app.py

import spaces
import gradio as gr
import numpy as np
import torch
from PIL import Image
from diffusers import DDPMScheduler, StableDiffusionPipeline, DDIMScheduler, UNet2DConditionModel
from diffusers import StableDiffusionInstructPix2PixPipeline, LCMScheduler

# InstructPix2Pix with LCM specified scheduler
pipe = StableDiffusionInstructPix2PixPipeline.from_pretrained(
       "timbrooks/instruct-pix2pix", torch_dtype=torch.float16
       )
pipe = pipe.to("cuda")
pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)

# Adapt the InstructPix2Pix model using the LoRA parameters
adapter_id = "latent-consistency/lcm-lora-sdv1-5"
pipe.load_lora_weights(adapter_id)
pipe.to('cuda')

MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 1024

@spaces.GPU(duration=30)
def infer(image, edit_instruction, guidance_scale, n_steps):
    image = Image.fromarray(image).resize((512, 512))
    image = pipe(prompt=edit_instruction, 
             image=image,
             num_inference_steps=n_steps, 
             guidance_scale=guidance_scale,
             image_guidance_scale=1.0
             ).images[0]

    return image

css="""
#col-container {
    margin: 0 auto;
    max-width: 1024px;
}
"""

if torch.cuda.is_available():
    power_device = "GPU"
else:
    power_device = "CPU"

with gr.Blocks(css=css) as demo:
    
    with gr.Column(elem_id="col-container"):
        gr.Markdown(
            f"""
        # ⚡ Instruct-pix2pix with Consistency Distillation⚡ 
        Currently running on {power_device}
        """
        )
        gr.Markdown(
            "If you enjoy the space, feel free to give a ⭐ to the <a href='https://github.com/yandex-research/invertible-cd' target='_blank'>Github Repo</a>. [![GitHub Stars](https://img.shields.io/github/stars/quickjkee/instruct-pix2pix-distill?style=social)](https://github.com/quickjkee/instruct-pix2pix-distill)"
        )
        with gr.Row():
            
            edit_instruction = gr.Text(
                label="Edit instruction",
                max_lines=1,
                placeholder="Enter your prompt",
            )
            
        
        with gr.Row():
            
            with gr.Column():
                image = gr.Image(label="Input image", height=512, width=512, show_label=False)
            with gr.Column():
                result = gr.Image(label="Result", height=512, width=512, show_label=False)

        with gr.Accordion("Advanced Settings", open=True):
            
            with gr.Row():
                
                guidance_scale = gr.Slider(
                    label="guidance scale",
                    minimum=1.0,
                    maximum=5.0,
                    step=1.0,
                    value=2.0,
                )

                n_steps = gr.Slider(
                    label="inference steps",
                    minimum=1.0,
                    maximum=10.0,
                    step=1.0,
                    value=4.0,
                )

        with gr.Row():
            run_button = gr.Button("Edit", scale=0)

        with gr.Row():
            examples = [
                [
                    "examples/orig_3.jpg", #input_image
                    "turn apples into oranges", #tgt_prompt
                    2, #guidance_scale
                    4
                ],
                [
                    "examples/orig_1.jpg", #input_image
                    "Make it a Modigliani painting", #tgt_prompt
                    2, #guidance_scale
                    4
                ],
                [
                    "examples/orig_2.jpg", #input_image
                    "Turn a teddy bear into panda", #tgt_prompt
                    2, #guidance_scale
                    4
                ],
            ]
  
            gr.Examples(
               examples = examples,
               inputs =[image, edit_instruction, guidance_scale, n_steps],
               outputs=[
                        result
                        ],
               fn=infer, cache_examples=True
            )

    run_button.click(
        fn = infer,
        inputs=[image, edit_instruction, guidance_scale, n_steps],
        outputs = [result]
    )

demo.queue().launch()