utils.py

"""
Utility functions for Pixagram AI Pixel Art Generator
UPDATED VERSION with type safety helpers
"""
import numpy as np
import cv2
import math
from PIL import Image, ImageEnhance, ImageFilter, ImageDraw
from config import COLOR_MATCH_CONFIG, FACE_MASK_CONFIG, AGE_BRACKETS


def ensure_int(value):
    """Convert numpy.int64 or similar to Python int"""
    if isinstance(value, (int, float)):
        return int(value)
    return int(value.item()) if hasattr(value, 'item') else int(value)


def safe_image_size(image):
    """Get image size as pure Python ints to avoid numpy.int64 issues"""
    return (ensure_int(image.width), ensure_int(image.height))


def sanitize_text(text):
    """
    Remove or replace problematic characters (emojis, special unicode) 
    that might cause encoding errors.
    """
    if not text:
        return text
    try:
        # Encode/decode to remove invalid bytes
        text = text.encode('utf-8', errors='ignore').decode('utf-8')
        # Keep only characters within safe unicode range
        text = ''.join(char for char in text if ord(char) < 65536)
    except Exception as e:
        print(f"[WARNING] Text sanitization warning: {e}")
    return text


def color_match_lab(target, source, preserve_saturation=True):
    """
    LAB color space matching for better skin tones with saturation preservation.
    GENTLE version to prevent color fading.
    
    Args:
        target: Target image to adjust
        source: Source image to match colors from
        preserve_saturation: If True, preserves original saturation levels
    """
    try:
        target_lab = cv2.cvtColor(target.astype(np.uint8), cv2.COLOR_RGB2LAB).astype(np.float32)
        source_lab = cv2.cvtColor(source.astype(np.uint8), cv2.COLOR_RGB2LAB).astype(np.float32)
        
        result_lab = np.copy(target_lab)
        
        # Very gentle L channel matching
        t_mean, t_std = target_lab[:,:,0].mean(), target_lab[:,:,0].std()
        s_mean, s_std = source_lab[:,:,0].mean(), source_lab[:,:,0].std()
        if t_std > 1e-6:
            matched = (target_lab[:,:,0] - t_mean) * (s_std / t_std) * 0.5 + s_mean
            result_lab[:,:,0] = target_lab[:,:,0] * (1 - COLOR_MATCH_CONFIG['lab_lightness_blend']) + matched * COLOR_MATCH_CONFIG['lab_lightness_blend']
        
        if preserve_saturation:
            # Minimal adjustment to A and B channels
            for i in [1, 2]:
                t_mean, t_std = target_lab[:,:,i].mean(), target_lab[:,:,i].std()
                s_mean, s_std = source_lab[:,:,i].mean(), source_lab[:,:,i].std()
                if t_std > 1e-6:
                    matched = (target_lab[:,:,i] - t_mean) * (s_std / t_std) + s_mean
                    blend_factor = COLOR_MATCH_CONFIG['lab_color_blend_preserved']
                    result_lab[:,:,i] = target_lab[:,:,i] * (1 - blend_factor) + matched * blend_factor
        else:
            # Gentle full matching
            for i in [1, 2]:
                t_mean, t_std = target_lab[:,:,i].mean(), target_lab[:,:,i].std()
                s_mean, s_std = source_lab[:,:,i].mean(), source_lab[:,:,i].std()
                if t_std > 1e-6:
                    matched = (target_lab[:,:,i] - t_mean) * (s_std / t_std) + s_mean
                    blend_factor = COLOR_MATCH_CONFIG['lab_color_blend_full']
                    result_lab[:,:,i] = target_lab[:,:,i] * (1 - blend_factor) + matched * blend_factor
        
        return cv2.cvtColor(result_lab.astype(np.uint8), cv2.COLOR_LAB2RGB)
    except Exception as e:
        print(f"LAB conversion error: {e}")
        return target.astype(np.uint8)


def enhance_saturation(image, boost=1.05):
    """
    Minimal saturation enhancement (disabled by default).
    
    Args:
        image: PIL Image
        boost: Saturation multiplier (1.0 = no change, >1.0 = more saturated)
    """
    if boost <= 1.0:
        return image
    enhancer = ImageEnhance.Color(image)
    return enhancer.enhance(boost)


def enhanced_color_match(target_img, source_img, face_bbox=None, preserve_vibrance=False):
    """
    Enhanced color matching with face-aware processing.
    Very gentle to prevent color fading.
    
    Args:
        target_img: Generated image to adjust
        source_img: Original image to match colors from
        face_bbox: Optional [x1, y1, x2, y2] for face region
        preserve_vibrance: If True, adds minimal saturation boost (disabled by default)
    """
    try:
        target = np.array(target_img).astype(np.float32)
        source = np.array(source_img).astype(np.float32)
        
        if face_bbox is not None:
            # Create face mask
            x1, y1, x2, y2 = [int(c) for c in face_bbox]
            x1, y1 = max(0, x1), max(0, y1)
            x2, y2 = min(target.shape[1], x2), min(target.shape[0], y2)
            
            face_mask = np.zeros((target.shape[0], target.shape[1]), dtype=np.float32)
            face_mask[y1:y2, x1:x2] = 1.0
            
            # Blur mask for smooth transition
            face_mask = cv2.GaussianBlur(
                face_mask, 
                COLOR_MATCH_CONFIG['gaussian_blur_kernel'], 
                COLOR_MATCH_CONFIG['gaussian_blur_sigma']
            )
            face_mask = face_mask[:, :, np.newaxis]
            
            # Match colors for face region with saturation preservation
            if y2 > y1 and x2 > x1:
                face_result = color_match_lab(
                    target[y1:y2, x1:x2], 
                    source[y1:y2, x1:x2],
                    preserve_saturation=True
                )
                target[y1:y2, x1:x2] = face_result
                
                # Blend with original using mask
                result = target * face_mask + target * (1 - face_mask)
            else:
                result = color_match_lab(target, source, preserve_saturation=True)
        else:
            # Standard LAB color matching with saturation preservation
            result = color_match_lab(target, source, preserve_saturation=True)
        
        result_img = Image.fromarray(result.astype(np.uint8))
        
        # NO saturation boost by default
        if preserve_vibrance:
            result_img = enhance_saturation(result_img, boost=COLOR_MATCH_CONFIG['saturation_boost'])
        
        return result_img
    
    except Exception as e:
        print(f"Enhanced color matching failed: {e}, returning target image")
        return target_img


def color_match(target_img, source_img, mode='mkl'):
    """
    Legacy color matching function - kept for compatibility.
    Use enhanced_color_match for better results.
    """
    try:
        target = np.array(target_img).astype(np.float32)
        source = np.array(source_img).astype(np.float32)
        
        if mode == 'simple':
            result = np.zeros_like(target)
            for i in range(3):
                t_mean, t_std = target[:,:,i].mean(), target[:,:,i].std()
                s_mean, s_std = source[:,:,i].mean(), source[:,:,i].std()
                
                result[:,:,i] = (target[:,:,i] - t_mean) * (s_std / (t_std + 1e-6)) + s_mean
                result[:,:,i] = np.clip(result[:,:,i], 0, 255)
        
        elif mode == 'mkl':
            result = color_match_lab(target, source)
        
        else:  # pdf mode
            result = np.zeros_like(target)
            for i in range(3):
                result[:,:,i] = np.interp(
                    target[:,:,i].flatten(),
                    np.linspace(target[:,:,i].min(), target[:,:,i].max(), 256),
                    np.linspace(source[:,:,i].min(), source[:,:,i].max(), 256)
                ).reshape(target[:,:,i].shape)
        
        return Image.fromarray(result.astype(np.uint8))
    
    except Exception as e:
        print(f"Color matching failed: {e}, returning target image")
        return target_img


def create_face_mask(image, face_bbox, feather=None):
    """
    Create a soft mask around the detected face for better blending.
    
    Args:
        image: PIL Image
        face_bbox: [x1, y1, x2, y2]
        feather: blur radius for soft edges (uses config default if None)
    """
    if feather is None:
        feather = FACE_MASK_CONFIG['feather']
    
    mask = Image.new('L', image.size, 0)
    draw = ImageDraw.Draw(mask)
    
    # Expand bbox slightly
    x1, y1, x2, y2 = face_bbox
    padding = int((x2 - x1) * FACE_MASK_CONFIG['padding'])
    x1 = max(0, x1 - padding)
    y1 = max(0, y1 - padding)
    x2 = min(image.width, x2 + padding)
    y2 = min(image.height, y2 + padding)
    
    # Draw ellipse for more natural face shape
    draw.ellipse([x1, y1, x2, y2], fill=255)
    
    # Apply gaussian blur for soft edges
    mask = mask.filter(ImageFilter.GaussianBlur(feather))
    
    return mask


def draw_kps(image_pil, kps, color_list=[(255, 0, 0), (0, 255, 0), (0, 0, 255), (255, 255, 0), (255, 0, 255)]):
    """Draw facial keypoints on image for InstantID ControlNet"""
    stickwidth = 4
    limbSeq = np.array([[0, 2], [1, 2], [3, 2], [4, 2]])
    kps = np.array(kps)

    w, h = image_pil.size
    out_img = np.zeros([h, w, 3])

    for i in range(len(limbSeq)):
        index = limbSeq[i]
        color = color_list[index[0]]

        x = kps[index][:, 0]
        y = kps[index][:, 1]
        length = ((x[0] - x[1]) ** 2 + (y[0] - y[1]) ** 2) ** 0.5
        angle = math.degrees(math.atan2(y[0] - y[1], x[0] - x[1]))
        polygon = cv2.ellipse2Poly(
            (int(np.mean(x)), int(np.mean(y))), (int(length / 2), stickwidth), int(angle), 0, 360, 1
        )
        out_img = cv2.fillConvexPoly(out_img.copy(), polygon, color)
    out_img = (out_img * 0.6).astype(np.uint8)

    for idx_kp, kp in enumerate(kps):
        color = color_list[idx_kp]
        x, y = kp
        out_img = cv2.circle(out_img.copy(), (int(x), int(y)), 10, color, -1)

    out_img_pil = Image.fromarray(out_img.astype(np.uint8))
    return out_img_pil


def get_facial_attributes(face):
    """
    Extract comprehensive facial attributes.
    Returns dict with age, gender, expression, quality metrics.
    """
    attributes = {
        'age': None,
        'gender': None,
        'expression': None,
        'quality': 1.0,
        'pose_angle': 0,
        'description': []
    }
    
    # Age extraction
    try:
        if hasattr(face, 'age'):
            age = int(face.age)
            attributes['age'] = age
            for min_age, max_age, label in AGE_BRACKETS:
                if min_age <= age < max_age:
                    attributes['description'].append(label)
                    break
    except (ValueError, TypeError, AttributeError) as e:
        print(f"[WARNING] Age extraction failed: {e}")
    
    # Gender extraction
    try:
        if hasattr(face, 'gender'):
            gender_code = int(face.gender)
            attributes['gender'] = gender_code
            if gender_code == 1:
                attributes['description'].append("male")
            elif gender_code == 0:
                attributes['description'].append("female")
    except (ValueError, TypeError, AttributeError) as e:
        print(f"[WARNING] Gender extraction failed: {e}")
    
    # Expression/emotion detection (if available)
    try:
        if hasattr(face, 'emotion'):
            # Some InsightFace models provide emotion
            emotion = face.emotion
            if isinstance(emotion, (list, tuple)) and len(emotion) > 0:
                emotions = ['neutral', 'happiness', 'surprise', 'sadness', 'anger', 'disgust', 'fear']
                emotion_idx = int(np.argmax(emotion))
                emotion_name = emotions[emotion_idx] if emotion_idx < len(emotions) else 'neutral'
                confidence = float(emotion[emotion_idx])
                
                if confidence > 0.4:  # Only add if confident
                    if emotion_name == 'happiness':
                        attributes['expression'] = 'smiling'
                        attributes['description'].append('smiling')
                    elif emotion_name not in ['neutral']:
                        attributes['expression'] = emotion_name
    except (ValueError, TypeError, AttributeError, IndexError) as e:
        # Expression not available in this model
        pass
    
    # Pose angle (profile detection)
    try:
        if hasattr(face, 'pose'):
            pose = face.pose
            if len(pose) > 1:
                yaw = float(pose[1])
                attributes['pose_angle'] = abs(yaw)
    except (ValueError, TypeError, AttributeError, IndexError):
        pass
    
    # Detection quality
    try:
        if hasattr(face, 'det_score'):
            attributes['quality'] = float(face.det_score)
    except (ValueError, TypeError, AttributeError):
        pass
    
    return attributes


def build_enhanced_prompt(base_prompt, facial_attributes, trigger_word):
    """
    Build enhanced prompt with facial attributes intelligently integrated.
    """
    prompt = base_prompt
    descriptions = facial_attributes['description']
    
    if not descriptions:
        return base_prompt
    
    # Check if demographics already in prompt
    prompt_lower = prompt.lower()
    has_demographics = any(desc.lower() in prompt_lower for desc in descriptions)
    
    if not has_demographics:
        # Insert after trigger word for better integration
        demographic_str = ", ".join(descriptions) + " person"
        prompt = prompt.replace(
            trigger_word,
            f"{trigger_word}, {demographic_str}",
            1
        )
        
        age = facial_attributes.get('age')
        quality = facial_attributes.get('quality')
        expression = facial_attributes.get('expression')
        
        print(f"[FACE] Detected: {', '.join(descriptions)}")
        print(f"  Age: {age if age else 'N/A'}, Quality: {quality:.2f}")
        if expression:
            print(f"  Expression: {expression}")
    
    return prompt


def get_demographic_description(age, gender_code):
    """
    Legacy function - kept for compatibility.
    Use get_facial_attributes() for new code.
    """
    demo_desc = []
    
    if age is not None:
        try:
            age_int = int(age)
            for min_age, max_age, label in AGE_BRACKETS:
                if min_age <= age_int < max_age:
                    demo_desc.append(label)
                    break
        except (ValueError, TypeError):
            pass
    
    if gender_code is not None:
        try:
            if int(gender_code) == 1:
                demo_desc.append("male")
            elif int(gender_code) == 0:
                demo_desc.append("female")
        except (ValueError, TypeError):
            pass
    
    return demo_desc


def calculate_optimal_size(original_width, original_height, recommended_sizes=None, max_dimension=1536):
    """
    Calculate optimal size maintaining aspect ratio with dimensions as multiples of 64.
    
    This updated version supports ANY aspect ratio (not just predefined ones),
    while ensuring dimensions are multiples of 64 and keeping total pixels reasonable.
    
    Args:
        original_width: Original image width
        original_height: Original image height  
        recommended_sizes: Optional list of (width, height) tuples (legacy support)
        max_dimension: Maximum allowed dimension (default 1536)
    
    Returns:
        Tuple of (optimal_width, optimal_height) as multiples of 64
    """
    # Ensure pure Python ints to avoid numpy.int64 issues
    original_width = ensure_int(original_width)
    original_height = ensure_int(original_height)
    
    aspect_ratio = original_width / original_height
    
    # Legacy mode: use recommended sizes if provided
    if recommended_sizes is not None:
        best_match = None
        best_diff = float('inf')
        
        for width, height in recommended_sizes:
            rec_aspect = width / height
            diff = abs(rec_aspect - aspect_ratio)
            if diff < best_diff:
                best_diff = diff
                best_match = (width, height)
        
        # Ensure dimensions are multiples of 64
        width, height = best_match
        width = int((width // 64) * 64)
        height = int((height // 64) * 64)
        
        return width, height
    
    # NEW: Support any aspect ratio
    # Strategy: Keep aspect ratio, scale to reasonable total pixels, round to multiples of 64
    
    # Target total pixels (around 1 megapixel for SDXL, adjustable)
    target_pixels = 1024 * 1024  # ~1MP, good balance for SDXL
    
    # Calculate dimensions that maintain aspect ratio and hit target pixels
    # width * height = target_pixels
    # width / height = aspect_ratio
    # => width = aspect_ratio * height
    # => aspect_ratio * height^2 = target_pixels
    # => height = sqrt(target_pixels / aspect_ratio)
    
    optimal_height = math.sqrt(target_pixels / aspect_ratio)
    optimal_width = optimal_height * aspect_ratio
    
    # Ensure we don't exceed max_dimension
    if optimal_width > max_dimension:
        optimal_width = max_dimension
        optimal_height = optimal_width / aspect_ratio
    
    if optimal_height > max_dimension:
        optimal_height = max_dimension
        optimal_width = optimal_height * aspect_ratio
    
    # Round to nearest multiple of 64
    width = int(round(optimal_width / 64) * 64)
    height = int(round(optimal_height / 64) * 64)
    
    # Ensure minimum size (at least 512 on shortest side)
    min_dimension = 512
    if min(width, height) < min_dimension:
        if width < height:
            width = min_dimension
            height = int(round((width / aspect_ratio) / 64) * 64)
        else:
            height = min_dimension
            width = int(round((height * aspect_ratio) / 64) * 64)
    
    # Final safety check: ensure multiples of 64
    width = max(64, int((width // 64) * 64))
    height = max(64, int((height // 64) * 64))
    
    print(f"[SIZING] Aspect ratio: {aspect_ratio:.3f}, Output: {width}x{height} ({width*height/1e6:.2f}MP)")
    
    return width, height


def enhance_face_crop(face_crop):
    """
    Multi-stage enhancement for better feature preservation.
    
    Args:
        face_crop: PIL Image of face region
    
    Returns:
        Enhanced PIL Image
    """
    # Stage 1: Resize to optimal size for CLIP (224x224)
    face_crop_resized = face_crop.resize((224, 224), Image.LANCZOS)
    
    # Stage 2: Enhance sharpness (helps with facial features)
    enhancer = ImageEnhance.Sharpness(face_crop_resized)
    face_crop_sharp = enhancer.enhance(1.5)
    
    # Stage 3: Enhance contrast slightly (helps with lighting)
    enhancer = ImageEnhance.Contrast(face_crop_sharp)
    face_crop_enhanced = enhancer.enhance(1.1)
    
    # Stage 4: Slight brightness adjustment to normalize lighting
    enhancer = ImageEnhance.Brightness(face_crop_enhanced)
    face_crop_final = enhancer.enhance(1.05)
    
    return face_crop_final


print("[OK] Utilities loaded")