If you are using ChatGPT, Gemini, Claude and so on for your prompt generation, you can feed it with this paste so the model knows directly how to use the LoRA - example prompts included.

This LoRA captures a sexy, fashion-focused uniform design language built around four distinct categories:

• military-inspired uniforms

• police-inspired uniforms

• captain and pilot uniforms

• stewardess and flight attendant uniforms

It focuses on fitted silhouettes, provocative cuts, structured tailoring, badges, patches, rank stripes, epaulettes, gold buttons, belts, insignia, scarves, caps and other uniform-specific details.

The LoRA was trained to understand the design language behind each category rather than reproduce one fixed outfit. Because of this, the styles can transfer well to many garment types, including:

• fitted uniforms and rompers

• mini dresses

• bodysuits

• lingerie sets

• cropped jackets and two-piece outfits

• long evening gowns

• fashion editorial looks

The four styles remain clearly separated through their characteristic details:

• military: camouflage, red piping, medals, epaulettes, tactical straps and officer-inspired elements

• police: black or navy uniforms, badges, shoulder patches, duty belts and utility details

• captain: white or black tailoring, gold buttons, sleeve stripes, epaulettes, wing insignia and nautical or aviation styling

• stewardess: navy or blue uniforms, neck scarves, wing pins, compact hats and polished cabin-crew tailoring

Trigger words

Always use:

commanding_uniforms

Then add one matching subtrigger:

• sexy_military_uniform

• sexy_police_uniform

• sexy_captain_uniform

• sexy_stewardess_uniform

The LoRA works best with clear descriptions of the garment type, material, cut, length, accessories and uniform details. Loose prompts usually create a recognizable classic uniform, while more detailed prompts allow the style to transfer to lingerie, couture dresses, bodysuits or other fashion concepts.

Would love if you post your creations here with it.

The LoRA was trained on the default flux1-dev model for maximum compatibility.

Dataset and training data:

• 52 images

• 7 steps per image

• 10 epochs

• bucks with 1152px max size, no auto scaling

• cosine scheduler with 0.2 warmup and 0.7 decay

• Captioning was done with OpenAI

• alpha == dim: 32

If you use it with other loras, for example a character lora, try to lower the weight - I had good results at around 0.5.

As always: I did not polish any of the images in the showcase, they all came out like this. I did 2 images per prompt and picked the better looking one.