Ciara Bravo, actress (if you’ve never seen the show ‘Wayne’, I highly suggest it.)
Description and prompt below!

PLEASE LEAVE A REVIEW! I'm still tweaking my process and it helps seeing how my TI's turn out in other people's results.

Using my own model in the examples here, still fine-tuning it a touch but will be releasing that later this week!
Also, I’m a HUGE FAN of Leosam's FilmGirl lora, although sometimes it fights my TI’s (believe it’s heavily trained on an asian dataset), it does a phenomenal job at adding a realistic film look. Check it out if you haven’t!
https://civitai.com/models/33208/filmgirl-film-grain-lora

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

I gotta mention that Ciara in real life looks much younger than she is, so the TI tends to lean that direction. Takes a decent amount of prompting to get away from that.

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_ciara>, fashion photography portrait of a pretty woman, sitting, (sweaty), mature body, flirty, 26 years old, in a lush rainforest, large exotic flowers, smiling, offshoulder, fluffy short hair in a (high ponytail), soft light, rim light, beautiful shadow, low key, (photorealistic, raw photo:1.2), (natural skin texture, realistic eye and face details), hyperrealism, ultra high res, 4K, Best quality, masterpiece, necklace, cleavage, in the dark

Negative -
Nude, nsfw, unclad, (Child, childlike), hat, headwear, cartoon, anime, sketches, (worst quality:1.5), (low quality:1,.5), lowres, ((monochrome)), (deformed, distorted, disfigured:1.3), poorly drawn,large hands, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, (mutated hands and fingers:1.4), disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation, <bad_prompt>




Trained at 1000 steps on 20 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 an experiment to see if I can make a TI embedding that gets the flavour of konyconi’s BohoAI LORA.

https://civitai.com/models/51966/bohoai

Thank you to @konyconi for sharing his dataset for the excellent BohoAI LORA.

https://civitai.com/models/52697/tutorial-konyconi-style-lora

The showcase uses 2 models:

revAnimated_v122.safetensors [4199bcdd14] with clip skip = 2

avalonTruvision_v2.safetensors [a4df55d292] with clip skip = 1

This TI can produce some decent Boho pix but it gets confused sometimes... eg asking for a spaceship and getting a truck. Perhaps for this sort of TI you need to use a lot more pictures in the training dataset, with more subject variation?

---------------------------

I’ve been struggling to work out how to make a style TI...

what makes a good training dataset?

what training settings should I use in automatic1111?

how long to cook the TI for?

For my training dataset I copied konyconi’s 76 1024x1024 images to a new folder without the associated TXT files, and reduced them all to 512x512. Then I renamed them “01 aeroplane.png”, “02 city.png”, “03 tank.png” etc.

Why? Because I was trying to match what I’ve done in the past for TIs that ended up usable. The reduced images dataset folder is what I used in the settings below.

The wikipage for automatic1111 Textual Inversion is here:

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Textual-Inversion

but way out of date. Last rev Jan 5, I’m writing this May 8.

I found this thread useful in parts. It’s a long read!

https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/1528

Training model: v1-5-pruned.ckpt [e1441589a6]

I used this because I don’t know any better, and it’s been useful in the past. Should I be using a different model for training, or is the base SD15 the best thing to use? No idea.

Create embedding:

name: kcboho07

initialization text: boho style photo

number of vectors per token: 4

Train embedding:

Embedding name: kcboho07

Embedding Learning Rate: 0.001:250, 0.0005:500, 0.00075:1000, 0.001

Gradient Clipping: disabled

Batch size: 1

Dataset directory: wherever you’ve put it on your computer

Log directory: textual_inversion

Prompt template: minimum_style_2.txt

The template has 3 lines:

<<<

[name] style, [filewords]

[name] style, a photo of [filewords]

[name] style, an illustration of [filewords]

>>>

Width = Height = 512

Do not resize images: OFF

Max steps: 3000

Save image steps: 25

Save embedding steps: 25

Use PNG alpha channel: OFF

Save images with embedding in PNG chunks: ON

Read parameters from txt2img tab: OFF

Shuffle tags: OFF

Drop out tags: 0

Latent sampling method: deterministic

Training time: about 50mins per 1000 steps on a 2060/6GB.

The TI at 3000 steps does produce a Boho style, although I think it’s a bit hit-and-miss compared to the BohoAI LORA.

If anyone has suggestions about what I should be doing differently please add a comment. Or if I’m doing anything obviously stupid! :-)

RIP

Alizée Lyonnet (née Jacotey) is a French singer who rose into stardom in the early 2000s.

Update: The TI has been updated after a 2-star review complaining about its oversaturation. Granted, version 1.0 was indeed oversaturated, but that was easily fixable by simply reducing the weight of the TI a little bit. Did it deserve a 2-star review just for that, basically ignoring any other merits? I very much doubt so, but to each their own, I guess. It's pushed me to create a new version of this TI, which was one of my first. It probably was about time anyway.

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

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

How to create a good prompt using my TIs

You're obviously free to experiment, but bear in mind that my TIs are trained with a more or less fixed phrasing, that normally starts with:

"photo of EMBEDDING_NAME, a woman"

So I recommend always starting your prompt like that and then building the rest of the prompt from there. For instance, "photo of (alizjacot:0.99), a woman, RAW, close portrait photo, long brown coat, turtleneck sweater, pale skin, slim body, (high detailed skin:1.2), 8k uhd, dslr, soft lighting, high quality, film grain, Fujifilm XT3 sharp focus, f 5.6".

Jelena Noura "Gigi" Hadid (born April 23, 1995) is an American model and television personality. In November 2014, she made her debut in the Top 50 Models ranking at Models.com. In 2016, she was named International Model of the Year by the British Fashion Council. Over the span of four years, she has made 35 appearances on international Vogue magazine covers.

If you like my work and you feel like it, you can invite me to a Ko-fi!

Textual Inversion of Daily Wire host Brett Cooper.

GOKU, muscles, without shirt, in the space, yellow hair, orange pants

Ashley Cipher is a person who doesn't exist. You can use her if you need a consistent character.

Please have a look at other Nobody models for contributions by other folks including those created with the LastName and NBDY tags.

Follow me if you'd like to be notified of future models!

For luscious, obviously fake lips.

Face Restoration may turn the fake lips into a mustache/"monkey lips".

To avoid this, you can change/lower/disable the face restoration setting or first generate the model then paint over the lips with face restoration disabled.

Or generate first without face restoration, then with face restoration then combine in photoshop/gimp.

Used in examples: SD1.4 / UberRealisticPornMerge / Clarity19

Asa Akira

Asa is an American adult film legend. LEGEND.

This is a commission from a user who wanted it created with a very specific set of source photographs. What can I say, I aim to please.

This is a textual inversion, if you want a LoRA of Asa look HERE

You can learn more about my process in my blog post HERE

I can create a custom Textual Inversion or LoRA just for you!

Do you like my work? Buy me a beer!

[ SDv1.4, 4e-03, 5100, v/s .201 ]

Miss Alice

Miss Alice is a Kiwi webcam and OnlyFans legend. LEGEND

This is a commission from a user who wanted it created with a very specific set of source photographs. What can I say, I aim to please.

This is a textual inversion, if you want a LoRA of Alice look HERE

You can learn more about my process in my blog post HERE

I can create a custom Textual Inversion or LoRA just for you!

Do you like my work? Buy me a beer!

[ SDv1.4, 4e-03, 5100, v/s .201 ]

Embedding of Canadian-Swedish actress Malin Akerman. You might know her from a variety of films like Rampage or Watchmen, but also series like Billions. Turned out pretty well in terms of likeness.

DPM++ 2M KARRAS 30 steps

7 CFG

Upscale through highres fix or Controlnet 1.1

Madison Nicole Ziegler is an American actress and dancer.

CyberRealistic v2.1 [BETA2] is used for the preview images.

Ela Rumeysa Cebeci is a Turkish news anchor and can also be found on Instagram.

She has over 1 million followers.

Still learning and improving the progress of the training, any feedback is welcome.

Feel free to leave me a tip if you like what I am doing :)

Yvonne Craig, aka Barbara Gordon/Batgirl from the 1960s Batman TV series

Yvonne Craig is best known as "Batgirl," but also played opposite Elvis Presley in a number of movies, and made numerous appearances in 60s and 70s TV shows, including roles in Star Trek, The Wild, Wild West and The Man from UNCLE.

Like with Erin Gray, this is draft version of the TI -- I will roll out a v1.1, etc. if I can source some better images. Nevertheless, as you can see, this TI can produce a pretty good likeness though it might take a few re-throws.

Let me know what you think!

This embed fixes some problems of 2FSV and makes it's style more consistent.

Intended for clip skip 1 and CFG scale below 10.

Alternative version aka "nncursedV0a" will burn your images to ashes on CFG 8+, so I can't recommend it, but base version of this embedding will eat all of non-generic styles and effects, like pixel-art for example, so for now I keep this alternative.

This piece trained and tested solely on 2fsv

I have no idea how it will work with other models. I can guess that it could work okay on models with 2.5D style.

A commission of @Zorglub.

Lesley-Ann Brandt is a South African actress, known for her roles in TV shows such as Spartacus and Lucifer.

This is a 1000-step TI trained on a dataset of 18 images with my usual settings.

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

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

How to create a good prompt using my TIs

You're obviously free to experiment, but bear in mind that my TIs are trained with a more or less fixed phrasing, that normally starts with:

"photo of EMBEDDING_NAME, a woman"

So I recommend always starting your prompt like that and then building the rest of the prompt from there. For instance, "photo of (lesanbrandt:0.99), a woman, RAW, close portrait photo, bodysuit with zipper, cleavage, fishnet stockings, pale skin, slim body, (high detailed skin:1.2), 8k uhd, dslr, soft lighting, high quality, film grain, Fujifilm XT3 sharp focus, f 5.6".

This is Negative Embedding. The concept was to improve quality, such as EasyNegative and veryBadImageNegative.

Please use it in the "\stable-diffusion-webui\embeddings" folder. Please use this embedding with trigger words of each version (ex.an5) in the NEGATIVE prompt form.

*It can be used with other models, but the effectiveness is not certain.

Veronika Fasterova AKA Verunka. Czech nude model.