app.py

import streamlit as st
from PIL import Image, ImageFont, ImageDraw
import io
import base64
import google.generativeai as genai
import os
import requests
import random
import numpy as np
import cv2
from rembg import remove
import textwrap
import easyocr
import pytesseract
from fontTools.ttLib import TTFont
from langchain_groq import ChatGroq
import logging
from together import Together


# Load environment variables
from dotenv import load_dotenv
load_dotenv()

# Configure the generative AI model
genai.configure(api_key=os.getenv("GOOGLE_API_KEY"))

generation_config = {
    "temperature": 0,
    "top_p": 1,
    "top_k": 1,
    "max_output_tokens": 2048,
}

model = genai.GenerativeModel(
    model_name="gemini-1.5-pro",
    generation_config=generation_config,
)


pytesseract.pytesseract.tesseract_cmd = r'Tesseract-OCR\tesseract.exe'

# Set up Groq LLM
llm = ChatGroq(
    temperature=0.7,
    groq_api_key=os.getenv('GROQ_API_KEY'),
    model_name="llama-3.3-70b-versatile"
)


# Content from agent.py
def generate_advertisement_prompt(description):
    prompt = f"""
    Based on the following detailed description for an advertisement post, image prompt:

    Description: "{description}"

    Generate a detailed image prompt for an AI image generation model, incorporating these elements:
    1. Header: Give one header based on the description.
    2. Sub Header: Give one sub header based on the description.

    3. Subject: Describe the main subject or product in the advertisement, including its key features and visual characteristics.

    4. Composition: Explain how the subject should be positioned within the frame, any specific angles or perspectives to highlight its best features.

    5. Background: Detail the setting or environment that complements the subject and reinforces the advertisement's message.

    6. Text Elements: text elemeny should be adverticement purpose like based on the description generate adverticement text.

    7. Style: Describe the overall visual style, color scheme, and mood that best represents the brand and appeals to the target audience.

    8. Additional Elements: List any supporting visual elements, such as logos, icons, or graphics that should be included to enhance the advertisement's impact.

    9. Target Audience: Briefly mention the intended audience to ensure the image resonates with them.

    10. Think in the basis of adverticment designer and combined all 9 points and make a final prompt.

    Please provide a cohesive image prompt that incorporates all these elements into a striking, attention-grabbing advertisement poster, based on the given description. The prompt should be detailed enough to generate a compelling and effective advertisement image.
    """

    response = llm.invoke(prompt)
    return response.content

def advertisement_generator():
    #st.title("Advertisement Post Generator")

    post_description = st.text_input("Enter a brief description for your advertisement post:")

    if st.button("Generate Image Prompt"):
        if post_description:
            with st.spinner("Prompt Enhancer..."):
                final_prompt = generate_advertisement_prompt(post_description)
                st.subheader("Generated Image Prompt:")
                
                st.text_area(label="Final Prompt", value=final_prompt.strip(), height=200)
        else:
            st.warning("Please enter a description for your post.")

# Content from image_generation.py
logging.basicConfig(level=logging.DEBUG)
logger = logging.getLogger(__name__)

client = Together()

def generate_poster():
    #st.header("Generate Social Media Post")
    description = st.text_input("Enter prompt for Advertisement:")
    col1, col2 = st.columns(2)  # Equal width columns
    with col1:
        if st.button("✨Enhance My Prompt", key="ad_generator_button", use_container_width=True):
            st.session_state.show_ad_generator = True
    with col2:
        generate_button = st.button("Generate Graphics", use_container_width=True)

    st.markdown("<div style='height: 30px;'></div>", unsafe_allow_html=True)

    if st.session_state.get('show_ad_generator', False):
        with st.expander("Prompt Enhancer :arrow_right:", expanded=True):
            advertisement_generator()

    st.markdown("<div style='height: 30px;'></div>", unsafe_allow_html=True)

    col1, col2 = st.columns(2)
    with col1:
        post_type = st.selectbox("Select Post Type", ["Instagram advertisement post", "Facebook advertisement post", "Twitter advertisement post", "Other"])
        aspect_ratio = st.selectbox("Select Image Aspect Ratio", ["1:1", "16:9", "4:5", "9:16"])
    with col2:
        if aspect_ratio == "1:1":
            dimensions = st.selectbox("Select Image Dimensions", ["1024x1024", "1200x1200", "1504x1504"])
        elif aspect_ratio == "16:9":
            dimensions = st.selectbox("Select Image Dimensions", ["1024x576", "1280x720", "1792x1008"])
        elif aspect_ratio == "4:5":
            dimensions = st.selectbox("Select Image Dimensions", ["1024x1280", "1200x1500", "1600x2000"])
        elif aspect_ratio == "9:16":
            dimensions = st.selectbox("Select Image Dimensions", ["576x1024", "720x1280", "1008x1792"])
        
        design_style = st.selectbox("Select Design Style", [
            "Minimalistic", "Bold/Graphic", "Elegant", "Playful/Fun",
            "Corporate/Professional", "Retro/Vintage", "Modern/Contemporary", "Illustrative/Artistic"
        ])

    st.markdown("<div style='height: 30px;'></div>", unsafe_allow_html=True)

    # Extract width and height from the selected dimensions
    width, height = map(int, dimensions.split('x'))
    
    with st.expander("Add Content : Header, Sub-header and Descriptions", expanded=False):
        header = st.text_input("Enter Header for Advertisement:")
        sub_header = st.text_input("Enter Sub-header for Advertisement:")    
        # Allow multiple user prompts
        user_prompts = []
        num_prompts = st.number_input("Number of Text Descriptions", min_value=1, max_value=80, value=1)
        for i in range(num_prompts):
            user_prompt = st.text_area(f"Enter Descriptions to display in the image (Descriptions {i+1}):")
            user_prompts.append(user_prompt)

    st.markdown("<div style='height: 30px;'></div>", unsafe_allow_html=True)
    
    with st.expander("Add Branding : Logo and Color", expanded=False):
        # Add color selection with predefined options
        color_options = ["None", "Black", "White", "Red", "Blue", "Green", "Yellow", "Purple"]
        selected_color = st.selectbox("Choose a dominant color for the image", color_options)
        
        logo = st.file_uploader("Upload Logo (optional)", type=['png', 'jpg', 'jpeg'])
        
        # Add logo position selection
        logo_position = st.selectbox("Select Logo Position", [
            "None", "Top Left", "Top Middle", "Top Right",
            "Left Middle", "Right Middle",
            "Bottom Left", "Bottom Middle", "Bottom Right"
        ])

    st.markdown("<div style='height: 30px;'></div>", unsafe_allow_html=True)

    if generate_button:
        # Generate 4 different variations of the prompt with enhanced realism and attention-grabbing elements
        lighting_options = ['golden hour lighting', 'studio lighting', 'natural daylight', 'dramatic spotlights']
        visual_elements = ['3D elements', 'metallic accents', 'glass effects', 'neon highlights']
        
        prompt_variations = [
            f"Create a professional and eye-catching {post_type.lower()} advertisement. The image should feature impactful {selected_color.lower() if selected_color != 'None' else 'vibrant'} colors that align with brand identity. Header: \"{header}\". Sub-header: \"{sub_header}\". Implement a {design_style.lower()} design style with clean, commercial-grade visuals. Main focus: {description}. Compose in {aspect_ratio} aspect ratio at {width}x{height}. The design should incorporate modern {random.choice(visual_elements)} to enhance visual appeal. Ensure high resolution with perfect clarity and legibility. Text should be bold, clear and strategically placed for maximum impact. Create a compelling visual hierarchy that drives attention to key messaging and call-to-action elements. Make it look like a premium advertisement created by a professional design agency. Variation {i+1}/4."
            for i in range(4)
        ]

        generated_images = []
        for i, prompt in enumerate(prompt_variations):
            with st.spinner(f"Generating Graphic {i+1}..."):
                logger.info(f"Generating Graphic {i+1} with prompt: {prompt}")
                
                # Adjust dimensions if needed to stay within API limits
                adjusted_width = min(1792, max(64, width))
                adjusted_height = min(1792, max(64, height))
                
                # Maintain aspect ratio while adjusting dimensions
                if width > 1792 or height > 1792:
                    ratio = min(1792/width, 1792/height)
                    adjusted_width = int(width * ratio)
                    adjusted_height = int(height * ratio)
                
                # Generate image using Together API
                response = client.images.generate(
                    prompt=prompt,
                    model="black-forest-labs/FLUX.1-schnell-Free",
                    width=adjusted_width,
                    height=adjusted_height,
                    steps=4,
                    n=1,
                    response_format="b64_json"
                )
                
                if response.data:
                    # Convert base64 to image
                    image_data = base64.b64decode(response.data[0].b64_json)
                    image = Image.open(io.BytesIO(image_data))
                    
                    # Resize back to original dimensions if needed
                    if adjusted_width != width or adjusted_height != height:
                        image = image.resize((width, height), Image.LANCZOS)
                    
                    # Add logo if provided
                    if logo:
                        logo_image = Image.open(logo)
                        logo_width = int(image.width * 0.15)  # 15% of the image width
                        logo_height = int(logo_image.height * (logo_width / logo_image.width))
                        logo_image = logo_image.resize((logo_width, logo_height), Image.LANCZOS)
                        
                        padding = int(image.width * 0.02)  # Fixed 2% padding
                        
                        if logo_position == "None":
                            # Randomly choose a corner for logo placement
                            corner = random.choice(["Top Left", "Top Right", "Bottom Left", "Bottom Right"])
                            if corner == "Top Left":
                                position = (padding, padding)
                            elif corner == "Top Right":
                                position = (image.width - logo_width - padding, padding)
                            elif corner == "Bottom Left":
                                position = (padding, image.height - logo_height - padding)
                            else:  # Bottom Right
                                position = (image.width - logo_width - padding, image.height - logo_height - padding)
                        else:
                            if logo_position == "Top Left":
                                position = (padding, padding)
                            elif logo_position == "Top Middle":
                                position = ((image.width - logo_width) // 2, padding)
                            elif logo_position == "Top Right":
                                position = (image.width - logo_width - padding, padding)
                            elif logo_position == "Bottom Left":
                                position = (padding, image.height - logo_height - padding)
                            elif logo_position == "Bottom Middle":
                                position = ((image.width - logo_width) // 2, image.height - logo_height - padding)
                            else:  # Bottom Right
                                position = (image.width - logo_width - padding, image.height - logo_height - padding)
                    
                        # Create a new image with an alpha channel
                        combined_image = Image.new('RGBA', image.size, (0, 0, 0, 0))
                        combined_image.paste(image, (0, 0))
                        
                        # Convert logo to RGBA if it's not already
                        if logo_image.mode != 'RGBA':
                            logo_image = logo_image.convert('RGBA')
                        
                        combined_image.paste(logo_image, position, logo_image)
                        
                        # Convert back to RGB for compatibility
                        image = combined_image.convert('RGB')
                    
                    generated_images.append(image)
                    
                    # Display generated image
                    st.image(image, caption=f"Generated Poster {i+1}", use_column_width=True)
                    
                    # Provide download option for the generated image
                    buf = io.BytesIO()
                    image.save(buf, format="PNG")
                    byte_im = buf.getvalue()
                    st.download_button(
                        label=f"Download generated Graphic {i+1}",
                        data=byte_im,
                        file_name=f"generated_Graphic_{i+1}.png",
                        mime="image/png"
                    )
                else:
                    st.error(f"Failed to generate Graphic {i+1}")

        st.markdown("<div style='height: 15px;'></div>", unsafe_allow_html=True)

# Content from image_to_image.py
def encode_image(image):
    buffered = io.BytesIO()
    image.save(buffered, format="PNG")
    return base64.b64encode(buffered.getvalue()).decode('utf-8')

def generate_image_prompt(image):
    encoded_image = encode_image(image)
    
    prompt_parts = [
        {"mime_type": "image/png", "data": base64.b64decode(encoded_image)},
        "Analyze this image and generate a detailed prompt that could be used to recreate this image using an AI image generation model. Include key visual elements, style, composition,text element, and any other relevant details."
    ]

    response = model.generate_content(prompt_parts)
    return response.text

def generate_new_image(prompt):
    # Generate image using Together API
    response = client.images.generate(
        prompt=prompt,
        model="black-forest-labs/FLUX.1-schnell-Free",
        width=1024,
        height=768,
        steps=4,
        n=1,
        response_format="b64_json"
    )
    
    if response.data:
        image_data = base64.b64decode(response.data[0].b64_json)
        return Image.open(io.BytesIO(image_data))
    return None

# Part 2: Image Editing

# Content from image_assembled.py
def load_image(image_file):
    img = Image.open(image_file)
    return img

def crop_image(img, left, top, right, bottom):
    return img.crop((left, top, right, bottom))

def resize_image(img, max_size):
    width, height = img.size
    if width > height:
        if width > max_size:
            ratio = max_size / width
            new_size = (max_size, int(height * ratio))
    else:
        if height > max_size:
            ratio = max_size / height
            new_size = (int(width * ratio), max_size)
    return img.resize(new_size, Image.LANCZOS)

def assemble_images(background, images, positions, sizes):
    canvas = background.copy()
    for img, pos, size in zip(images, positions, sizes):
        resized_img = img.resize(size, Image.LANCZOS)
        canvas.paste(resized_img, pos, resized_img if resized_img.mode == 'RGBA' else None)
    return canvas

def drag_and_resize_images(background, images, positions, sizes):
    def on_mouse(event, x, y, flags, param):
        nonlocal dragging, resizing, active_image, offset_x, offset_y, start_size, resize_corner

        if event == cv2.EVENT_LBUTTONDOWN:
            for i, (image, pos, size) in enumerate(zip(images, positions, sizes)):
                ix, iy = pos
                if ix <= x <= ix + size[0] and iy <= y <= iy + size[1]:
                    active_image = i
                    offset_x = x - ix
                    offset_y = y - iy
                    start_size = size
                    
                    # Check if click is near a corner (within 10 pixels)
                    corner_size = 10
                    if (x - ix < corner_size and y - iy < corner_size) or \
                       (ix + size[0] - x < corner_size and y - iy < corner_size) or \
                       (x - ix < corner_size and iy + size[1] - y < corner_size) or \
                       (ix + size[0] - x < corner_size and iy + size[1] - y < corner_size):
                        resizing = True
                        resize_corner = (x - ix, y - iy)
                    else:
                        dragging = True
                    break

        elif event == cv2.EVENT_MOUSEMOVE:
            if dragging:
                positions[active_image] = (x - offset_x, y - offset_y)
            elif resizing:
                dx = x - (positions[active_image][0] + resize_corner[0])
                dy = y - (positions[active_image][1] + resize_corner[1])
                
                if resize_corner[0] < start_size[0] / 2:
                    dx = -dx
                if resize_corner[1] < start_size[1] / 2:
                    dy = -dy
                
                new_width = max(10, start_size[0] + dx)
                new_height = max(10, start_size[1] + dy)
                sizes[active_image] = (int(new_width), int(new_height))

        elif event == cv2.EVENT_LBUTTONUP:
            dragging = False
            resizing = False

    dragging = False
    resizing = False
    active_image = -1
    offset_x, offset_y = 0, 0
    start_size = (0, 0)
    resize_corner = (0, 0)

    window_name = "Drag and Resize Images"
    cv2.namedWindow(window_name)
    cv2.setMouseCallback(window_name, on_mouse)

    while True:
        img_copy = assemble_images(background, images, positions, sizes)
        cv2.imshow(window_name, cv2.cvtColor(np.array(img_copy), cv2.COLOR_RGB2BGR))
        key = cv2.waitKey(1) & 0xFF
        if key == 27:  # ESC key
            break

    cv2.destroyAllWindows()
    return positions, sizes, img_copy

# Content from text_replacer.py
def detect_text(image):
    reader = easyocr.Reader(['en'])
    img_array = np.array(image)
    results = reader.readtext(img_array)
    return [(text, box) for (box, text, _) in results]

def get_text_color(image, box):
    x, y = int(box[0][0]), int(box[0][1])
    rgb_image = image.convert('RGB')
    color = rgb_image.getpixel((x, y))
    return color

def detect_font(image, box):
    x, y, w, h = int(box[0][0]), int(box[0][1]), int(box[2][0] - box[0][0]), int(box[2][1] - box[0][1])
    cropped = image.crop((x, y, x+w, y+h))
    
    # Use Tesseract to detect font
    custom_config = r'--oem 3 --psm 6 -c tessedit_char_whitelist=ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789'
    font_info = pytesseract.image_to_data(cropped, config=custom_config, output_type=pytesseract.Output.DICT)
    
    # Check if 'font' key exists in font_info
    if 'font' in font_info:
        # Get the most common font
        fonts = [f for f in font_info['font'] if f != '']
        if fonts:
            most_common_font = max(set(fonts), key=fonts.count)
            return most_common_font
    
    return "arialbd"  # Default to Bold Arial if no font detected or 'font' key not present

def replace_text_in_image(image, text_to_replace, new_text, new_color):
    img_array = np.array(image)
    for (text, box) in detect_text(image):
        if text == text_to_replace:
            x, y, w, h = int(box[0][0]), int(box[0][1]), int(box[2][0] - box[0][0]), int(box[2][1] - box[0][1])
            
            # Detect the font of the original text
            detected_font = detect_font(image, box)
            
            # Create a mask for the text area
            mask = np.zeros(img_array.shape[:2], dtype=np.uint8)
            cv2.rectangle(mask, (x, y), (x+w, y+h), 255, -1)

            # Inpaint the text area
            img_array = cv2.inpaint(img_array, mask, 3, cv2.INPAINT_TELEA)

            image = Image.fromarray(img_array)
            draw = ImageDraw.Draw(image)
            font_size = int(h * 0.8) 
            
            supported_extensions = ['.ttf', '.otf', '.woff', '.woff2']
            font_path = None
            for ext in supported_extensions:
                if os.path.exists(f"{detected_font}{ext}"):
                    font_path = f"{detected_font}{ext}"
                    break
            if not font_path:
                font_path = "arialbd.ttf"  # Default to Bold Arial if no supported font found
            
            font = ImageFont.truetype(font_path, font_size)
            
            draw.text((x, y), new_text, font=font, fill=new_color)
            return image

    return Image.fromarray(img_array)

# Content from text.py
def put_text(img, text, x_value, y_value, color):
    if img is None:
        raise ValueError("Image not found or could not be loaded.")
    
    font = cv2.FONT_HERSHEY_DUPLEX
    wrapped_text = textwrap.wrap(text, width=30)
    font_size = 1
    font_thickness = 2

    for i, line in enumerate(wrapped_text):
        textsize = cv2.getTextSize(line, font, font_size, font_thickness)[0]

        gap = textsize[1] + 10
        y = y_value + i * gap
        x = x_value
        cv2.putText(img, line, (x, y), font,
                    font_size,
                    color,
                    font_thickness,
                    lineType = cv2.LINE_AA)

def drag_text(img, texts):
    def on_mouse(event, x, y, flags, param):
        nonlocal dragging, active_text, offset_x, offset_y

        if event == cv2.EVENT_LBUTTONDOWN:
            for i, (text, pos, color) in enumerate(texts):
                tx, ty = pos
                wrapped_text = textwrap.wrap(text, width=30)
                text_height = len(wrapped_text) * 30  # Approximate text height
                text_width = max(cv2.getTextSize(line, cv2.FONT_HERSHEY_DUPLEX, 1, 2)[0][0] for line in wrapped_text)
                if tx <= x <= tx + text_width and ty - 30 <= y <= ty + text_height:  # Expanded clickable area
                    dragging = True
                    active_text = i
                    offset_x = x - tx
                    offset_y = y - ty
                    break

        elif event == cv2.EVENT_MOUSEMOVE:
            if dragging:
                texts[active_text] = (texts[active_text][0], (x - offset_x, y - offset_y), texts[active_text][2])

        elif event == cv2.EVENT_LBUTTONUP:
            dragging = False

    dragging = False
    active_text = -1
    offset_x, offset_y = 0, 0

    window_name = "Drag Text"
    cv2.namedWindow(window_name)
    cv2.setMouseCallback(window_name, on_mouse)

    while True:
        img_copy = img.copy()
        for text, (x, y), color in texts:
            put_text(img_copy, text, x, y, color)

        cv2.imshow(window_name, img_copy)
        key = cv2.waitKey(1) & 0xFF
        if key == 27:  # ESC key
            break

    cv2.destroyAllWindows()
    return texts, img_copy

# Content from background_remove.py
def remove_background(image):
    # Convert PIL Image to numpy array
    img_array = np.array(image)
    
    # Remove background
    result = remove(img_array)
    
    # Convert back to PIL Image
    return Image.fromarray(result)

# Main Streamlit App
def main():    
    # Add logo to the center of the sidebar
    logo = Image.open("Mark8 AI.png")  # Replace with your logo path
    st.sidebar.markdown(
        """
        <style>
        .sidebar-logo {
            display: flex;
            justify-content: center;
            align-items: center;
            padding: 1rem 0;
        }
        </style>
        """,
        unsafe_allow_html=True
    )
    st.sidebar.markdown('<div class="sidebar-logo">', unsafe_allow_html=True)
    st.sidebar.image(logo, width=150)
    st.sidebar.markdown('</div>', unsafe_allow_html=True)
    
    # Initialize session state for page
    if 'page' not in st.session_state:
        st.session_state.page = "poster_generation"
    
    # Function to display title and description
    def display_title_and_description(title, description):
        st.title(title)
        st.write(description)
    
    # Create even-shaped buttons in the sidebar
    button_style = """
        <style>
        div.stButton > button {
            width: 100%;
            height: 3em;
            margin-bottom: 10px;
        }
        </style>
    """
    st.sidebar.markdown(button_style, unsafe_allow_html=True)
    
    if st.sidebar.button("Designer"):
        st.session_state.page = "poster_generation"
    
    if st.sidebar.button("Image to Image Generation"):
        st.session_state.page = "text_to_image"
    
    if st.sidebar.button("Image Editing"):
        st.session_state.page = "image_editing"
    
    if st.sidebar.button("Advertisement Generator"):
        st.session_state.page = "advertisement_generator"

    if st.session_state.page == "text_to_image":
        display_title_and_description("Mark8 Designer", "Transform your ideas into stunning visuals.")
        text_to_image_generation()
    elif st.session_state.page == "image_editing":
        display_title_and_description("Mark8 Designer", "Enhance and modify your images with powerful tools.")
        image_editing()
    elif st.session_state.page == "poster_generation":
        display_title_and_description("Mark8 Designer", "Create eye-catching posters for various platforms.")
        generate_poster()
    elif st.session_state.page == "advertisement_generator":
        display_title_and_description("Mark8 Designer", "Create compelling advertisements with AI assistance.")
        advertisement_generator()

def text_to_image_generation():
   # st.header("Text to Image Generation")

    # Image to Image Generation
    st.subheader("Image to Image Generation")
    uploaded_file = st.file_uploader("Choose an image:", type=["png", "jpg", "jpeg"])
    if uploaded_file is not None:
        image = Image.open(uploaded_file)
        st.image(image, caption="Uploaded Image", use_column_width=True)

        if st.button("Generate Image Prompt"):
            with st.spinner("Analyzing image and generating prompt..."):
                generated_prompt = generate_image_prompt(image)
                st.subheader("Generated Image Prompt:")
                st.text_area(label="", value=generated_prompt.strip(), height=200, key="generated_prompt", disabled=True)
                st.session_state['saved_prompt'] = generated_prompt.strip()

        # User input prompt
        st.subheader("Additional Prompt")
        user_prompt = st.text_input("Enter additional prompt details:")

        # Combine prompts
        saved_prompt = st.session_state.get('saved_prompt', '')
        final_prompt = f"{saved_prompt}, {user_prompt}".strip()
        
        st.subheader("Final Prompt")
        final_prompt_area = st.text_area("Final prompt for image generation:", value=final_prompt, height=150, key="final_prompt")

        if st.button("Generate New Images"):
            with st.spinner("Generating new images..."):
                col1, col2 = st.columns(2)
                for i in range(4):
                    new_image = generate_new_image(final_prompt_area)
                    if i % 2 == 0:
                        with col1:
                            st.image(new_image, caption=f"Generated Image {i+1}", use_column_width=True)
                    else:
                        with col2:
                            st.image(new_image, caption=f"Generated Image {i+1}", use_column_width=True)

def image_editing():
    #st.header("Image Editing")

    # Background Removal
    st.subheader("Background Removal")
    bg_remove_file = st.file_uploader("Choose an image for background removal", type=["jpg", "jpeg", "png"])
    if bg_remove_file is not None:
        image = Image.open(bg_remove_file)
        st.image(image, caption="Original Image", use_column_width=True)
        if st.button("Remove Background"):
            result = remove_background(image)
            st.image(result, caption="Image with Background Removed", use_column_width=True)
            buf = io.BytesIO()
            result.save(buf, format="PNG")
            byte_im = buf.getvalue()
            st.download_button(label="Download Result", data=byte_im, file_name="result.png", mime="image/png")

    # Image Assembly
    st.subheader("Image Assembly")
    background_file = st.file_uploader("Choose a background image", type=['png', 'jpg', 'jpeg'])
    if background_file:
        background = load_image(background_file)
        background = resize_image(background, 800)
        st.image(background, caption="Background Image", use_column_width=True)

        uploaded_files = st.file_uploader("Choose foreground images", accept_multiple_files=True, type=['png', 'jpg', 'jpeg'])
        if uploaded_files:
            images = [load_image(file) for file in uploaded_files]
            cropped_images = []
            for i, img in enumerate(images):
                st.subheader(f"Image {i+1}")
                st.image(img, use_column_width=True)
                
                st.write(f"Crop image {i+1}")
                col1, col2 = st.columns(2)
                with col1:
                    left = st.slider(f"Left crop for image {i+1}", 0, img.width, 0)
                    right = st.slider(f"Right crop for image {i+1}", 0, img.width, img.width)
                with col2:
                    top = st.slider(f"Top crop for image {i+1}", 0, img.height, 0)
                    bottom = st.slider(f"Bottom crop for image {i+1}", 0, img.height, img.height)
                
                cropped_img = crop_image(img, left, top, right, bottom)
                resized_img = resize_image(cropped_img, 200)
                cropped_images.append(resized_img)
                st.image(resized_img, caption=f"Cropped and Resized Image {i+1}", use_column_width=True)

            if 'positions' not in st.session_state:
                st.session_state.positions = [(0, 0) for _ in cropped_images]
            if 'sizes' not in st.session_state:
                st.session_state.sizes = [img.size for img in cropped_images]

            if st.button("Drag, Resize, and Assemble Images"):
                positions, sizes, assembled_image = drag_and_resize_images(background, cropped_images, st.session_state.positions, st.session_state.sizes)
                st.session_state.positions = positions
                st.session_state.sizes = sizes
                st.image(assembled_image, caption="Assembled Image", use_column_width=True)

            if st.button("Finalize Assembly"):
                assembled_image = assemble_images(background, cropped_images, st.session_state.positions, st.session_state.sizes)
                st.image(assembled_image, caption="Final Assembled Image", use_column_width=True)
                
                buf = io.BytesIO()
                assembled_image.save(buf, format="PNG")
                byte_im = buf.getvalue()
                st.download_button(label="Download Assembled Image", data=byte_im, file_name="assembled_image.png", mime="image/png")

    # Text Overlay
    st.subheader("Text Overlay")
    text_overlay_file = st.file_uploader("Choose an image for text overlay", type=["jpg", "jpeg", "png"])
    if text_overlay_file is not None:
        image = Image.open(text_overlay_file)
        img_array = np.array(image)
        
        st.image(image, caption='Uploaded Image', use_column_width=True)
        
        texts = []
        num_texts = st.number_input("Number of text overlays", min_value=1, value=1)
        
        for i in range(num_texts):
            text = st.text_area(f"Enter text to overlay #{i+1} (multiple lines supported):")
            x_value = st.slider(f"X position for text #{i+1}", 0, img_array.shape[1], 50)
            y_value = st.slider(f"Y position for text #{i+1}", 0, img_array.shape[0], 50 + i*50)
            color = st.color_picker(f"Choose color for text #{i+1}", '#000000')
            color = tuple(int(color.lstrip('#')[i:i+2], 16) for i in (0, 2, 4))
            texts.append((text, (x_value, y_value), color))
        
        if st.button("Add Text and Drag"):
            img_with_text = img_array.copy()
            updated_texts, updated_img = drag_text(img_with_text, texts)
            result_image = Image.fromarray(updated_img)
            st.image(result_image, caption='Image with Dragged Text Overlay', use_column_width=True)
            
            buf = io.BytesIO()
            result_image.save(buf, format="PNG")
            st.download_button(label="Download Updated Image", data=buf.getvalue(), file_name="image_with_dragged_text.png", mime="image/png")

    # Text Replacement
    st.subheader("Text Replacement")
    text_replace_file = st.file_uploader("Choose an image for text replacement", type=["jpg", "jpeg", "png"])
    if text_replace_file is not None:
        image = Image.open(text_replace_file)
        st.image(image, caption='Current Image', use_column_width=True)

        text_results = detect_text(image)
        
        st.subheader("Detected Text:")
        
        for i, (text, box) in enumerate(text_results):
            if text.strip() and text not in st.session_state.get('replaced_texts', []):
                st.text(f"{i+1}. {text}")
                
                new_text = st.text_input(f"Enter new text for '{text}':", value=text, key=f"new_text_{i}")
                new_color = st.color_picker(f"Choose color for new text '{new_text}':", '#000000', key=f"color_{i}")
                
                if st.button(f"Replace '{text}'", key=f"replace_{i}"):
                    st.session_state.edited_image = replace_text_in_image(image, text, new_text, new_color)
                    if 'replaced_texts' not in st.session_state:
                        st.session_state.replaced_texts = []
                    st.session_state.replaced_texts.append(text)
                    st.image(st.session_state.edited_image, caption='Edited Image', use_column_width=True)
                    text_results[i] = (new_text, box)
        
        if hasattr(st.session_state, 'edited_image') and st.session_state.edited_image is not None:
            buf = io.BytesIO()
            st.session_state.edited_image.save(buf, format="PNG")
            byte_im = buf.getvalue()
            st.download_button(label="Download edited image", data=byte_im, file_name="edited_image.png", mime="image/png")

if __name__ == "__main__":
    main()