Sign In

Bogna Harkusha - Natural Character LoRA

0

Not recently loaded

Download

1 variant available

bf16 SafeTensor

bognaharkusha_v1.safetensors

BF16, good balance • 138.43 MB

Verified:

Type
LoRA
Stats

19

0

Reviews

No reviews yet

Published

Nov 17, 2026

Base Model
Training
Steps: 3,000Epochs: 1
Usage Tips
Strength: 1
Hash
AutoV2E2EDB88A81
Trigger Words
b0gn4h4rkush4
1794915759760_0.png

My second character LoRA - meet Bogna Harkusha!

Bogna is entirely fictional female character in her early 20s with short brunette bangs and eastern-european look to her. As with my previous LoRA - being tired of seeing same faces with LoRAs, i aimed to produce another character that will at least try to stand out a little from perfect AI-proportions. For now I trained Bogna for MOSS, although i could also do other models in some time. I look forward for all feedback, both positive and negative, as i'm not 100% happy with this project, which i describe a little further below.

Keyword: b0gn4h4rkush4

Description and other yapping:
I trained Bogna with a high-quality dataset (mostly 1792x2400 pictures). I tried to include both studio shots and amateur-style pictures (in terms of technical aspects) so she can be used in various scenarios and prompts. Overall, this LoRA can produce both 3/4 and 1/1 ratio generations, both amateur and more "professional photo" styles.

By default, Bogna is rather natural looking - no make-up should be included, unless you prompt it.

What Bogna can handle: various types of outfits, angles, poses, mimics and emotions, lighting.

What may not be all-time consisent: sadly, facial structure with some prompts - this LoRA gave me much more trouble when training, despite trying several times with different parameters. I'm still learning to train LoRAs, Bogna is a project I'm not 100% satisified with in terms of accuracy to her reference look and consistency. In some prompts and angles Bogna can look a little different on her face. I plan on making v2 version of her, when i learn more about preparing good dataset and training parameters.

Despite her flaws, I encourage to try Bogna, post your generations and give feedback - It'll allow me to improve future models.