app.py

import gradio as gr
from transformers import CLIPProcessor, CLIPModel
import torch
from PIL import Image
from io import BytesIO
import requests
import numpy as np
from sklearn.cluster import KMeans

# Load CLIP
model = CLIPModel.from_pretrained("openai/clip-vit-base-patch16")
processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch16")
device = "cpu"
model.to(device)

PROMPT_CATEGORIES = [
    "Anime & Manga", "TV Shows & Movies", "Video Games", "Cartoons & Animated Characters",
    "Pop Culture & Music", "K-Pop & Idol Groups", "Celebrities & Influencers",
    "Floral & Botanical", "Scenery & Landscapes", "Abstract & Minimalist",
    "Cats & Dogs", "Wildlife & Exotic Animals", "Fantasy Creatures",
    "Football & Basketball", "Extreme Sports", "Fitness & Gym",
    "Motivational & Inspirational", "Funny & Meme-Based", "Dark & Gothic",
    "Cyberpunk & Sci-Fi", "Glitch & Vaporwave", "AI & Robotics",
    "Flags & National Pride", "Traditional Art", "Astrology & Zodiac Signs"
]

CATEGORIES = [
    "Anime & Manga", "TV Shows & Movies", "Video Games", "Cartoons & Animated Characters",
    "Pop Culture & Music", "K-Pop & Idol Groups", "Celebrities & Influencers",
    "Floral & Botanical", "Scenery & Landscapes", "Abstract & Minimalist",
    "Cats & Dogs", "Wildlife & Exotic Animals", "Fantasy Creatures",
    "Football & Basketball", "Extreme Sports", "Fitness & Gym",
    "Motivational & Inspirational", "Funny & Meme-Based", "Dark & Gothic",
    "Cyberpunk & Sci-Fi", "Glitch & Vaporwave", "AI & Robotics",
    "Flags & National Pride", "Traditional Art", "Astrology & Zodiac Signs"
]

TARGET_COLORS = {
    'Red': (255, 107, 107),
    'Teal': (78, 205, 196),
    'Blue': (69, 183, 209),
    'Green': (150, 206, 180),
    'Yellow': (255, 234, 167),
    'Orange': (225, 112, 85),
    'Black': (45, 52, 54),
    'Gray': (223, 230, 233),
    'Pink': (253, 121, 168),
    'Purple': (155, 89, 182),
    'Light Purple': (200, 162, 200),
    'Dark Blue': (52, 73, 94),
    'Beige': (245, 222, 179),
    'Brown': (139, 69, 19),
    'White': (255, 255, 255),
}

def closest_palette_color(rgb):
    min_diff = float('inf')
    closest_name = "Unknown"
    for name, ref_rgb in TARGET_COLORS.items():
        diff = sum((comp1 - comp2) ** 2 for comp1, comp2 in zip(rgb, ref_rgb))
        if diff < min_diff:
            min_diff = diff
            closest_name = name
    return closest_name

def get_dominant_colors(img, num_colors=3):
    img = img.resize((128, 128))
    np_pixels = np.array(img).reshape(-1, 3)
    filtered = [p for p in np_pixels if not (
        np.all(p > [240, 240, 240]) or np.all(p < [15, 15, 15])
    )]
    if not filtered:
        return ["Unknown"]
    filtered = np.array(filtered)
    kmeans = KMeans(n_clusters=5, random_state=42, n_init=10)
    kmeans.fit(filtered)
    centers = kmeans.cluster_centers_
    mapped_colors = [closest_palette_color(tuple(map(int, center))) for center in centers]
    unique_colors = []
    for color in mapped_colors:
        if color not in unique_colors and color != "Unknown":
            unique_colors.append(color)
        if len(unique_colors) == num_colors:
            break
    return unique_colors or ["Unknown"]

def classify_and_tag(image_url):
    try:
        response = requests.get(image_url)
        image = Image.open(BytesIO(response.content)).convert("RGB")
    except Exception:
        return "", "", "", "Error: Could not load image."

    # Classification
    inputs = processor(text=PROMPT_CATEGORIES, images=image, return_tensors="pt", padding=True).to(device)
    outputs = model(**inputs)
    probs = outputs.logits_per_image.softmax(dim=1)
    top_idx = torch.argmax(probs).item()
    category = CATEGORIES[top_idx]  # ✅ plain text, not hashtag

    # Colors
    colors = get_dominant_colors(image)
    color1 = colors[0] if len(colors) > 0 else ""
    color2 = colors[1] if len(colors) > 1 else ""
    color3 = colors[2] if len(colors) > 2 else ""

    # ✅ Return category + 3 color names
    return category, color1, color2, color3

demo = gr.Interface(
    fn=classify_and_tag,
    inputs=gr.Text(label="Image URL"),
    outputs=[
        gr.Text(label="Category"),  # <== plain text
        gr.Text(label="Color 1"),
        gr.Text(label="Color 2"),
        gr.Text(label="Color 3"),
    ],
    title="🎨 ArtCase Design Analyzer",
    description="Paste an image URL to get category and top 3 colors."
)

demo.launch()