app.py

import streamlit as st
from diffusers import UNet2DConditionModel, DiffusionPipeline, LCMScheduler
import torch
from PIL import Image

# Function to generate and display image
def generate_and_display_image(prompt):
    # Initialize the UNet model
    unet = UNet2DConditionModel.from_pretrained("path/to/fine-tuned/weight", torch_dtype=torch.float16, variant="fp16")

    # Initialize the diffusion pipeline
    pipeline = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", unet=unet, torch_dtype=torch.float16, variant="fp16")
    pipeline.safety_checker = None
    pipeline.requires_safety_checker = False

    # Set the loaded scheduler in the pipeline
    pipeline.scheduler = LCMScheduler.from_config(pipeline.scheduler.config)
    pipeline.to("cuda")

    # Set the number of inference steps
    inference_steps = 4

    # Generate image
    image = pipeline(prompt, num_inference_steps=inference_steps, guidance_scale=2).images[0]
    image = image.resize((512, 512))

    # Display the generated image
    st.image(image, caption="Generated Image", use_column_width=True)

# Main function
def main():
    st.title("Image Generation with Diffusion Models")

    # Input prompt
    prompt = st.text_input("Enter your prompt")

    # Button to generate and display image
    if st.button("Generate Image"):
        if prompt:
            generate_and_display_image(prompt)
        else:
            st.warning("Please provide a prompt.")

if __name__ == "__main__":
    main()