app.py

import os
import streamlit as st
import requests
from transformers import pipeline
from typing import Dict
from together import Together

# Image-to-text
def img2txt(url: str) -> str:
    print("Initializing captioning model...")
    captioning_model = pipeline("image-to-text", model="Salesforce/blip-image-captioning-base")
    
    print("Generating text from the image...")
    text = captioning_model(url, max_new_tokens=20)[0]["generated_text"]
    
    print(text)
    return text

# Text-to-story generation with LLM model
def txt2story(prompt: str, top_k: int, top_p: float, temperature: float) -> str:
    # Load the Together API client
    client = Together(api_key=os.environ.get("TOGETHER_API_KEY"))

    # Modify the prompt based on user inputs and ensure a 250-word limit
    story_prompt = f"Write a short story of no more than 250 words based on the following prompt: {prompt}"

    # Call the LLM model
    stream = client.chat.completions.create(
        model="meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo",
        messages=[
            {"role": "system", "content": '''As an experienced short story writer, write a meaningful story influenced by the provided prompt.
            Ensure the story does not exceed 250 words.'''},
            {"role": "user", "content": story_prompt}
        ],
        top_k=top_k,
        top_p=top_p,
        temperature=temperature,
        stream=True
    )

    # Concatenate story chunks
    story = ''
    for chunk in stream:
        story += chunk.choices[0].delta.content

    return story

# Text-to-speech
def txt2speech(text: str) -> None:
    print("Initializing text-to-speech conversion...")
    API_URL = "https://api-inference.huggingface.co/models/espnet/kan-bayashi_ljspeech_vits"
    headers = {"Authorization": f"Bearer {os.environ['HUGGINGFACEHUB_API_TOKEN']}"}
    payloads = {'inputs': text}
    
    response = requests.post(API_URL, headers=headers, json=payloads)
    
    with open('audio_story.mp3', 'wb') as file:
        file.write(response.content)

# Get user preferences for the story
def get_user_preferences() -> Dict[str, str]:
    preferences = {}

    preferences['continent'] = st.selectbox("Continent", ["North America", "Europe", "Asia", "Africa", "Australia"])
    preferences['genre'] = st.selectbox("Genre", ["Science Fiction", "Fantasy", "Mystery", "Romance"])
    preferences['setting'] = st.selectbox("Setting", ["Future", "Medieval times", "Modern day", "Alternate reality"])
    preferences['plot'] = st.selectbox("Plot", ["Hero's journey", "Solving a mystery", "Love story", "Survival"])
    preferences['tone'] = st.selectbox("Tone", ["Serious", "Light-hearted", "Humorous", "Dark"])
    preferences['theme'] = st.selectbox("Theme", ["Self-discovery", "Redemption", "Love", "Justice"])
    preferences['conflict'] = st.selectbox("Conflict Type", ["Person vs. Society", "Internal struggle", "Person vs. Nature", "Person vs. Person"])
    preferences['twist'] = st.selectbox("Mystery/Twist", ["Plot twist", "Hidden identity", "Unexpected ally/enemy", "Time paradox"])
    preferences['ending'] = st.selectbox("Ending", ["Happy", "Bittersweet", "Open-ended", "Tragic"])
    
    return preferences

# Main function
def main():
    st.set_page_config(page_title="🎨 Image-to-Audio Story 🎧", page_icon="🖼️")
    st.title("Turn the Image into Audio Story")

    # Allows users to upload an image file
    uploaded_file = st.file_uploader("# 📷 Upload an image...", type=["jpg", "jpeg", "png"])

    # Parameters for LLM model (in the sidebar)
    st.sidebar.markdown("# LLM Inference Configuration Parameters")
    top_k = st.sidebar.number_input("Top-K", min_value=1, max_value=100, value=5)
    top_p = st.sidebar.number_input("Top-P", min_value=0.0, max_value=1.0, value=0.8)
    temperature = st.sidebar.number_input("Temperature", min_value=0.1, max_value=2.0, value=1.5)

    # Get user preferences for the story
    st.markdown("## Story Preferences")
    preferences = get_user_preferences()

    if uploaded_file is not None:
        # Reads and saves uploaded image file
        bytes_data = uploaded_file.read()
        with open("uploaded_image.jpg", "wb") as file:
            file.write(bytes_data)

        st.image(uploaded_file, caption='🖼️ Uploaded Image', use_column_width=True)

        # Initiates AI processing and story generation
        with st.spinner("## 🤖 AI is at Work! "):
            scenario = img2txt("uploaded_image.jpg")  # Extracts text from the image
            
            # Modify the prompt to include user preferences
            prompt = f"Based on the image description: '{scenario}', create a {preferences['genre']} story set in {preferences['setting']} in {preferences['continent']}. " \
                     f"The story should have a {preferences['tone']} tone and explore the theme of {preferences['theme']}. " \
                     f"The main conflict should be {preferences['conflict']}. " \
                     f"The story should have a {preferences['twist']} and end with a {preferences['ending']} ending."
            
            story = txt2story(prompt, top_k, top_p, temperature)  # Generates a story based on the image text, LLM params, and user preferences
            
            txt2speech(story)  # Converts the story to audio

            st.markdown("---")
            st.markdown("## 📜 Image Caption")
            st.write(scenario)

            st.markdown("---")
            st.markdown("## 📖 Story")
            st.write(story)

            st.markdown("---")
            st.markdown("## 🎧 Audio Story")
            st.audio("audio_story.mp3")

if __name__ == '__main__':
    main()

# Credits
st.markdown("### Credits")
st.caption('''
            Made with ❤️ by @Aditya-Neural-Net-Ninja\n 
            Utilizes Image-to-Text, Text Generation, Text-to-Speech Transformer Models\n
            Gratitude to Streamlit, 🤗 Spaces for Deployment & Hosting
            ''')