app.py

import gradio as gr
from PIL import Image
import os
import tempfile
from transformers import pipeline

def optimize_image(image, png_optimize, jpeg_quality, jpeg_resolution, webp_quality):
    img = Image.open(image)
    
    original_size = os.path.getsize(image.name) / 1024  # tamaño en KB
    
    # Crear un directorio seguro para almacenar las imágenes optimizadas
    output_dir = "/tmp/optimized_images"
    os.makedirs(output_dir, exist_ok=True)
    
    # Generar archivos temporales en un directorio seguro
    lossless_output_path = os.path.join(output_dir, "lossless.png")
    lossy_output_path = os.path.join(output_dir, "lossy.jpg")
    reduced_output_path = os.path.join(output_dir, "reduced_resolution.jpg")
    webp_lossy_output_path = os.path.join(output_dir, "lossy.webp")

    # 1. Compresión sin pérdida (PNG)
    img.save(lossless_output_path, format="PNG", optimize=png_optimize)
    lossless_size = os.path.getsize(lossless_output_path) / 1024
    lossless_diff = original_size - lossless_size
    lossless_img = Image.open(lossless_output_path)
    
    # 2. Compresión con pérdida (JPEG)
    img.save(lossy_output_path, format="JPEG", quality=jpeg_quality, optimize=True)
    lossy_size = os.path.getsize(lossy_output_path) / 1024
    lossy_diff = original_size - lossy_size
    lossy_img = Image.open(lossy_output_path)
    
    # 3. Reducción de resolución (JPEG)
    new_resolution = (img.width * jpeg_resolution // 100, img.height * jpeg_resolution // 100)
    reduced_img = img.resize(new_resolution, Image.LANCZOS)
    reduced_img.save(reduced_output_path, format="JPEG", quality=jpeg_quality, optimize=True)
    reduced_size = os.path.getsize(reduced_output_path) / 1024
    reduced_diff = original_size - reduced_size
    reduced_img = Image.open(reduced_output_path)

    # 4. Compresión con WebP (con pérdida)
    img.save(webp_lossy_output_path, format="WEBP", quality=webp_quality, optimize=True)
    webp_lossy_size = os.path.getsize(webp_lossy_output_path) / 1024
    webp_lossy_diff = original_size - webp_lossy_size
    webp_lossy_img = Image.open(webp_lossy_output_path)
    
    # Retornar las imágenes optimizadas, tamaños y rutas para descarga
    return [
        lossless_img, f"Sin pérdida: {lossless_size:.2f} KB \n(diferencia: {-lossless_diff:.2f} KB)", lossless_output_path,
        lossy_img, f"Con pérdida: {lossy_size:.2f} KB \n(diferencia: {-lossy_diff:.2f} KB)", lossy_output_path,
        reduced_img, f"Reducción de resolución: {reduced_size:.2f} KB \n(diferencia: {-reduced_diff:.2f} KB)", reduced_output_path,
        webp_lossy_img, f"WebP con pérdida: {webp_lossy_size:.2f} KB \n(diferencia: {-webp_lossy_diff:.2f} KB)", webp_lossy_output_path
    ]

# Función para aplicar un modelo seleccionado desde Hugging Face
def apply_model(image, model_name):
    model_pipeline = pipeline("image-super-resolution", model=model_name)
    return model_pipeline(image)

with gr.Blocks() as demo:
    with gr.Tab("Optimización Tradicional"):
        image_input = gr.File(label="Sube tu imagen", file_types=['image'])
        optimize_button = gr.Button("Optimizar")

        with gr.Row():
            with gr.Column():
                optimized_output1 = gr.Image(label="Optimización sin pérdida")
                png_optimize = gr.Checkbox(label="Optimizar PNG", value=True)
                download_button1 = gr.File(label="Descargar", visible=True)
                optimized_size1 = gr.Text(value="", interactive=False, show_label=False)

            with gr.Column():
                optimized_output2 = gr.Image(label="Optimización con pérdida (JPEG)")
                jpeg_quality = gr.Slider(label="Calidad JPEG", minimum=10, maximum=100, value=50, step=1)
                download_button2 = gr.File(label="Descargar", visible=True)
                optimized_size2 = gr.Text(value="", interactive=False, show_label=False)
                
            with gr.Column():
                optimized_output3 = gr.Image(label="Reducción de resolución (JPEG)")
                jpeg_resolution = gr.Slider(label="Resolución JPEG (%)", minimum=10, maximum=100, value=50, step=1)
                download_button3 = gr.File(label="Descargar", visible=True)
                optimized_size3 = gr.Text(value="", interactive=False, show_label=False)

            with gr.Column():
                optimized_output4 = gr.Image(label="Optimización WebP con pérdida")
                webp_quality = gr.Slider(label="Calidad WebP", minimum=10, maximum=100, value=50, step=1)
                download_button4 = gr.File(label="Descargar", visible=True)
                optimized_size4 = gr.Text(value="", interactive=False, show_label=False)

        # Conectar cada control a la función de optimización
        optimize_button.click(
            fn=optimize_image,
            inputs=[image_input, png_optimize, jpeg_quality, jpeg_resolution, webp_quality],
            outputs=[
                optimized_output1, optimized_size1, download_button1,
                optimized_output2, optimized_size2, download_button2,
                optimized_output3, optimized_size3, download_button3,
                optimized_output4, optimized_size4, download_button4
            ]
        )

    with gr.Tab("Optimización con Modelos de Hugging Face"):
        hf_image_input = gr.File(label="Sube tu imagen para optimización avanzada", file_types=['image'])
        model_selector = gr.Dropdown(
            label="Selecciona un modelo",
            choices=["xinntao/Real-ESRGAN", "google/ddpm-cifar10-32", "facebook/ddpm"],  # Añade los modelos disponibles
            value="xinntao/Real-ESRGAN"
        )
        hf_output = gr.Image(label="Resultado")
        hf_button = gr.Button("Aplicar Modelo")

        hf_button.click(
            fn=apply_model,
            inputs=[hf_image_input, model_selector],
            outputs=hf_output
        )

demo.launch()