sd3-lora-celebrities

This is a LoRA derived from stabilityai/stable-diffusion-3-medium-diffusers.

The main validation prompt used during training was:

a studio portrait photograph of emma watson. she looks relaxed and happy.

Validation settings

• CFG: 5.0
• CFG Rescale: 0.2
• Steps: 50
• Sampler: euler
• Seed: 2
• Resolution: 1280x768

Note: The validation settings are not necessarily the same as the training settings.

You can find some example images in the following gallery:

Prompt

a studio portrait photograph of emma watson. she looks relaxed and happy.

Negative Prompt

blurry, cropped, ugly

The text encoder was not trained. You may reuse the base model text encoder for inference.

Training settings

• Training epochs: 5
• Training steps: 9316
• Learning rate: 0.0001
• Effective batch size: 1
  • Micro-batch size: 1
  • Gradient accumulation steps: 1
  • Number of GPUs: 1
• Prediction type: v_prediction
• Rescaled betas zero SNR: True
• Optimizer: AdamW, stochastic bf16
• Precision: Pure BF16
• Xformers: Not used
• LoRA Rank: 16
• LoRA Alpha: 16
• LoRA Dropout: 0.1
• LoRA initialisation style: default

Datasets

celebrities-sd3

• Repeats: 0
• Total number of images: 1830
• Total number of aspect buckets: 27
• Resolution: 0.5 megapixels
• Cropped: False
• Crop style: None
• Crop aspect: None

Inference

import torch
from diffusers import StableDiffusion3Pipeline



model_id = "sd3-lora-celebrities"
prompt = "a studio portrait photograph of emma watson. she looks relaxed and happy."
negative_prompt = "malformed, disgusting, overexposed, washed-out"

pipeline = DiffusionPipeline.from_pretrained(model_id)
pipeline.to('cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu')
image = pipeline(
    prompt=prompt,
    negative_prompt='blurry, cropped, ugly',
    num_inference_steps=50,
    generator=torch.Generator(device='cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu').manual_seed(1641421826),
    width=1152,
    height=768,
    guidance_scale=5.0,
    guidance_rescale=0.2,
).images[0]
image.save("output.png", format="PNG")