import gradio as gr
from transformers import pipeline

# pipeline 1: zero-shot to identify movie genres
genre_classifier = pipeline("zero-shot-classification", model="facebook/bart-large-mnli")

# pipeline 2: text2text-generation for movie descriptions
desc_pipeline = pipeline("text2text-generation", model="google/flan-t5-base")

# fine-tuned model (pipeline 3): local movie recommender model
rec_pipeline = pipeline("text-classification", model="model")

candidate_movies = [
    "Inception", "The Matrix", "Interstellar", "Titanic", "The Dark Knight",
    "The Godfather", "Pulp Fiction", "The Shawshank Redemption", "Forrest Gump", "Avengers"
]

movie_genres = ["sci-fi", "drama", "romance", "action", "crime", "thriller", "adventure"]

def recommend_movies(input_movies):
    if not input_movies.strip():
        return "⚠️ Please enter at least one movie.", []

    genres = genre_classifier(input_movies, candidate_labels=movie_genres, multi_label=True)
    top_genres = genres["labels"][:2]

    # Use fine-tuned model to recommend a movie label (simplified)
    try:
        rec_results = rec_pipeline(f"{input_movies} | genres: {', '.join(top_genres)}")
    except Exception as e:
        return f"❌ Recommendation model error: {e}", []

    top_rec = rec_results[:5]
    gallery = []
    for item in top_rec:
        title = item["label"].replace("LABEL_", "").strip()
        score = item["score"]
        try:
            desc = desc_pipeline(f"Describe the movie {title} in one sentence.", max_length=40)[0]["generated_text"]
        except:
            desc = "No description available."
        img_url = f"https://via.placeholder.com/200x300?text={title.replace(' ', '+')}"
        label = f"🎞️ **{title}**\n\n{desc}\n\nConfidence: {score:.2f}"
        gallery.append((img_url, label))

    summary = f"🎬 Based on your input and detected genres ({', '.join(top_genres)}), we recommend:"
    return summary, gallery

with gr.Blocks(title="🎥 Movie Recommender") as demo:
    gr.Markdown("# 🎬 Personalized Movie Recommendation\n_Using Hugging Face pipelines + fine-tuned model_")
    with gr.Row():
        input_box = gr.Textbox(label="Enter up to 3 movies you like", placeholder="e.g. Inception, Titanic")
        btn = gr.Button("🎯 Recommend")
    output_text = gr.Markdown()
    output_gallery = gr.Gallery(columns=2)
    btn.click(fn=recommend_movies, inputs=input_box, outputs=[output_text, output_gallery])

demo.launch()