import cv2
import numpy as np
from PIL import Image
import tempfile
import os

def process_image(image) -> str:
    """Basic image description"""
    try:
        captioner = ModelLoader().load_image_captioner()  # singleton-like
        result = captioner(image)
        return result[0]['generated_text']
    except:
        return "Image processed. Describe what you see or ask a question about it."

def process_video(video_path) -> str:
    """Extract key frames and describe"""
    cap = cv2.VideoCapture(video_path)
    frames_desc = []
    frame_count = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
    step = max(1, frame_count // 5)  # ~5 key frames

    for i in range(0, frame_count, step):
        cap.set(cv2.CAP_PROP_POS_FRAMES, i)
        ret, frame = cap.read()
        if ret:
            pil_img = Image.fromarray(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))
            desc = process_image(pil_img)
            frames_desc.append(f"Frame {i//step + 1}: {desc}")
    
    cap.release()
    return "\n".join(frames_desc) + "\n\nVideo summarized from key frames."

def generate_references(query: str) -> str:
    """Fake realistic references (static or rule-based)"""
    refs = [
        f"https://en.wikipedia.org/wiki/{query.replace(' ', '_')}",
        "https://arxiv.org/search/?query=" + query.replace(" ", "+"),
        "https://scholar.google.com/scholar?q=" + query.replace(" ", "+")
    ]
    return "\n".join([f"- [{r.split('//')[1].split('/')[0]}]({r})" for r in refs])