🤎🍑🤎🍑🤎🍑🤎🍑🤎🍑🤎🍑🤎🍑🤎🍑🤎🍑🤎🍑🤎🍑🤎🍑🤎🍑🤎🍑

First of all I would like to start off by saying this is a experiment in tricking Stable Diffusion. I have longed for mixed race bodies that more resemble what I am fond of, bodies like my significant other. SD didn't seem to do it right, and was had to control to get the effect I wanted.

I also didn't want to release a model that would reproduce identifiable faces for privacy and respect, and just wanted the artistic form of the generalized body type.

To defeat this I put together an idea to experiment with:

1. Gather images that represent the dataset I wanted

2. Specifically crop images so that only the body is in the frame.

3. Upscale the images with 4x-UltraSharp.pth, downsize to 512x512
   3.a For images that the face could not be cropped out, I masked and filled with latent noise (using ComfyUI and my WAS Node Suite)

4. Train with prompts that focus on the body type I am going for
   4.a Add specific emphasis on faces, which don't exist, again in the style I am going for
   

The idea here is that while training, and using my prompts, Stable Diffusion is going to be force to make up unique faces on it's own. To me the frequency of cut-off faces seems very low, and so far, my theory seems to have panned out.

🤎🍑🤎🍑🤎🍑🤎🍑🤎🍑🤎🍑🤎🍑🤎🍑🤎🍑🤎🍑🤎🍑🤎🍑🤎🍑🤎🍑

Notes:

• Best used with VAE vae-ft-mse-840000

• Best used with good ol' negative embeds for composition, and bad hands

• Best used with models like Dreamlike, Protogen, TheAlly's Mix's, etc.

• Due to the nature of the TI, if you want clothed models, you'll have to play with weighting to taste.

Gavid first found fame as the hard rocking front-man of seminal 80's LA rock band 'Gans & Posers'. Then he met and married model Stefani Mourford, and things got a little weird. First they decided to name their daughter "Grapefruit", then they announced to the world they had decided to "consciously uncouple", and finally Gavid dropped the bombshell news he was leaving the band. Since then he's reinvented himself as a successful (and ever-changing) solo artist. Now he's ready to feature in your SD creations. As a minor side-note ... Gavid doesn't exist, and is an AI OC, a character not based on any living person, in personality or looks. Gavid may LOOK like somebody you know, but he's a Nobody.

He's happy to be your hard-rock hero, your gnarly barbarian, or homeless bum, when you need a consistent character not based on any one living person. He works in every model I've tested (which is quite a few), likes to keep his hair long, and seems to wear a permanent scowl (so appropriate prompts will be needed to make him smile).

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

Alexa Bliss is an American professional wrestler. She is signed to WWE, where she performs on the Raw brand under the ring name Alexa Bliss. In 2013, Bliss signed a contract with WWE and was assigned to their Performance Center and developmental brand NXT. She made her main roster debut on the SmackDown brand in 2016, later becoming a two-time SmackDown Women's Champion and the first woman to hold the title twice.

Bliss transferred to the Raw brand in 2017 and would go on to become a three-time Raw Women's Champion, with her initial reign making her the first woman to win both the Raw and SmackDown Women's titles

Random model

Amy Smart

British instagram/twitter model

Geena Davis is an American actress, winner of an Oscar award, known for her roles in many notable films, including Beetlejuice and Thelma & Louise.

1000-step TI trained on a dataset of 18 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. :)

Now that I have your attention.

Trained using GrapeLikeDreamFruit

Sample images using ConsistentFactorV32

This is an embedding trained on sample images of embeddings, mostly my own and those of JernauGurgeh. I also generated new images using modified prompts from all of your embeddings. So all it takes is "art by CFStyle" to get high quality sample images. Of course, you can add more detail to your prompts for more fine tuning.

Erin Timony, aka GoodnightMoon ASMR.

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_goodnightmoon>, hyper realistic photograph, photo of a beautiful girl, full body, wearing leggings and a t-shirt, outside in LA, highly detailed, large eyes, photorealism, sharp focus, best quality, 4k, vibrant colors, backlit, rim light, (looking at viewer), shot on Canon, detailed skin,

Negative -
far away, ugly, low-res, indoors, blurry, wrinkles, 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,


Using mostly Realistic Vision v1.4 in these, but it works well in most models trained on diverse datasets. Big fan of Epi_Noiseoffset as well.

Trained at 2000 steps on 40 images. Before anyone asks, it's a .bin file because it was trained using the Google Colab, and that's what it gives you.

Victoria Pedretti, actress from The Haunting of Hill House, Bly Manor, You, etc.

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_victoria>, hyper realistic photograph, photo of a beautiful girl, long straight hair, full body, wearing leggings and a t-shirt, magazine cover, poster art, highly detailed, large eyes, photorealism, sharp focus, best quality, 4k, vibrant colors, backlit, rim light, key light, low key, (looking at viewer), shot on Canon, detailed skin, studio lighting, (solid color background), synthwave,

Negative -
far away, ugly, low-res, graphic tee, blurry, wrinkles, 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,


Using mostly Realistic Vision v1.4 in these, but it works well in most models trained on diverse datasets. Big fan of Epi_Noiseoffset as well.

Trained at 2000 steps on 40 images. Before anyone asks, it's a .bin file because it was trained using the Google Colab, and that's what it gives you.

follow me on Instagram, Youtube, Patreon and my website.

NOTE: Try to vote on the quality of the model/training an not based on how easily you are offended by the subject.

Ron DeSantis

Suggested Weight: 1.0

Suggested prompts

• photo of (d354nt15:1.0), sharp focus, intricate detail, light bokeh, hasselblad h6d-400c 1.4, natural lighting

Training

• trained on ~20 images

• Learning Rate: 0.005:1000, 0.001:2000, 0.0001:5000, 0.00005

• Total Steps: 15,000

• Batch Size: 1

• Gradient Steps: 1

This is a textual inversion of character "Erin" - just add to your "embeddings" folder and then use the command "erinOobleksd15-Final" in your prompt to call it.

This is my first upload.

Seltin Sweet is a 23 year old Russian instagram/only fans model.

介绍（中文）

它的作用很简单——让图像更清晰。它会使图像中所有模糊的区域更加清晰，如果这破坏了你原本的画风，则停止使用它。我不建议调整它的权重。

使用方法

将下载得到的 .pt 文件放入您 stable diffusion 的 embeddings 文件夹内（无需重启 webui）。在生成图像时，于负面提示词内输入你下载的 .pt 的文件名（不包括扩展名）。例如，你下载的文件名叫 'lr.pt' 则输入 'lr' 即可。

Introduction (English)

What it does is simple - it makes the image clearer. It will sharpen any blurred areas in the image, if this ruins your original style, stop using it. I don't recommend adjusting its weight.

Instructions

Put the downloaded .pt file into the embeddings folder of your stable diffusion (no need to restart the webui). When generating the image, input the filename (extension is excluded) of the .pt file you downloaded in the negative prompt. For example, if the file name you downloaded is 'lr.pt', just enter 'lr'.

Littel Caprice

Markéta Štroblová-Schlögl, known professionally as Little Caprice, is a Czech pornographic actress, model and producer.

I merged a 300 and 2000 step inversion of her, that I trained.

Shailene Woodley, actress from Big Little Lies, Divergent, Secret Life of an American Teenager, etc.

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_shailene>, hyper realistic photograph, waist up, portrait of a beautiful woman, adorable, highly detailed, photorealism, sharp focus, best quality, 4k, full body, (solar eclipse, outside, dark sky), cinematic lighting, risqué, boudoir, dark eye makeup, reflections, (looking at viewer, symmetrical), shot on Canon, detailed skin, sharp focus, rim lighting, two tone lighting, dimly lit, low key

Negative -
far away, ugly, low-res, blurry, 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, extra nipples, far away shot, monochrome,


Using mostly Realistic Vision v1.4 in these, but it works well in most models trained on diverse datasets. Big fan of Epi_Noiseoffset as well.

Trained at 2009 steps on 41 images. Before anyone asks, it's a .bin file because it was trained using the Google Colab, and that's what it gives you.

This is my sixth upload, please, for the love of god, leave a review or something.

Rowan Blanchard, 21 year old actress

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 + 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_rowan>, hyper realistic photograph, portrait of a woman, waist up, highly detailed, photorealism, sharp focus, best quality, 4k, neon colors, backlit, rim light, synthwave, (looking at viewer), 80s inspired, shot on Canon, detailed skin, sharp focus, shallow depth

Negative -
far away, ugly, low-res, blurry, bad anatomy, anime, cartoon, 3d render, illustration, disfigured, child, childlike, 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


Using mostly Realistic Vision in these, but it works well in most models trained on diverse datasets.

Trained at 2009 steps on 41 images. Before anyone asks, it's a .bin file because it was trained using the Google Colab, and that's what it gives you.

This is my third upload, please, for the love of god, leave a review or something.