You can deploy this model and start earning money today!

anonhs/
sdxl-kream-clothes-lora

Name: sdxl-kream-clothes-lora
Availability: InStock

stabilityai/stable-diffusion-xl-base-1.0

adapter

creativeml-openrail-m

Views

No views yet

SDXL LoRA: KREAM Fashion Aesthetic

This LoRA was trained on a massive dataset of 14,904 high-quality images from the KREAM (KoreAn Market) ecosystem. It is designed to capture the clean, high-end, and modern streetwear aesthetic typical of premium fashion photography.

Model Description

Developed by: [anonhs]
Model type: LoRA (Low-Rank Adaptation) for Stable Diffusion XL
Base Model: SDXL 1.0 Base
Training Resolution: 1024x1024
Trigger Word: kream style

Training Details

This model was trained using QLoRA (4-bit quantization) to ensure high precision while maintaining efficient VRAM usage.

Epochs/Steps: 1 Full pass over 14,904 images (~14,904 steps at batch size 1).
Learning Rate: 1e-4
Optimizer: AdamW
Precision: bfloat16 compute with nf4 weights.

Usage Instructions

Trigger Words

To activate the specific aesthetic, use: kream style.

Recommended Settings

LoRA Weight: 0.7 - 0.9 (Adjust based on how strong you want the fashion styling).
CFG Scale: 5.0 - 7.5
Sampler: DPM++ 2M Karras or Euler a.

Example Prompt

Prompt: A professional studio fashion shot of a person wearing a luxury streetwear puffer jacket and sneakers, kream style, high-end photography, 8k, minimalist background.

Intended Use

Fashion design visualization.
Generating high-quality streetwear lookbooks.
Exploring modern Korean-inspired minimalist aesthetics.

Disclaimer

This LoRA is intended for creative and educational purposes. Please ensure you follow the license agreements of the base SDXL model.

python

import gradio as gr
from diffusers import DiffusionPipeline
import torch

# Load SDXL and your LoRA
pipe = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16)
pipe.load_lora_weights("anonhs/sdxl-kream-clothes-lora")
pipe.to("cuda")

def generate(prompt, lora_scale):
    image = pipe(prompt, cross_attention_kwargs={"scale": lora_scale}).images[0]
    return image

gr.Interface(fn=generate, inputs=["text", gr.Slider(0, 1)], outputs="image").launch()