OFASmallmodel.py

import gradio as gr
from diffusers import DiffusionPipeline
import torch

# 1. USE A SMALLER MODEL (CPU-FRIENDLY)
model_id = "OFA-Sys/small-stable-diffusion-v0"  # Lightweight model

# 2. SIMPLIFIED PIPELINE FOR CPU
pipe = DiffusionPipeline.from_pretrained(model_id)
pipe = pipe.to("cpu")  # Force CPU usage

# 3. FASTER GENERATION SETTINGS
def generate_image(prompt, negative_prompt="", steps=13):
    return pipe(
        prompt=prompt,
        negative_prompt=negative_prompt,
        num_inference_steps=steps,
        guidance_scale=7.5
    ).images[0]

# 4. STREAMLINED UI
with gr.Blocks() as demo:
    gr.Markdown("# Lightweight CPU Image Generator using OFA Small model")
    
    with gr.Row():
        with gr.Column():
            prompt = gr.Textbox(label="Your Prompt", value="a beautiful flower")
            negative = gr.Textbox(label="Avoid (Optional)", value="low-resolution")
            steps = gr.Slider(1, 30, value=13, label="Quality Steps")
            btn = gr.Button("Generate →")
        
        output = gr.Image(label="Result", height=400)
    
    btn.click(fn=generate_image, inputs=[prompt, negative, steps], outputs=output)
    
    gr.Examples(
        examples=[
            ["cityscape at night, red lights", "people", 12],
            ["watercolor painting of a flower", "photorealistic", 8]
        ],
        inputs=[prompt, negative, steps]
    )

demo.launch()