import streamlit as st
from transformers import pipeline
from huggingface_hub import InferenceClient
from PIL import Image
import os

def setup_session():
    if 'app_ready' not in st.session_state:
        print("Powering up the Dragon Radar...")
        st.session_state['app_ready'] = True
        st.session_state['hf_token'] = os.getenv("HUGGINGFACE_TOKEN")
        st.session_state['client'] = InferenceClient(api_key=st.session_state['hf_token'])

def main():
    setup_session()
    
    st.header("Anime & Friends Image Commentary")
    st.write("Let your favorite characters react to any image!")
    
    character = st.selectbox(
        "Select your commentator", 
        ["goku", "elmo", "kirby", "pikachu"]
    )
    
    uploaded_img = st.file_uploader("Share your image!")
    
    if uploaded_img is not None:
        image = Image.open(uploaded_img)
        st.image(image)
        
        
        caption_model = pipeline("image-to-text", model="Salesforce/blip-image-captioning-large")
        base_caption = caption_model(image)[0]['generated_text']
        
        
        character_reactions = {
            "goku": f"Describe this image like you're Goku from Dragon Ball Z, mentioning power levels: {base_caption}",
            "elmo": f"Describe this image like you're Elmo from Sesame Street, speaking in third person: {base_caption}",
            "kirby": f"Describe this image like you're Kirby, being cute and mentioning food: {base_caption}",
            "pikachu": f"Describe this image like you're Pikachu, using 'pika' frequently: {base_caption}"
        }
        
        messages = [
            {
                "role": "user",
                "content": character_reactions[character]
            }
        ]
        
        # Generate character response using Llama
        response_stream = st.session_state['client'].chat.completions.create(
            model="meta-llama/Llama-3.2-3B-Instruct",
            messages=messages,
            max_tokens=500,
            stream=True
        )
        
        character_response = ''
        for chunk in response_stream:
            character_response += chunk.choices[0].delta.content
        
        st.write(character_response)

if __name__ == '__main__':
    main()