Justice Dress — MVD Uniform LoRA

I'm creating a LoRA for the first time as an experiment, since I've never done this before.
A Krea 2 LoRA trained to reproduce the formal uniform dress worn by female students of a Russian internal affairs / MVD educational institution where I study. This LoRA is designed to reproduce a distinctive dark blue uniform dress worn by female students at certain Russian internal affairs universities. The uniform includes designated areas for shoulder boards and various insignia patches, such as those identifying the student's faculty or university. This type of uniform may also be seen on female personnel of various Russian agencies and services, such as the Ministry of Internal Affairs (MVD), the Investigative Committee of the Russian Federation, the police, and airport services but they differ in color - ranging from dark blue and black to dark green (like those worn by Customs workers in Russia).

The LoRA was trained on 39 reference images using Krea 2 Raw and AI Toolkit. It is designed to reproduce the distinctive appearance and structure of the uniform while allowing it to be used in different characters, poses, environments, lighting conditions and photographic styles.

Sometimes, low-quality Russian flag patches and shoulder boards with unreadable text may appear in the image; I’ll touch those up later.

Trigger words: Russian uniform of the internal affairs agencies, MVD uniform, blue scarf around neck

These descriptions can be incorporated naturally into your prompt rather than using a separate short activation token.

Recommended usage

Describe the character and scene normally, then include one of the trigger phrases.

Example:

a young woman standing in a university hallway, wearing Russian uniform of the internal affairs agencies, full body, realistic photography, natural lighting

The LoRA can be tested with different poses, camera angles, environments and lighting conditions to evaluate how well the uniform transfers to new situations.

Training details

• Base model: Krea 2 Raw

• Training images: 39

• Training steps: 2000

• Resolution: 1024

• Batch size: 1

• LoRA rank: 32

• LoRA alpha: 32

• Optimizer: AdamW8bit

• UNet learning rate: 0.0001

• Text Encoder learning rate: 0

• Training engine: AI Toolkit 0.13.22

Intended use

This LoRA was created as an experimental educational project for studying generative AI and LoRA training for my University, as well as for creating illustrative and informational images related to university activities.

The model is intended to reproduce the uniform itself, not a particular person.