app.py

import gradio as gr
from diffusers import StableDiffusionControlNetPipeline, ControlNetModel
import torch


model_id = "runwayml/stable-diffusion-v1-5"
controlnet_id = "lllyasviel/control_v11p_sd15_openpose"


controlnet = ControlNetModel.from_pretrained(controlnet_id, torch_dtype=torch.float32)
pipe = StableDiffusionControlNetPipeline.from_pretrained(
    model_id,
    controlnet=controlnet,
    safety_checker=None, 
    torch_dtype=torch.float32
)
pipe = pipe.to("cpu")


pipe.enable_attention_slicing()

def generate_image(prompt, control_image, num_inference_steps=25, guidance_scale=7.5, controlnet_conditioning_scale=1.0):
    """
    Generate an image using the ControlNet pipeline.
    
    Args:
      prompt (str): Your text prompt for image generation.
      control_image (PIL.Image): A control image to guide generation.
      num_inference_steps (int): Number of denoising steps.
      guidance_scale (float): Classifier-free guidance scale.
      controlnet_conditioning_scale (float): How strongly to condition on the control image.
      
    Returns:
      PIL.Image: The generated image.
    """
    result = pipe(
        prompt=prompt,
        image=control_image,
        num_inference_steps=num_inference_steps,
        guidance_scale=guidance_scale,
        controlnet_conditioning_scale=controlnet_conditioning_scale
    )
    return result.images[0]


with gr.Blocks() as demo:
    gr.Markdown("# ControlNet Image Generator on CPU\nThis demo uses a ControlNet pipeline (openpose variant) with Stable Diffusion to generate images guided by a control image. Note: Running on CPU can be slow!")
    
    with gr.Row():
        prompt_input = gr.Textbox(label="Prompt", placeholder="Enter your image prompt here", value="A futuristic cityscape at dusk")
    
    with gr.Row():
        
        control_image_input = gr.Image(label="Control Image", type="pil")
        output_image = gr.Image(label="Generated Image", type="pil")
    
    with gr.Row():
        num_steps = gr.Slider(minimum=10, maximum=50, value=25, step=1, label="Inference Steps")
        guidance = gr.Slider(minimum=1.0, maximum=15.0, value=7.5, step=0.5, label="Guidance Scale")
        control_scale = gr.Slider(minimum=0.1, maximum=2.0, value=1.0, step=0.1, label="ControlNet Conditioning Scale")
    
    generate_btn = gr.Button("Generate Image")
    generate_btn.click(
        fn=generate_image,
        inputs=[prompt_input, control_image_input, num_steps, guidance, control_scale],
        outputs=output_image
    )

demo.launch()