import torch
from diffusers import StableDiffusionPipeline
from peft import PeftModel
from transformers import CLIPTextModel, CLIPTokenizer
from diffusers import UNet2DConditionModel, AutoencoderKL, DDPMScheduler

# Load base model components
tokenizer = CLIPTokenizer.from_pretrained("runwayml/stable-diffusion-v1-5", subfolder="tokenizer")
text_encoder = CLIPTextModel.from_pretrained("runwayml/stable-diffusion-v1-5", subfolder="text_encoder")
vae = AutoencoderKL.from_pretrained("runwayml/stable-diffusion-v1-5", subfolder="vae")
unet = UNet2DConditionModel.from_pretrained("runwayml/stable-diffusion-v1-5", subfolder="unet")
scheduler = DDPMScheduler.from_pretrained("runwayml/stable-diffusion-v1-5", subfolder="scheduler")

# Load LoRA weights
unet = PeftModel.from_pretrained(unet, "YOUR_USERNAME/leonardo_dicaprio_lora")

# Create pipeline
pipeline = StableDiffusionPipeline(
    unet=unet,
    text_encoder=text_encoder,
    vae=vae,
    tokenizer=tokenizer,
    scheduler=scheduler,
    safety_checker=None,
    requires_safety_checker=False,
)

# Move to GPU if available
if torch.cuda.is_available():
    pipeline = pipeline.to("cuda")

# Generate image
prompt = "leonardo dicaprio eating"
image = pipeline(prompt, num_inference_steps=50, guidance_scale=7.5).images[0]
image.save("dicaprio_eating.png")

Using with Automatic1111 WebUI

Download the model files
Place them in your stable-diffusion-webui/models/Lora/ directory
Use the trigger word: leonardo dicaprio
Example prompt: leonardo dicaprio eating, high quality, detailed

Example Prompts

leonardo dicaprio eating
leonardo dicaprio smiling
leonardo dicaprio in a suit
leonardo dicaprio portrait
leonardo dicaprio on a red carpet
leonardo dicaprio with sunglasses

Training Information

This model was trained using PEFT (Parameter-Efficient Fine-Tuning) with LoRA on a dataset of 100 Leonardo DiCaprio face images. The training was optimized for realistic portrait generation while maintaining the base model's general capabilities.

Training Parameters

LoRA Rank: 16
LoRA Alpha: 16
Target Modules: ["to_q", "to_k", "to_v", "to_out.0"]
Learning Rate: 1e-4
Batch Size: 1
Gradient Accumulation Steps: 4
Mixed Precision: FP16
Training Steps: 100

Model Performance

Generation Time: ~50 seconds per image on RTX 3050
Image Quality: High-quality 512x512 portraits
Consistency: Good facial feature consistency
Style: Realistic Leonardo DiCaprio portraits

License

This model is for educational and research purposes. Please respect copyright and use responsibly.

Acknowledgments

Base model: runwayml/stable-diffusion-v1-5
Training framework: PEFT (Parameter-Efficient Fine-Tuning)
Hardware: NVIDIA RTX 3050

Citation

If you use this model in your research or projects, please cite:

bibtex

@misc{leonardo_dicaprio_lora,
  title={Leonardo DiCaprio LoRA Model},
  author={Your Name},
  year={2025},
  url={https://huggingface.co/YOUR_USERNAME/leonardo_dicaprio_lora}
}

KarunyaRonith29/leonardo_dicaprio_lora

Leonardo DiCaprio LoRA Model

Model Details

Usage

Using with Diffusers