face_lora_sd15

pytorch_lora_weights.safetensors

<Gallery />

Model description

This model is a fine-tuned version of the Stable Diffusion architecture, leveraging the Low-Rank Adaptation (LoRA) technique. It has been trained using the CelebA-HQ and FFHQ datasets, both renowned for their high-quality images of human faces.

Training Details:

• Base Model: Stable Diffusion
• Adaptation Technique: Low-Rank Adaptation (LoRA)
• Datasets: CelebA-HQ (30,000 images), FFHQ (70,000 images)
• Resolution: resolution : 512*512 fine-tuning for detailed facial synthesis

Example Usages:

import torch
from diffusers import StableDiffusionPipeline,UNet2DConditionModel

pipeline = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5").to("cuda")

pipeline.load_lora_weights("phil329/face_lora_sd15", weight_name="pytorch_lora_weights.safetensors")

NEGATIVE_PROMPT = "worst quality, low quality, bad anatomy, watermark, text, blurry, cartoon, unreal"
text = 'A young woman with smile, wearing a purple hat.'

lora_image = pipeline(text,negative_prompt=NEGATIVE_PROMPT).images[0]

display(lora_image)

Results

We use four prompts as follows:

• 'A young woman with smile, wearing a purple hat.'
• 'A middle-aged man,beard ,attractive'
• 'A girl with long blonde hair'
• 'An young man with curry hair'

The negative prompt are the same as the example codes. All the results are randomly generated and not cherry-picked.

If the generation effect is not good, try adding a negative prompt, or try different prompts and seeds.

Download model

Weights for this model are available in Safetensors format.

Download them in the Files & versions tab.