rzgar/messy-shag-hairstyles-lora-flux

Messy Shag Hairstyles

Prompt

messy_shaggy, A cinematic photo of a plus size woman with a grey messy shag hair style.her eyes are half-lidded. She islooking away as if lost in thoughts. fill lips, parted lips, half-lidded eyes. She is wearing an elegant royal blue dress. Background a cinematic photo backdrop with gradient colors and flower paterns, blured backdrop and depth of field.

Prompt

messy_shaggy, A cinematic photo of a plus size woman with a short messy pixie cut with bangs. Her face is orgasmic. her eyes are half-lidded.She is looking directly at the viewer. fill lips, parted lips, half-lidded eyes. She is wearing a elegant yellow dress detailed with white rose patern. Background a cinematic photo backdrop with gradient colors and flower paterns, blured backdrop and depth of field.

Prompt

messy_shaggy, a woman with long brown messy hair and freckles. Her hair is styled in a short, layered cut with bangs, giving her a stylish and modern look. she has an orgasmic face.

Prompt

messy_shaggy, a woman with long ginger hair, bangs and freckles. Her hair is styled in a medium length, messy, layered cut with bangs framing her face.

Model description

Description

This LoRA model is a fine-tuned adaptation of Flux.1 Dev, designed to generate messy bob and shaggy hairstyles with a distinct aesthetic popular in East Kurdistan (Kurdish region). It is specifically tailored for semi- and plus-size women aged 35 and above, reflecting the dominant hairstyle trends for this demographic in the region. The model emphasizes heavier, slightly messier hair with a natural, lived-in look, which contrasts with the default Flux.1 outputs that often produce hair that feels too light.

I created this LoRA to pair with another LoRA, rzgar/kurdish_east_clothing_styles, to achieve a cohesive look for East Kurdish women. I was pleased with the results and decided to share this model in case others are looking for the same aesthetic, messy, voluminous hairstyles with depth and a casual, effortless vibe.

Limitations

This LoRA serves a specific need and introduces subtle changes from the base Flux.1 Dev model. Here are some considerations:

• Dataset Bias: I intentionally prioritized images with fuller, more voluminous hair to achieve the heavier, messier look popular in East Kurdistan. As a result, this LoRA may not perform well for sleek, lightweight, or overly polished hairstyles. If your prompt deviates from this style, consider lowering the LoRA weight (e.g., to 0.5) or adjusting the CFG scale (e.g., to 5-7) to reduce its influence.
• Demographic Focus: The model is optimized for semi- and plus-size women over 35, reflecting East Kurdish hairstyle trends. It may require prompt adjustments for other demographics.

Demo Images

Below are some example outputs generated using this LoRA:

Training Parameters

The LoRA was trained with the following settings:

Parameter	Value
Engine	kohya
Network Dim	32
Network Alpha	16
Num Repeats	6
Resolution	1024
LR Scheduler	cosine_with_restarts
Min SNR Gamma	5
Noise Offset	0.05
Target Steps	3060
Enable Bucket	True
Optimizer Type	AdamW8Bit
Max Train Epochs	10

License

This LoRA is released under the MIT License. You are free to use, modify, and distribute it, provided you include the license and copyright notice.

Trigger words

You should use messy_shaggy to trigger the image generation.

Download model

Weights for this model are available in Safetensors format.

Download them in the Files & versions tab.