yuurei002/smart-medical-lora

SMART Medical LoRA

LoRA fine-tuned từ SD 1.5 trên 235 ảnh từ SMART Medical Illustration series.

Cách dùng

from diffusers import StableDiffusionPipeline
import torch

pipe = StableDiffusionPipeline.from_pretrained(
    'stable-diffusion-v1-5/stable-diffusion-v1-5', torch_dtype=torch.float16
).to('cuda')
pipe.unet.load_adapter('YOUR_HF_USERNAME/smart-medical-lora')

# QUAN TRỌNG: dùng trigger word 'smsart' trong prompt!
image = pipe(
    'smsart, medical illustration of human heart, anatomical diagram',
    negative_prompt='blurry, low quality, watermark',
    num_inference_steps=30, guidance_scale=7.5,
).images[0]

Training

• Base: SD 1.5 | LoRA rank=4 | Epochs=30
• Dataset: SMART Medical Illustration (235 ảnh, 8 chủ đề)
• Best loss: 0.0438