app.py

import gradio as gr
from diffusers import DiffusionPipeline
import dask
from dask import delayed

# Load model
pipe = DiffusionPipeline.from_pretrained("kandinsky-community/kandinsky-2-1")

def generate_image(prompt, num_inference_steps=50):
    """
    Generate an image based on a text prompt using diffusion with optimizations.
    The number of inference steps is reduced for faster generation.
    """
    # Reduce steps for faster processing
    image = pipe(prompt, num_inference_steps=num_inference_steps).images[0]
    return image

# Dask-delayed function to utilize multi-core CPU processing
@delayed
def dask_generate(prompt):
    return generate_image(prompt)

def parallel_generate(prompt):
    # Execute the generation using Dask to potentially improve processing speed
    image = dask.compute(dask_generate(prompt))[0]
    return image

# Gradio interface
iface = gr.Interface(
    fn=parallel_generate,
    inputs=gr.Textbox(label="Prompt", placeholder="Enter your prompt here"),
    outputs=gr.Image(type="pil"),
    title="CPU Optimized Image Generation",
    description="Enter a prompt to generate an image efficiently using CPU optimization."
)

# Launch the Gradio app
if __name__ == "__main__":
    iface.launch()