app.py

# import part
import streamlit as st
from transformers import pipeline
import torch
from PIL import Image
import io
import os
from huggingface_hub import InferenceClient
import numpy as np
import base64

# function part
# img2text
def img2text(image_path):
    image_to_text = pipeline("image-to-text", model="noamrot/FuseCap_Image_Captioning")
    text = image_to_text(image_path)[0]["generated_text"]
    return text

# text2story
def text2story(text):
    # Using Llama model through API to avoid GGUF format complexities in Streamlit
    client = InferenceClient(model="MaziyarPanahi/Llama-3.2-1B-Instruct-GGUF")
    
    # Create a prompt for the story generation
    prompt = f"""Write a fun, engaging children's story of about 100 words based on this caption: 
    "{text}"
    
    The story should be suitable for kids aged 3-10 years old, with simple language, positive themes, and a clear beginning, middle, and end.
    """
    
    # Generate the story
    story_text = client.text_generation(
        prompt,
        max_new_tokens=250,
        temperature=0.7,
        top_p=0.9,
        repetition_penalty=1.2
    )
    
    return story_text

# text2audio
def text2audio(story_text):
    # Using Bark text-to-speech model
    tts = pipeline("text-to-speech", model="suno/bark")
    
    # Generate audio with a voice suitable for children's stories
    audio_output = tts(
        text=story_text,
        forward_params={"speaker": "v2/en_speaker_6", "text_temp": 0.7}
    )
    
    return {
        "audio": audio_output["audio"],
        "sampling_rate": audio_output["sampling_rate"]
    }

# Function to save temporary image file
def save_uploaded_image(uploaded_file):
    # Create a temp directory if it doesn't exist
    if not os.path.exists("temp"):
        os.makedirs("temp")
    
    # Define the path to save the image
    image_path = os.path.join("temp", uploaded_file.name)
    
    # Save the image
    with open(image_path, "wb") as f:
        f.write(uploaded_file.getvalue())
    
    return image_path

# main part
st.set_page_config(
    page_title="Kids Storytelling Magic",
    page_icon="📚",
    layout="centered"
)

# Add some CSS for a child-friendly interface
st.markdown("""
    <style>
    .main {
        background-color: #f0f8ff;
    }
    h1, h2, h3 {
        color: #1e90ff;
    }
    .stButton>button {
        background-color: #ff6b6b;
        color: white;
        font-size: 1.2rem;
        border-radius: 10px;
        padding: 0.5rem 1rem;
    }
    </style>
    """, unsafe_allow_html=True)

st.title("🧸 Kids Storytelling Magic 🦄")
st.subheader("Upload a picture and hear a magical story!")

uploaded_file = st.file_uploader("Choose a fun picture...", type=["jpg", "jpeg", "png"])

if uploaded_file is not None:
    # Display a loading spinner
    with st.spinner("Working on your magical story..."):
        # Display the uploaded image
        st.image(uploaded_file, caption="Your magical picture", use_column_width=True)
        
        # Save the image temporarily
        image_path = save_uploaded_image(uploaded_file)
        
        # Stage 1: Image to Text
        with st.spinner("Looking at your picture..."):
            caption = img2text(image_path)
            st.markdown("### 📝 I see...")
            st.write(caption)
        
        # Stage 2: Text to Story
        with st.spinner("Creating your story..."):
            story = text2story(caption)
            st.markdown("### 📖 Your Story")
            st.write(story)
        
        # Stage 3: Story to Audio data
        with st.spinner("Making your story speak..."):
            try:
                audio_data = text2audio(story)
                
                # Add a play button with cute icon
                st.markdown("### 🔊 Listen to your story")
                if st.button("🎵 Play Story"):
                    st.audio(
                        audio_data["audio"],
                        format="audio/wav",
                        start_time=0,
                        sample_rate=audio_data["sampling_rate"]
                    )
            except Exception as e:
                st.error(f"Oops! Something went wrong with the audio: {str(e)}")
                st.write("But you can still read the story above!")
        
        # Clean up - delete the temporary image
        try:
            os.remove(image_path)
        except:
            pass

else:
    # Show instructions with a friendly message
    st.markdown("""
    ### How to use:
    1. Click the button above to upload a picture
    2. Wait for the magical story to appear
    3. Press play to hear your story!
    
    Try pictures of animals, nature, toys, or anything fun!
    """)
    
    # Show a placeholder image
    st.image("https://placehold.co/600x400/9370db/ffffff?text=Upload+an+image+to+start+the+magic!", 
             caption="Ready for your picture!", use_column_width=True)

# Add a footer
st.markdown("---")
st.markdown("Made for kids to enjoy the stories")