app.py

import gradio as gr
from transformers import pipeline
from PIL import ImageDraw, ImageFont
import textwrap

# --- LOAD MODELS ---
print("Loading Models...")
caption_pipeline = pipeline("image-to-text", model="Salesforce/blip-image-captioning-base")
classification_pipeline = pipeline("image-classification", model="google/vit-base-patch16-224")
sentiment_pipeline = pipeline("sentiment-analysis")

# --- DRAWING FUNCTION ---
def add_caption_to_image(image, text):
    draw = ImageDraw.Draw(image)
    image_width, image_height = image.size
    
    # 1. Setup Font
    try:
        font = ImageFont.truetype("DejaVuSans.ttf", 20)
    except IOError:
        font = ImageFont.load_default()
    
    # 2. Wrap Text
    avg_char_width = 12 
    chars_per_line = max(10, int((image_width - 40) / avg_char_width))
    lines = textwrap.wrap(text, width=chars_per_line)
    
    # 3. Calculate Box Size
    line_height = 24
    total_text_height = len(lines) * line_height
    y_start = image_height - total_text_height - 20
    
    max_line_width = 0
    for line in lines:
        bbox = draw.textbbox((0, 0), line, font=font)
        w = bbox[2] - bbox[0]
        if w > max_line_width: max_line_width = w
    
    box_x = (image_width - max_line_width) / 2
    
    # 4. Draw Box
    padding = 10
    draw.rectangle(
        [
            (box_x - padding, y_start - padding), 
            (box_x + max_line_width + padding, y_start + total_text_height + padding)
        ],
        fill=(0, 0, 0, 180) 
    )
    
    # 5. Draw Text
    current_y = y_start
    for line in lines:
        bbox = draw.textbbox((0, 0), line, font=font)
        line_width = bbox[2] - bbox[0]
        line_x = (image_width - line_width) / 2
        draw.text((line_x, current_y), line, font=font, fill="white")
        current_y += line_height 
        
    return image

# --- ANALYSIS FUNCTION ---
def multimodal_analysis(input_image):
    if input_image is None: return None, "Upload image first", "N/A"
    
    processed_image = input_image.copy()
    
    # 1. Caption
    try:
        caption = caption_pipeline(input_image)[0]['generated_text']
    except:
        return processed_image, "Error", "Error"
        
    # 2. Draw 
    final_img = add_caption_to_image(processed_image, caption)
    
    # 3. Classify
    try:
        res = classification_pipeline(input_image)
        cls_str = f"{res[0]['label']} ({res[0]['score']:.2f})"
    except:
        cls_str = "Error"
        
    # 4. Sentiment
    try:
        sent = sentiment_pipeline(caption)[0]['label']
    except:
        sent = "Error"
        
    return final_img, cls_str, sent

# --- INTERFACE (Removed Theme to fix crash) ---
with gr.Blocks() as demo:
    gr.Markdown("# 🤖 Multimodal AI Analyst")
    gr.Markdown("Select an example image below to see: **Image Captioning**, **Vision Classification**, and **NLP Sentiment Analysis** working together.")
    
    with gr.Row():
        with gr.Column():
            image_input = gr.Image(type="pil", label="Input Image")
            submit_btn = gr.Button("🔍 Analyze Image", variant="primary")
        
        with gr.Column():
            output_image = gr.Image(label="AI Caption Result")
            with gr.Row():
                output_class = gr.Textbox(label="Object Class")
                output_sent = gr.Textbox(label="Caption Sentiment")

    # EXACT FILES FROM YOUR LIST
    examples = [
        ["Ashe Catcum with Pikachu.png"],
        ["Beautiful sunrise over ocean.png"],
        ["Cat on a couch.png"],
        ["Female Crying.png"],
        ["Lions Football team huddle.png"],
        ["michael jordan trophy.png"],
        ["Puppies playing in grass.png"],
        ["Red Ferrari.png"],
        ["Siamese cat.png"],
        ["Stormy dark sky lightning.png"]
    ]
    
    gr.Examples(examples=examples, inputs=image_input)
    submit_btn.click(fn=multimodal_analysis, inputs=image_input, outputs=[output_image, output_class, output_sent])

demo.launch()