app.py

import os
import torch
import numpy as np
import zipfile
import gradio as gr
from PIL import Image

# Load the model directly from the source repository using torch.hub
# This bypasses the need for local .py files
device = "cuda" if torch.cuda.is_available() else "cpu"
model = torch.hub.load('SkyTNT/anime-segmentation', 'isnetis', pretrained=True, trust_repo=True)
model.to(device)
model.eval()

def process_images(file_paths):
    if not file_paths:
        return None, None
    
    res_dir, mask_dir = "results_out", "masks_out"
    os.makedirs(res_dir, exist_ok=True)
    os.makedirs(mask_dir, exist_ok=True)
    
    res_list, mask_list = [], []

    for path in file_paths:
        img = Image.open(path).convert("RGB")
        original_size = img.size # Store original dimensions
        
        # Prepare image for model (the model expects 1024x1024 internally)
        input_img = img.resize((1024, 1024))
        img_np = np.array(input_img).astype(np.float32) / 255.0
        img_tensor = torch.from_numpy(img_np).permute(2, 0, 1).unsqueeze(0).to(device)
        
        with torch.no_grad():
            # Model returns a list of tensors; the first one is the main mask
            outputs = model(img_tensor)
            mask = outputs[0][0][0].cpu().numpy()
        
        # Resize mask back to original resolution to avoid quality loss
        mask_img = Image.fromarray((mask * 255).astype(np.uint8)).resize(original_size, resample=Image.BILINEAR)
        
        # Create Transparent Result
        result_img = img.copy()
        result_img.putalpha(mask_img)

        # Save files for zipping
        base_name = os.path.splitext(os.path.basename(path))[0]
        res_path = os.path.join(res_dir, f"{base_name}_no_bg.png")
        mask_path = os.path.join(mask_dir, f"{base_name}_mask.png")
        
        result_img.save(res_path)
        mask_img.save(mask_path)
        
        res_list.append(res_path)
        mask_list.append(mask_path)

    # Create ZIP files
    res_zip, mask_zip = "transparent_results.zip", "grayscale_masks.zip"
    with zipfile.ZipFile(res_zip, 'w') as z:
        for f in res_list: z.write(f, os.path.basename(f))
    with zipfile.ZipFile(mask_zip, 'w') as z:
        for f in mask_list: z.write(f, os.path.basename(f))

    return res_zip, mask_zip

# Build the Interface
with gr.Blocks(title="Anime Background Remover (Bulk)") as demo:
    gr.Markdown("## 🏮 Bulk Anime Background Remover")
    gr.Markdown("Upload multiple images. This tool processes them at their **original resolution** and provides separate ZIP downloads.")
    
    input_files = gr.File(label="Select Images", file_count="multiple", file_types=["image"])
    btn = gr.Button("🚀 Start Processing", variant="primary")
    
    with gr.Row():
        out_res = gr.File(label="1. Download Transparent Images (ZIP)")
        out_mask = gr.File(label="2. Download Grayscale Masks (ZIP)")

    btn.click(process_images, inputs=input_files, outputs=[out_res, out_mask])

demo.launch()