giga_App.py

import gradio as gr
from PIL import Image
import numpy as np
from aura_sr import AuraSR
import torch
import os
import time
from pathlib import Path
import argparse

# Force CPU usage
torch.set_default_tensor_type(torch.FloatTensor)

# Override torch.load to always use CPU
original_load = torch.load
torch.load = lambda *args, **kwargs: original_load(*args, **kwargs, map_location=torch.device('cpu'))

# Initialize the AuraSR model
aura_sr = AuraSR.from_pretrained("fal/AuraSR-v2")

# Restore original torch.load
torch.load = original_load

def process_single_image(input_image_path):
    if input_image_path is None:
        raise gr.Error("Please provide an image to upscale.")

    # Load the image
    pil_image = Image.open(input_image_path)

    # Upscale the image using AuraSR
    start_time = time.time()
    upscaled_image = aura_sr.upscale_4x(pil_image)
    processing_time = time.time() - start_time

    print(f"Processing time: {processing_time:.2f} seconds")

    # Save the upscaled image
    output_folder = "outputs"
    os.makedirs(output_folder, exist_ok=True)
    
    input_filename = os.path.basename(input_image_path)
    output_filename = os.path.splitext(input_filename)[0]
    output_path = os.path.join(output_folder, output_filename + ".png")
    
    counter = 1
    while os.path.exists(output_path):
        output_path = os.path.join(output_folder, f"{output_filename}_{counter:04d}.png")
        counter += 1
    
    upscaled_image.save(output_path)

    return [input_image_path, output_path]

def process_batch(input_folder, output_folder=None):
    if not input_folder:
        raise gr.Error("Please provide an input folder path.")

    if not output_folder:
        output_folder = "outputs"

    os.makedirs(output_folder, exist_ok=True)

    input_files = [f for f in os.listdir(input_folder) if f.lower().endswith(('.png', '.jpg', '.jpeg', '.bmp', '.tiff'))]
    total_files = len(input_files)
    processed_files = 0
    results = []

    for filename in input_files:
        input_path = os.path.join(input_folder, filename)
        pil_image = Image.open(input_path)

        start_time = time.time()
        upscaled_image = aura_sr.upscale_4x(pil_image)
        processing_time = time.time() - start_time

        output_filename = os.path.splitext(filename)[0] + ".png"
        output_path = os.path.join(output_folder, output_filename)
        
        counter = 1
        while os.path.exists(output_path):
            output_path = os.path.join(output_folder, f"{os.path.splitext(filename)[0]}_{counter:04d}.png")
            counter += 1

        upscaled_image.save(output_path)

        processed_files += 1
        print(f"Processed {processed_files}/{total_files}: {filename} in {processing_time:.2f} seconds")
        
        results.append(output_path)

    print(f"Batch processing complete. {processed_files} images processed.")
    return results

title = """<h1 align="center">AuraSR Giga Upscaler V1 by SECourses - Upscales to 4x</h1>
<p><center>AuraSR: new open source super-resolution upscaler based on GigaGAN. Works perfect on some images and fails on some images so give it a try</center></p>
<p><center>Works very fast and very VRAM friendly</center></p>
<h2 align="center">Latest version on : <a href="https://www.patreon.com/posts/110060645">https://www.patreon.com/posts/110060645</a></h1>
"""

def create_demo():
    with gr.Blocks() as demo:
        gr.HTML(title)
        
        with gr.Tab("Single Image"):
            with gr.Row():
                with gr.Column(scale=1):
                    input_image = gr.Image(label="Input Image", type="filepath")
                    process_btn = gr.Button(value="Upscale Image", variant="primary")
                with gr.Column(scale=1):
                    output_gallery = gr.Gallery(label="Before / After", columns=2)

            process_btn.click(
                fn=process_single_image,
                inputs=[input_image],
                outputs=output_gallery
            )

        with gr.Tab("Batch Processing"):
            with gr.Row():
                input_folder = gr.Textbox(label="Input Folder Path")
                output_folder = gr.Textbox(label="Output Folder Path (Optional)")
            batch_process_btn = gr.Button(value="Process Batch", variant="primary")
            output_gallery = gr.Gallery(label="Processed Images")

            batch_process_btn.click(
                fn=process_batch,
                inputs=[input_folder, output_folder],
                outputs=output_gallery
            )

    return demo

if __name__ == "__main__":
    parser = argparse.ArgumentParser(description="AuraSR Image Upscaling")
    parser.add_argument("--share", action="store_true", help="Create a publicly shareable link")
    args = parser.parse_args()

    demo = create_demo()
    demo.launch(debug=True, inbrowser=True, share=args.share)