app.py

import os
import streamlit as st
import requests
from transformers import pipeline
import openai
from langchain import LLMChain, PromptTemplate
from langchain import HuggingFaceHub

# Suppressing all warnings
import warnings
warnings.filterwarnings("ignore")

api_token = os.getenv('H_TOKEN')

# Image-to-text
def img2txt(url):
    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

model = "tiiuae/falcon-7b-instruct"
llm = HuggingFaceHub(
    huggingfacehub_api_token = api_token,
    repo_id = model,
    verbose = False,
    model_kwargs = {"temperature":0.2, "max_new_tokens": 4000})

def generate_story(scenario, llm):
  template= """You are a story teller.
               You get a scenario as an input text, and generates a short story out of it.
               Context: {scenario}
               Story:
               """
  prompt = PromptTemplate(template=template, input_variables=["scenario"])
  #Let's create our LLM chain now
  chain = LLMChain(prompt=prompt, llm=llm)
  story = chain.predict(scenario=scenario)
  start_index = story.find("Story:") + len("Story:")

  # Extract the text after "Story:"
  story = story[start_index:].strip()
  return story


# Text-to-speech
def txt2speech(text):
    print("Initializing text-to-speech conversion...")
    API_URL = "https://api-inference.huggingface.co/models/espnet/kan-bayashi_ljspeech_vits"
    headers = {"Authorization": f"Bearer {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)
        

        
# Streamlit web app 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)

    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
            story = generate_story(scenario, llm)  # Generates a story based on the image text, LLM params
            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()