 3

Gradient Steps: 2

The TI is nowhere near where I'd like it to be, with the right prompts you can get some decent looking images however they will tend to be low-resolution.

Mya Mays

Mya Mays is an American adult film actress.

This will be the last porn star I do for a little while, just sayin.

Do you like my work? Buy me a beer!

[ SDv1.4, 5e-03, 3600, .201 ]

Nia Nacci

Nia Nacci is an American adult film star

Do you like my work? Buy me a beer!

[ SDv1.4, 5e-03, 3400, .205 ]

Sarah Banks

Sarah Banks is an American adult film star

Do you like my work? Buy me a beer!

[ SDv1.4, 5e-03, 3400, .198 ]

Kira Noir

Kira Noir is a magnificent American adult film actress. Highly recommended.

Do you like my work? Buy me a beer!

[ SDv1.4, 5e-03, 3000, .199 ]

TRY OUT ALL VERSONS

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. This TI embedding works great with most fine tuned checkpoints.

PS: Use the generation parameters of the sample images provided and then try experimenting with different combinations. Do share your generations, I would love to see them. Thank You.

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

Michael C. Hall is an actor most recognizable and known for playing Dexter Morgan on Showtime's TV series Dexter.

I saw another TI of Dexter, and thought it was decent, but I wanted to see if I could do any better since I was already planning to try my hand at him.

Trained on 25 images of Michael C. Hall as Dexter. Mostly screencaptures from the TV series.

I mostly used closeup for the dataset and trained his appearance, but not his attire or other elements from the show. I haven't messed around with LoRAs yet, but this seems like the type of thing that one of those would be useful for.

For what it's worth, I was going to showcase more images with Dexter stalking different people, but ran out of time so figured the couple I had were good enough. I have to learn Latent Couple or other techniques better to make better use of these types of photos.

Faith is the sixth member of the LastName family, for anyone who wants a consistent character for their SD 1.5 renders.

Faith is almost certainly out of your league. She's a musician and gymnast, who dreams of running away to join Cirque du Soleil (and almost certainly will). She was born during golden hour. She’ll take you camping on the beach, then wear a leather jacket over a loose-fitting top and somehow make it impossibly cool. As I said: she is almost certainly out of your league.

To use this embedding with the Automatic1111 web UI, save the downloaded file to /embeddings/faithlastnamesd15.pt, and then include the trigger word faithlastnamesd15 in your prompt. (All of the sample images have full prompt metadata, and are straight out of Standard Diffusion  - no upscaling, inpainting, or img2img was used.)

A female character i designed within VirtaAMate and ported into SD.

Lana Sin

A Australian influencer, who loves travelling and meeting new people.

(All characters i upload are fictional)

Enjoy :)

Hello! This is my first TI, an attempt to teach the diffuser how to generate the [two?] face-framing tendrils hair style.

This is an early release that may not be followed up upon, depending on popularity/usage/issues.

Keyword for each individual file is the "Trigger word" in the file details on the upper right hand part of this screen.

V07 info: "Photo" version is a more trained version of the "Illustration" inversion, and has a heavier application strength. If the photo version is used for illustration/anime-style checkpoints, it may start to override racial identity and facial features without further prompting.

V07 Known issues:

• May require a higher-strength version for higher application rate in models like Deliberate and Realistic.

• Mediumish-strength bias towards a neutral gray background (or a majority-gray/brown/boring background even when prompted). I can't seem to drop this bias yet, sorry.

• If it applies too strongly, it can start to affect facial details, and will definitely start to weight the hair color towards black or red. Both of these should (should..) be fairly easy to override without much extra prompting.

If you have any issues or feedback, please comment below!

Model was trained on a fairly-consistent Irish face with black hair (with attempts to de-bias the final result against both of those features afterwards) on a neutral gray background from a variety of zoom and angle shots, with varying amounts of hair contained in the "tendrils" (anywhere from a few strands to straight-up curtain bangs).

Huge thank you to BelieveDiffusion's textual inversion training tutorial located at https://github.com/BelieveDiffusion/tutorials/blob/main/consistent_character_embedding/README.md!

Thanks!

Sarah Snook is an Australian actress, mainly known for her role as Shiv Roy in the HBO series Succession.

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

A female character i designed within VirtaAMate and ported into SD.

Jamie Cole

A New Zealand Instagram Model, who loves food and animals?

(All characters i upload are fictional)

Enjoy :)

A female character i designed within VirtaAMate and ported into SD.

Jenna Cole

A New Zealand business lawyer, who has a step sister named Jamie Cole!

(All characters i upload are fictional)

Enjoy :)

A request of @Sonet.

Lindsey Pelas is an American model, actress and social media celebrity.

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. :)

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

Julia 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.

Julia has light skin, thick eyebrows, blue/grey eyes, and brown/red brown 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-julia-sd15.pt, and then include the trigger word nbdy-julia-sd15 in your prompt.

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

Jasmine Webb

Jasmine is an American adult film actress

Do you like my work? Buy me a beer!

[ SDv1.4, 5e-03, 3800, v/s .204 ]

Brittany White

Brittany White is an American adult film actress

Do you like my work? Buy me a beer!

[ SDv1.4, 5e-03, 3600, v/s .201 ]

A request of @DCl.

Jameela Jamil is a British actress and tv/radio host.

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. :)

Character Embedding of me, intended to be used in a comic project idea.

Jessica Trunk is a gamer and television personality. She got her start in the industry hosting gaming programming for IGN and Microsoft covering video games as well as attending many industry events. She currently can be seen on various programs on the discovery network of shows, including a show called Expedition X.

Aside from all that, it should be obvious that she's a very attractive lady. She looks just as good now as she always has, but I did choose the dataset based on much of the time earlier in her career (mostly only noticeable if you pay attention to the hairstyle).

There's probably going to be exactly 1, maybe 2 other people that understand the reference, but for diehard Chobot fans, I gave it a shot at recreatingthepicture. You know, the one that basically kicked off her entire career in the industry.

Tested with a good number of models with decent results. Mostly realistic style models. Not too many anime models, but a couple artistic oreinted ones though such as Dreamscaper, Galaxy Time Machine, and DucHaiten's models.

Trained up to 6000 steps on the base 1.5 model so it should work with most 1.5 models.

Some of the prompts in my images might have a different name than the actual filename, they were images I liked when I was running XYZ plot tests that I liked the result, all images have the same training strength as the 1.0 version of this TI.

I have made a decent number of Textual Inversions for personal use for projects, but not many that I have actually released to the public, so hopefully this is up to the mark for people's standards.

Huge shoutout to JernauGurgeh for being such a good soul and being not only helpful, but enthusiastic and a great member of the community. You should check out their work, they're one of the people I respect and look to most as a member of the community when it comes to creating good TIs and sharing good info.

rkt

Enju Aihara trained on 11400 steps

some models works fine, somes not

use such as words black framed glasses , small beard may make it more similar

Jade Aspen

Jade is an American adult film actress

Do you like my work? Buy me a beer!

[ SDv1.4, 5e-03, 3400, v/s .201 ]

Harley Dean

Harley Dean is an American adult film actress

Do you like my work? Buy me a beer!

[ SDv1.4, 5e-03, 3400, v/s .198 ]

original description

What are these

These are the textual inversion you'll often find on the posts of my models and LoRAs.

They are nothing special, but I've been requested multiple times to post them, so here we are :)

These have been tested a lot with my custom checkpoints:

• https://civitai.com/models/24149

• https://civitai.com/models/26332

• https://civitai.com/models/28322

• https://civitai.com/models/32022

The Inversions

Right now there are two inversions:

• Hotify: changes a girl appearence to make it "sexier"

• Prettify: adds makeup and change the perspective to a closeup

Guides

Learn how to create pictures like mine with my step-by-step tutorials:

• Beginner's Guide

• Prompting

• My Workflow

• Inpainting

If you want to create images with a high LoRA weight, I highly suggest you to check the tutorial.

Support

If you want to support my work, you can check out my music here:

https://inzaniak.bandcamp.com/

You can also check my AI Art accounts, where you'll find daily pictures made with my models (and previews of new ones coming):

• Instagram

• DeviantArt

What are these

These are the textual inversion you'll often find on the posts of my models and LoRAs.

They are nothing special, but I've been requested multiple times to post them, so here we are :)

These have been tested a lot with my custom checkpoints:

• https://civitai.com/models/24149

• https://civitai.com/models/26332

• https://civitai.com/models/28322

• https://civitai.com/models/32022

The Inversions

Right now there are two inversions:

• Hotify: changes a girl appearence to make it "sexier"

• Prettify: adds makeup and change the perspective to a closeup

Guides

Learn how to create pictures like mine with my step-by-step tutorials:

• Beginner's Guide

• Prompting

• My Workflow

• Inpainting

If you want to create images with a high LoRA weight, I highly suggest you to check the tutorial.

Support

If you want to support my work, you can check out my music here:

https://inzaniak.bandcamp.com/

You can also check my AI Art accounts, where you'll find daily pictures made with my models (and previews of new ones coming):

• Instagram

• DeviantArt

Introduction (Chinese)

Basic Information

This page is recommended for all text embeddings (Embedding) for the AnimeIllustDiffusion [1] model. You can view the information embedded in this text from the release description.

Instructions

You should place the downloaded negative text embedding file into the embeddings folder of your stable diffusion directory. After that, you just need to enter badv4 in the negative prompt word field.

Evaluation parameters

Each text embedding will describe its characteristics from the following parameters:

Positive/Negative Text Embedding：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。

Scope of action：一个文本嵌入主要影响的画面内容。

vector number：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。

vector strength：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.

Quote / References

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

Sommer Ray is an American fitness model and social media celebrity with more than 25M followers on Instagram.

1850-step TI trained on a dataset of 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. :)

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

Stephanie 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!

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 Schulman - d4p

original description

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:

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_v3Lightand therealisticVisionV20_v20models.

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 ]

Mother 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 Maxwellis 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 garcia 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:
+ 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.)

, 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_v3Lightmodel.

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_v3Lightmodel.

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_v3Lightmodel.

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 assetscan 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 assetsmade 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,nowwould we?).

• I've found that including open mouth and teethin 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,andthe 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!

Note: The new religious image I uploaded uses a lot of tricks. Just showing what it's possible to do. Happy Easter!

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 adult film star 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;

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_v20model

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.

This combination of ti+lora was my work a long time ago, and it still seems to work well. Of course, you can also only use my ti, but use other people's 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

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

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

Why ti+lora?

Using ti and lora together can work better on photo-based models, and you don't have to write a lot of tags.

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

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), , (ulzzang-6500:0.3)

normal:

1girl,(weihua:1.0),

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 ]

Sky 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 ]

Riley 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 ]