🧠Model Description
LoRA Name: TATIANA
Base Model: KREA 2
Type: Character LoRA
Age: 27
Gender: FEMALE
This LoRA is designed to consistently and accurately recreate the character TATIANA, including facial features, body type, visual style, typical appearance, and micro-expressions.
Sometimes a small mole or spot may appear on the face because I wanted to experiment with a retouched character. I'm not sure if it will be released or how. We're here to experiment!
⚙️ Recommended Parameters (ComfyUI)
Steps: 10-12
Sampler: Euler ANCESTRAL
Prompt Format: simple
Recommended settings:
LoRA weight: 0.9 – 1.1
Aspect Ratio: 4:3 , 5:8 ,9:16 (PORTRAIT)
Suggested terms for prompt: T4T14N4 ,and the rest of the prompt
This model was trained using an image that I had discarded but then retrieved. My search continues for unique faces that stand out from the glossy, perfect standards typical of AI. Any suggestions or advice are welcome!
If you like my Lora and want to support me, you can consider sending me buzzes, or even better, you can sign up for Runpod using my link. Runpod referrals: https://runpod.io?ref=72iobbrk
You would be helping me greatly to continue this work and make the new models available to everyone.
Description
FAQ
Comments (3)
Another really good model - thanks for sharing!
Similarity to the Rose LORA from yesterday, but a bit younger and more attactive overall. Also has a better and more adaptive hair/style/color.
I also noticed how crisp these two models rendered with my checkpoint, so i think i'll check out your other chara LORAs later tonight. Keep up the good work.
These are all faces that I process using old Gan networks like ArtBreeder, and then I run them through Flux Klein to create the dataset. Not all of them turn out the same way, though. I noticed that you’re using another checkpoint based on Krea 2. What’s special about it? Is it less censored?
@rolenzio Sick Ollie INT8 is what i'm using - It's very NSFW friendly, but also produces much better images in general. I never really bothered checking out the base checkpoints, so it's hard to say how each image would look otherwise. It's certainly a far cry from the usual Turbo images which tend to look either grainy or blurry with muted colors, and homogenize everything (in particular the character LORAs).
The author also seems to be a great and helpful guy (Sickollie), doesn't spam "alternate" versions just to leech buzz and clicks.