ℹ️ Babes 1.2i12 Illustrious marks an exciting evolution in the series, transitioning from its Pony-based foundations to the Illustrious architecture. This shift brings a refined, luminous quality that feels both fresh and deeply intentional - a harmonious blend of the signature Babes aesthetic with the elegant strengths of Illustrious. The result is a style that retains its characteristic glamour and polish while gaining new levels of clarity, harmony, and expressive finesse.

Visually, Babes 1.2i12 delivers radiant, high-fidelity portraits with luminous skin tones, softly sculpted features, and a gentle yet captivating presence. Colors bloom with rich, balanced saturation - jewel-like hues in hair and attire dance alongside natural, flattering lighting that creates depth without harshness. Linework is crisp yet graceful, backgrounds gain subtle storytelling detail, and overall compositions feel more cinematic and cohesive. Faces carry a refined, almost ethereal charm: expressive eyes, harmonious proportions, and an inviting warmth that draws the viewer in.

This version captures the essence of timeless elegance reimagined through modern digital artistry - glamorous, vibrant, and effortlessly stylish, yet always tasteful and sophisticated. Whether crafting fashion-forward concepts, fantasy-inspired scenes, or serene character studies, Babes 1.2i12 offers a versatile, premium aesthetic that feels both luxurious and alive.

ℹ️ Babes 1.2p11 Pony - This version introduces several refinements. A significantly expanded color range brings richer saturation, cleaner white balance, and more accurate, natural hues across a broader spectrum of shades. Shapes and linework appear sharper and more defined, enhancing clarity without sacrificing softness. Backgrounds receive greater attention, with increased depth and fine detail that elevate overall composition. The semi-realistic style is now more consistent throughout, and facial features lean subtly cuter - balanced, expressive, and more harmonized with the overall aesthetic.

How it works?

Choose Product or Subscribe:
- Get this model via a Ko-fi product, or
- Join the appropriate membership tier:
  - Model released during current month: Current Month Content
  - If model released in previous months: All Months Content
You will be prompted to link your Hugging Face and Ko-fi emails.
Hugging Face access is checked by email, which might be different from your Ko-fi email. You will receive a link from Ko-fi to add the email and continue.
Request Access on Hugging Face:
After you acquired product or subscription, you’ll receive a link to a gated Hugging Face repository.
- Click "Request Access".
- Your access will be approved automatically within a minute.
Download & Use:
You can then download and use the model under the license provided in the repository (LICENSE.md).

Why I Created My Own Access Control

I have been a creator on Civitai since December 2022. My goal has always been to help make model training, design, and development a sustainable, professional activity, not only for myself, but for other creators too. Supporting people who want to create models professionally is good for users, for creators, for the entire community, including Civitai itself.

Civitai has shared some of this vision and has tried to create ways to support creators. Unfortunately, their current Creators Program has serious problems:

Civitai was deplatformed by payment processors and operates at a loss.
The current compensation model pays creators mainly for model usage in the generator.
Early Access limits paid access to only 15 days, after which models must be free.
Buzz extraction value has dropped to ~45% of its original worth, making withdrawals basically a 55% fee.
Bugs and creator feedback are ignored for months, and several team members who used to listen have left.

I believe this system is unfair, harmful, and exploitative. Creators should be able to:

Sell models without arbitrary time limits.
Pay a fair platform fee of 15% or less.

Civitai briefly tested a much better system called Clubs in early 2024, which worked well for creators, but it was canceled after backlash from a loud minority that opposed fair creator compensation. I have long hoped Civitai would fix these issues, as I’ve suggested in my articles:

But nothing has changed. Bugs remain unresolved, feedback is ignored, and I’ve lost optimism about Civitai’s willingness to improve.

That’s why I’ve decided to experiment with my own alternative access control system, a fairer way to share models with users who value my work.

ℹ️ Babes 1.2p10 Pony - This version introduces a few interesting changes, continuing my effort to refine and evolve Babes with every version. In this version, I focused on adding a few hand gestures, such as: hand on chest, hand v sign, hand open palm, hand thumb up, hand pointing up, and both hands on waist. These gestures are significantly better now, but they are definitely far from perfect.
I also tried to improve the teeth, aiming for more defined, clear details and more consistent results, and I think that effort was successful.
This version includes a few adjustments to the semi-realistic base style and default faces. My goal was to make the model feel more rounded and distinct, while staying aligned with the established look of earlier versions. I feel that the fidelity remains very high, and I’m pleased with the direction the model is heading.
As always, each release is an opportunity for me to add creative flavor to the model. I’ve continued to refine the model and hope to be able to keep making frequent updates, while Civitai is going through some worrying changes.
I really love what the model can do now, and I hope that people will enjoy this latest addition.
ℹ️ Renamed versions: 1.2-1.29 Pony versions changed to 1.2p00-1.2p09 or 1.2p0-1.2p9.
ℹ️ Babes 1.2p09 Pony – A new training and a few new components added, aiming to polish and improve semi-realistic style further, adjustment to faces and skin, a little bit more details, enhancement of sharpness and contrast.
ℹ️ Babes 1.2p08 Pony – Incorporated a few new components and new training, sharpened backgrounds, added more details, slicker 3D effects, expanded tonal range, and lowered close-up bias. This version can create truly incredible images—it's my pleasure and privilege to be able to create and publish it.
ℹ️ Babes 1.2p07 Pony – This version comes with enhanced semi-realistic base style, adjustment to default faces, and improvement to saturation and lighting.
ℹ️ Babes 1.2p06 Pony – This version brings significant improvements to details and sharpness, along with major enhancements to compositions and backgrounds. For a more alternative look, I recommend using "tattoos" in the prompt. I have to admit, though, that hands are more challenging in this version.
⚠️ This model has been on Civitai’s generator since the first day of generator existence. However, after the recent attempt to support all checkpoints in the generator did not work as expected, Civitai has decided to stop supporting this model in the generator. I’m disappointed by this decision—this is my first and most beloved model.
Ultimately, during the period when all models were briefly available in the generator, neither of the Babes versions met some required threshold. Maybe this happened because the latest version was in Early Access, splitting users between two versions—I don’t know.
I checked the data from this model for exact numbers. I also found at least four models that are currently in the generator, which have a lower number of images generated with them than versions of this model (v1.2p04, v1.2p05) - despite being measured over a shorter period of time. So unfortunately, the decision by Civitai is not based on merit, unlike how it was presented.
Civitai promised that the new Creators Program would be more fair, but after reverting changes to support all checkpoints, we now see that two creators hold around 20 checkpoints in the generator—meaning just two people have 10% of all available slots—including a single model with six versions. Other models have three or even four versions in the generator. This situation is an unfair for other creators and not helpful to users.
ℹ️ Babes 1.2p05 Pony - This version looks amazing. It creates gorgeous faces and bodies, and overall, I like it a lot. Some of the changes include: less shiny skin, rebalanced contrast with renormalized colors, better proportions, and more realistic lighting.
ℹ️ Babes 1.2p04 Pony - Additional polishing of base style, look and feel. Adjustment to more volumetric and less flat shading, slightly more metallic textures, can produce more darker images, has a wider variety of colors - particularly for clothes and hair.
🍊 I want to thank all the people who are using and enjoying this model, all my other models, and supporting my work. As a token of appreciation to all the lovely people, I decided to discount Babes 1.2p03 Early Access bigly and boldly. Hope y’all are having a great 2025. Let’s Make AI Great Again!
ℹ️ Babes 1.2p03 Pony - Adjustment to base style, default faces, more vivid colors and more colorful images. I love the results, somehow I keep surprising myself all the time. 😆
ℹ️ Babes 1.2p02 Pony - Incorporating new components, including training from Galena 1.3 and Babes Kissable Lips 3.7, and additional new finetuning. The base style is a continuation of Babes 1.2p00 and 1.2p01 with a slight variation. The results look amazing in my humble opinion. 😊
ℹ️ Babes 1.2p01 Pony - Changes to the base style of Babes 1.2p00, intended to offer additional complementary options.
ℹ️ Babes 1.2p0 Pony - Finetuning based on Babes Kissable Lips 3.4.
ℹ️ Babes 3.1 includes new training, a new recipe, rebalanced styles, and enhancements to the base style.
ℹ️ Babes 3.0 was trained with 43,000 images, including many new styles and subjects.
ℹ️ Styles: Babes 1.1 - "basety style", Babes 2 - "abbe bi style", Sassy Girls - "sassy style", Midjourney - "midjourney style".
ℹ️ Cartoon styles:

othalama style, ronidu style, seviechan style, samdoesart style, thepit style, owler style, cherrmous style, arosen style, uodenim style, stanleylau style, amime style

ℹ️ Recommended tags in the negative prompt: nose stud, piercing
ℹ️ Misc trigger words: wild nature, suicidegirl, interior design, digital art

📌 Are your results not 100% identical to any specific picture?

Make sure to use Hires-fix from example SwinIR_4x / 4x-UltraSharp / 4x-AnimeSharp / RealESRGAN_x4plus_anime_6B (Upscaler Download), it is what I usually use for hires-fix.
SD 1.5 - Use VAE: vae-ft-mse-840000-ema-pruned for better colors. Download it into "stable-diffusion-webui/models/VAE" folder. Select it in the settings.
I use xformers - it's a small performance improvement that might change the results. It is not a must to have and can be hard to install. Can be enabled with a command argument "--xformers" when launching WebUI.
WebUI is updated constantly with some changes that influence image generation. Many times technological progress is prioritized over backward compatibility.
Hardware differences may influence changes. I've heard that a bunch of people tested the same prompt with the same settings, and the results weren't identical.
I have seen on my own system, that when running as part of a batch, may change a little bit the results.
I suspect there are hidden variables inside modules we can't change that produce slightly different results due to internal state changes.
Any change in image dimension, steps, sampler, prompt, and many other things, can cause small or huge differences in results.

📌 Do you really want to get the exact result from the image? There are a few things that you can do, and possibly get even better results.

Make a single word changes to prompt/negative prompt and test, and push it slowly to your desired direction.
If the image has too much of something or doesn't have enough of something, try to use emphasis. For example, too glossy? use "(glossy:0.8)", or less, or remove it from the prompt, or add it to the negative. Want more, use values 1.1-1.4, then additional descriptors in the same direction.
Use variations - use the same seed, and to the right of the seed check "Extra". Set "Variation strength" to a low value of 0.05, generate a few images, and watch how big the changes are. Increase if you want more changes, and reduce if you want fewer changes. That way you can generate a huge amount of images that are very similar to the original, but some of them will be even better.

📌 Recommendations to improve your results:

SD 1.5 - Use VAE for better colors and details. You can use VAE that comes with the model or download "vae-ft-mse-840000-ema-pruned from (https://huggingface.co/stabilityai/sd-vae-ft-mse-original/tree/main) , ckpt or safetensors file into "stable-diffusion-webui/models/VAE" folder. In the settings find "SD VAE", refresh it, and select "vae-ft-mse-840000-ema-pruned"(or the version included with the model). Click "Apply settings" button on the top. The VAE that comes with the model is "vae-ft-mse-840000-ema-pruned", you don't need both, use the one that you downloaded, it will work very well with most of the other models too.
Use hires-fix, SwinIR_4x / 4x-UltraSharp / 4x-AnimeSharp / RealESRGAN_x4plus_anime_6B (Upscaler Download), first pass around 512x512, second above 960x960, and keep the ratio between the two passes the same if possible.
Use negatives, but not too much. Add them when you see something you don't like.
Use CFG 7.5 or lower, with heavy prompts, that use many emphases and are long, you can go as low as 3.5. And generally try to minimize the usage of emphasis, you can just put the more important things at the begging of the prompt. If everything is important, just don't use emphasis at all.
Make changes cautiously, changes made at the beginning of the prompt have more influence. So every concept can throw your results drastically.
Read and use the manual (https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features).
Learn from others, copy prompts from images that look good, and play with them.
DPM++ 2M Karras is the sampler of choice for many people, including me. 40 steps are plenty, and I usually use 20.
Discord server for help, sharing, show-offs, experiments, and challenges

Files with SHA256: 4c954069c75045239b7d2a24db3056cbe80f81d90390a50590e15dc9c8b73c5c