app.py

import gradio as gr
from transformers import CLIPProcessor, CLIPModel
from PIL import Image
import torch

# Load a pre-trained CLIP model and processor from Hugging Face
model_name = "openai/clip-vit-base-patch32"
model = CLIPModel.from_pretrained(model_name)
processor = CLIPProcessor.from_pretrained(model_name)

# Define possible personas mapped to interests
predefined_personas = {
    "Fashion": {
        "Age": "18-30 years",
        "Gender": "Likely Female",
        "Interests": ["Fashion", "Trendy Clothing", "Social Media"],
        "Income Level": "Medium",
        "Psychographics": ["Fashion-conscious", "Socially active", "Trend-seeker"],
        "Behavioral Traits": ["Follows fashion influencers", "Buys seasonal items", "Active on Instagram"]
    },
    "Technology": {
        "Age": "20-40 years",
        "Gender": "Likely Male",
        "Interests": ["Technology", "Gadgets", "Software"],
        "Income Level": "High",
        "Psychographics": ["Tech-savvy", "Innovative", "Early adopter"],
        "Behavioral Traits": ["Follows tech blogs", "Interested in product launches", "Buys online frequently"]
    },
    "Sports": {
        "Age": "15-40 years",
        "Gender": "Likely Male",
        "Interests": ["Sports", "Fitness", "Outdoor Activities"],
        "Income Level": "Medium",
        "Psychographics": ["Active lifestyle", "Health-conscious", "Competitive"],
        "Behavioral Traits": ["Watches sports events", "Goes to the gym", "Buys athletic gear"]
    },
    "Travel": {
        "Age": "25-45 years",
        "Gender": "Likely Female or Male",
        "Interests": ["Travel", "Adventure", "Culture"],
        "Income Level": "Medium to High",
        "Psychographics": ["Experience-seeker", "Open-minded", "Curious"],
        "Behavioral Traits": ["Follows travel blogs", "Likes outdoor activities", "Prefers unique experiences"]
    },
    "Luxury": {
        "Age": "30-50 years",
        "Gender": "Likely Female or Male",
        "Interests": ["Luxury Goods", "High-end Fashion", "Fine Dining"],
        "Income Level": "High",
        "Psychographics": ["Status-conscious", "Exclusive-taste", "Prestige-oriented"],
        "Behavioral Traits": ["Buys luxury brands", "Frequent traveler", "Invests in high-end products"]
    },
    "Fitness": {
        "Age": "20-40 years",
        "Gender": "Likely Female or Male",
        "Interests": ["Fitness", "Healthy Living", "Nutrition"],
        "Income Level": "Medium",
        "Psychographics": ["Health-conscious", "Goal-oriented", "Disciplined"],
        "Behavioral Traits": ["Goes to the gym regularly", "Buys fitness equipment", "Follows nutrition plans"]
    },
    "Education": {
        "Age": "18-35 years",
        "Gender": "Likely Female or Male",
        "Interests": ["Education", "Learning", "Career Development"],
        "Income Level": "Medium",
        "Psychographics": ["Growth-oriented", "Knowledge-seeker", "Curious"],
        "Behavioral Traits": ["Follows online courses", "Interested in higher education", "Learns new skills"]
    },
    "Entertainment": {
        "Age": "15-35 years",
        "Gender": "Likely Female or Male",
        "Interests": ["Movies", "Music", "Video Games"],
        "Income Level": "Medium",
        "Psychographics": ["Fun-seeking", "Relaxed", "Entertainment-driven"],
        "Behavioral Traits": ["Watches Netflix", "Plays video games", "Follows pop culture trends"]
    }
}

# List of interest categories to compare against
interest_categories = list(predefined_personas.keys())

# Function to calculate similarity between the image and each interest category
def get_most_similar_persona(image):
    """Infers the most similar predefined persona for the given image."""
    inputs = processor(images=image, return_tensors="pt")
    with torch.no_grad():
        image_features = model.get_image_features(**inputs)
    
    # Define descriptions of interest categories (texts)
    category_descriptions = [
        f"A photo that represents {category.lower()}." for category in interest_categories
    ]
    
    # Tokenize the category descriptions and get text features
    text_inputs = processor(text=category_descriptions, return_tensors="pt", padding=True)
    text_features = model.get_text_features(**text_inputs)

    # Calculate similarity between image and text features
    image_features /= image_features.norm(dim=-1, keepdim=True)
    text_features /= text_features.norm(dim=-1, keepdim=True)
    similarities = torch.matmul(image_features, text_features.T)

    # Find the most similar category
    most_similar_idx = similarities.argmax().item()
    most_similar_category = interest_categories[most_similar_idx]
    
    return predefined_personas[most_similar_category]

# Gradio interface function
def persona_analysis(image):
    """Generates a marketing persona based on the image."""
    persona = get_most_similar_persona(image)
    return "\n".join([f"{key}: {value}" for key, value in persona.items()])

# Build the Gradio interface
iface = gr.Interface(
    fn=persona_analysis,
    inputs=gr.Image(type="pil"),  # Accept image input in PIL format
    outputs=gr.Textbox(),  # Display persona as text output
    title="Marketing Persona Generator",
    description="Upload an image to generate a marketing persona based on the ad subject."
)

# Launch the Gradio interface
iface.launch()