batch_remove_watermarks.py

import os
import re
import sys
import shutil
import time
import socket
import urllib.request
import json
import torch
import torch.nn.functional as F
import numpy as np
from PIL import Image, ImageDraw

REPORT_PATH = r"C:\ComfyUI\Articles\watermark_scan_report.md"
LAMA_MODEL_PATH = os.path.join(os.path.expanduser("~"), ".cache", "torch", "hub", "checkpoints", "big-lama.pt")

COMFYUI_URL = "http://127.0.0.1:7777"
COMFYUI_INPUT_DIR = r"C:\ComfyUI\input"
COMFYUI_OUTPUT_DIR = r"C:\ComfyUI\output"
WORKFLOW_PATH = r"C:\ComfyUI\Articles\watermark_inpaint_workflow.json"

def is_port_open(host, port):
    try:
        with socket.create_connection((host, port), timeout=1):
            return True
    except OSError:
        return False

def check_comfy_online():
    return is_port_open("127.0.0.1", 7777)

def create_mask(image_path, flagged_corners, mask_path):
    img = Image.open(image_path)
    w, h = img.size
    
    # Create black image
    mask = Image.new("L", (w, h), 0)
    
    # Detect if Guardian to use larger mask
    is_guardian = "guardian" in image_path.lower()
    
    if is_guardian:
        # Guardian banner is massive in bottom right
        rect_w = int(w * 0.38)
        rect_h = int(h * 0.35)
    else:
        # 25% width, 15% height is usually enough for standard watermarks
        rect_w = int(w * 0.25)
        rect_h = int(h * 0.15)
    
    draw = ImageDraw.Draw(mask)
    
    for corner in flagged_corners:
        if corner == 'top_left':
            draw.rectangle([0, 0, rect_w, rect_h], fill=255)
        elif corner == 'top_right':
            draw.rectangle([w - rect_w, 0, w, rect_h], fill=255)
        elif corner == 'bottom_left':
            draw.rectangle([0, h - rect_h, rect_w, h], fill=255)
        elif corner == 'bottom_right':
            draw.rectangle([w - rect_w, h - rect_h, w, h], fill=255)
            
    mask.save(mask_path)
    return mask

def get_all_jpgs(root_dir):
    jpgs = []
    for root, dirs, files in os.walk(root_dir):
        for f in files:
            if f.endswith('.jpg'):
                jpgs.append(os.path.join(root, f))
    return jpgs

def simplify_string(s):
    s = s.replace('.jpg', '')
    s = re.sub(r'[^a-zA-Z0-9]', '', s).lower()
    return s

def process_report():
    if not os.path.exists(REPORT_PATH):
        print(f"Report not found at {REPORT_PATH}")
        return []
        
    with open(REPORT_PATH, 'r', encoding='utf-8') as f:
        lines = f.readlines()
        
    tasks = []
    current_filename = None
    current_corners = []
    
    for line in lines:
        line = line.strip()
        m_file = re.match(r"^### \d+\. (.+)$", line)
        if m_file:
            if current_filename and current_corners:
                tasks.append({
                    'image': current_filename,
                    'corners': current_corners
                })
            current_filename = m_file.group(1)
            current_corners = []
            continue
            
        m_flag = re.search(r"\*\*(top_left|top_right|bottom_left|bottom_right)\*\*:.*\(Flagged\)", line)
        if m_flag:
            current_corners.append(m_flag.group(1))
            
    if current_filename and current_corners:
        tasks.append({
            'image': current_filename,
            'corners': current_corners
        })
        
    all_jpgs = get_all_jpgs(r"C:\ComfyUI\Articles")
    jpg_map = []
    for path in all_jpgs:
        filename = os.path.basename(path)
        simple = simplify_string(filename)
        m = re.search(r'img(\d+)$', simple)
        img_num = m.group(1) if m else None
        jpg_map.append({'path': path, 'simple': simple, 'img_num': img_num})
        
    matched_tasks = []
    for task in tasks:
        rep_filename = os.path.basename(task['image'])
        rep_simple = simplify_string(rep_filename)
        m = re.search(r'img(\d+)$', rep_simple)
        rep_img_num = m.group(1) if m else None
        
        best_match = None
        for j in jpg_map:
            if rep_img_num and j['img_num']:
                if rep_img_num != j['img_num']: continue
            elif rep_img_num or j['img_num']: continue
                
            j_clean = j['simple'].replace('original', '')
            rep_clean = rep_simple.replace('original', '')
            if j_clean == rep_clean:
                best_match = j['path']
                break
                
        if best_match:
            matched_tasks.append({
                'image': best_match,
                'corners': task['corners']
            })
            
    return matched_tasks

def process_via_comfyui(img_path, corners):
    # Create mask locally first
    mask_path = img_path + "_mask.png"
    mask_pil = create_mask(img_path, corners, mask_path)
    
    # Save original backup
    backup_path = img_path.replace(".jpg", "_original.jpg").replace(".png", "_original.png")
    if not os.path.exists(backup_path):
        shutil.copy2(img_path, backup_path)
    
    # Copy input and mask to ComfyUI input directory
    unique_id = str(int(time.time())) + "_" + str(np.random.randint(1000, 9999))
    comfy_in_name = f"watermark_in_{unique_id}.jpg"
    comfy_mask_name = f"watermark_mask_{unique_id}.png"
    
    shutil.copy2(backup_path, os.path.join(COMFYUI_INPUT_DIR, comfy_in_name))
    shutil.copy2(mask_path, os.path.join(COMFYUI_INPUT_DIR, comfy_mask_name))
    
    # Load workflow template
    with open(WORKFLOW_PATH, 'r', encoding='utf-8') as f:
        workflow = json.load(f)
        
    # Update inputs
    workflow["39"]["inputs"]["image"] = comfy_in_name
    workflow["40"]["inputs"]["image"] = comfy_mask_name
    
    # Generate seed
    seed = int(np.random.randint(1, 10000000))
    workflow["43"]["inputs"]["seed"] = seed
    
    # Queue prompt
    print(f"  Sending to ComfyUI API...")
    payload = {"prompt": workflow, "client_id": "watermark_remover_agent"}
    data = json.dumps(payload).encode('utf-8')
    req = urllib.request.Request(f"{COMFYUI_URL}/prompt", data=data, headers={'Content-Type': 'application/json'})
    
    try:
        with urllib.request.urlopen(req) as res:
            res_data = json.loads(res.read().decode('utf-8'))
            prompt_id = res_data["prompt_id"]
    except Exception as e:
        print(f"  Failed to queue prompt: {e}")
        return False
        
    print(f"  Prompt queued. ID: {prompt_id}. Waiting for completion...")
    
    # Poll for completion
    completed = False
    output_filename = None
    
    # Wait up to 120 seconds
    for _ in range(60):
        time.sleep(2)
        # Check history
        req_hist = urllib.request.Request(f"{COMFYUI_URL}/history/{prompt_id}")
        try:
            with urllib.request.urlopen(req_hist) as res_hist:
                hist_data = json.loads(res_hist.read().decode('utf-8'))
                if prompt_id in hist_data:
                    # Completed!
                    outputs = hist_data[prompt_id].get("outputs", {})
                    # Find output from node 99 (SaveImage)
                    save_node_out = outputs.get("99", {})
                    if "images" in save_node_out and len(save_node_out["images"]) > 0:
                        output_filename = save_node_out["images"][0]["filename"]
                        completed = True
                        break
        except Exception as e:
            # ComfyUI might still be processing
            pass
            
    if not completed:
        print("  Timeout waiting for ComfyUI to process.")
        return False
        
    # Copy completed image back
    comfy_out_path = os.path.join(COMFYUI_OUTPUT_DIR, output_filename)
    if os.path.exists(comfy_out_path):
        shutil.copy2(comfy_out_path, img_path)
        print(f"  Successfully removed watermark using ComfyUI! Saved to {img_path}")
        # Clean up input folder files
        try:
            os.remove(os.path.join(COMFYUI_INPUT_DIR, comfy_in_name))
            os.remove(os.path.join(COMFYUI_INPUT_DIR, comfy_mask_name))
            os.remove(mask_path)
        except:
            pass
        return True
    else:
        print(f"  ComfyUI output file not found at {comfy_out_path}")
        return False

def process_via_local_fallback(img_path, corners, model, device):
    mask_path = img_path + "_mask.png"
    mask_pil = create_mask(img_path, corners, mask_path)
    
    backup_path = img_path.replace(".jpg", "_original.jpg").replace(".png", "_original.png")
    if not os.path.exists(backup_path):
        shutil.copy2(img_path, backup_path)
        
    try:
        img_pil = Image.open(backup_path).convert("RGB")
    except Exception as e:
        print(f"  -> Error opening image: {e}")
        return False
        
    w, h = img_pil.size
    
    img_np = np.array(img_pil, dtype=np.float32) / 255.0
    mask_np = np.array(mask_pil, dtype=np.float32) / 255.0
    mask_np = (mask_np > 0.1).astype(np.float32)
    
    image_tensor = torch.from_numpy(img_np).permute(2, 0, 1).unsqueeze(0).to(device)
    mask_tensor = torch.from_numpy(mask_np).unsqueeze(0).unsqueeze(0).to(device)
    
    pad_mod = 8
    pad_h = (pad_mod - h % pad_mod) % pad_mod
    pad_w = (pad_mod - w % pad_mod) % pad_mod
    
    if pad_h > 0 or pad_w > 0:
        image_tensor = F.pad(image_tensor, (0, pad_w, 0, pad_h), mode='reflect')
        mask_tensor = F.pad(mask_tensor, (0, pad_w, 0, pad_h), mode='constant', value=0.0)
        
    with torch.no_grad():
        output = model(image_tensor, mask_tensor)
        
    if pad_h > 0 or pad_w > 0:
        output = output[:, :, :h, :w]
        
    output = output.squeeze(0).permute(1, 2, 0).cpu().numpy()
    output = np.clip(output * 255.0, 0, 255).astype(np.uint8)
    
    output_pil = Image.fromarray(output)
    output_pil.save(img_path)
    print("  -> Success (Local Fallback)")
    
    if os.path.exists(mask_path):
        os.remove(mask_path)
    return True

def main():
    import sys
    sys.stdout.reconfigure(encoding='utf-8')
    test_mode = "--test" in sys.argv
    tasks = process_report()
    print(f"Found {len(tasks)} images to process.")
    
    if test_mode:
        tasks = tasks[:2]
        print(f"Running in TEST mode. Only processing {len(tasks)} images.")
    
    if not tasks:
        sys.exit(0)

    comfy_online = check_comfy_online()
    local_model = None
    device = None
    
    if comfy_online:
        print("ComfyUI is ONLINE. Using ComfyUI API for high-quality removal.")
    else:
        print("ComfyUI is OFFLINE. Initializing local PyTorch LaMa fallback...")
        if not os.path.exists(LAMA_MODEL_PATH):
            print("LaMa model not found. Please run lama_inpaint.py once to download it.")
            sys.exit(1)
        device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
        print(f"Using device: {device}")
        local_model = torch.jit.load(LAMA_MODEL_PATH, map_location=device)
        local_model.eval()

    for i, task in enumerate(tasks):
        img_path = task['image']
        corners = task['corners']
        print(f"[{i+1}/{len(tasks)}] Processing {img_path}")
        
        if not os.path.exists(img_path):
            print("  -> File not found!")
            continue
            
        if comfy_online:
            success = process_via_comfyui(img_path, corners)
            if not success:
                print("  Failed via ComfyUI. Attempting local fallback...")
                if local_model is None:
                    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
                    local_model = torch.jit.load(LAMA_MODEL_PATH, map_location=device)
                    local_model.eval()
                process_via_local_fallback(img_path, corners, local_model, device)
        else:
            process_via_local_fallback(img_path, corners, local_model, device)

if __name__ == "__main__":
    main()