app.py

# story_generator.py
import re
import time
import tempfile
import streamlit as st
from PIL import Image
from gtts import gTTS
from transformers import pipeline

# --- Initialize Streamlit Config ---
st.set_page_config(
    page_title="Magic Story Generator",
    layout="centered",
    page_icon="📖"
)

# --- Model Loading (Cached) ---
@st.cache_resource(show_spinner=False)
def load_models():
    # Image captioning model
    captioner = pipeline(
        "image-to-text",
        model="Salesforce/blip-image-captioning-base",
        device=-1
    )
    
    # Story generation model with optimized settings
    storyteller = pipeline(
        "text-generation",
        model="Qwen/Qwen3-1.7B",
        device_map="auto",
        trust_remote_code=True,
        torch_dtype="auto",
        model_kwargs={
            "revision": "main",
            "temperature": 0.7,
            "top_p": 0.9,
            "repetition_penalty": 1.1,
            "pad_token_id": 151645
        }
    )
    
    return captioner, storyteller

# --- Text Processing Utilities ---
def clean_generated_text(raw_text):
    # Split at first assistant marker
    clean_text = raw_text.split("<|im_start|>assistant\n", 1)[-1]
    
    # Remove any subsequent chat turns
    clean_text = clean_text.split("<|im_start|>")[0]
    
    # Remove special tokens and whitespace
    clean_text = clean_text.replace("<|im_end|>", "").strip()
    
    # Regex cleanup for remaining markers
    clean_text = re.sub(
        r'^(assistant[\s\-\:>]*)+',
        '',
        clean_text,
        flags=re.IGNORECASE
    ).strip()
    
    # Format punctuation and capitalization
    sentences = []
    for sent in re.split(r'(?<=[.!?]) +', clean_text):
        sent = sent.strip()
        if not sent:
            continue
        if sent[-1] not in {'.', '!', '?'}:
            sent += '.'
        sentences.append(sent[0].upper() + sent[1:])
    
    return ' '.join(sentences)

# --- Main Application UI ---
st.title("📖✨ Magic Story Generator")

uploaded_image = st.file_uploader(
    "Upload a children's book style image:",
    type=["jpg", "jpeg", "png"]
)

if uploaded_image:
    # Display uploaded image
    image = Image.open(uploaded_image).convert("RGB")
    st.image(image, use_column_width=True)
    
    # Load models only when needed
    caption_pipe, story_pipe = load_models()
    
    # Generate image caption
    with st.spinner("🔍 Analyzing image..."):
        try:
            caption_result = caption_pipe(image)
            image_caption = caption_result[0].get("generated_text", "").strip()
            
            if not image_caption:
                raise ValueError("Couldn't generate caption")
                
            st.success(f"**Image Understanding:** {image_caption}")
        except Exception as e:
            st.error("❌ Failed to analyze image. Please try another.")
            st.stop()
    
    # Create story prompt
    story_prompt = (
        f"<|im_start|>system\n"
        f"You are a children's book author. Create a 150-word story based on: {image_caption}\n"
        "Include these elements:\n"
        "- Friendly characters\n"
        "- Simple vocabulary\n"
        "- Positive lesson\n"
        "- Clear story structure\n"
        "<|im_end|>\n"
        f"<|im_start|>user\n"
        f"Write an engaging story suitable for ages 6-8.<|im_end|>\n"
        f"<|im_start|>assistant\n"
    )
    
    # Generate story text
    with st.spinner("📝 Crafting magical story..."):
        try:
            story_result = story_pipe(
                story_prompt,
                max_new_tokens=300,
                do_sample=True,
                num_return_sequences=1
            )
            raw_story = story_result[0]['generated_text']
        except Exception as e:
            st.error("❌ Story generation failed. Please try again.")
            st.stop()
    
    # Process and display story
    final_story = clean_generated_text(raw_story)
    
    st.subheader("✨ Your Story")
    st.write(final_story)
    
    # Generate audio version
    with st.spinner("🔊 Creating audio version..."):
        try:
            tts = gTTS(text=final_story, lang='en', slow=False)
            with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as fp:
                tts.save(fp.name)
                st.audio(fp.read(), format="audio/mp3")
        except Exception as e:
            st.warning("⚠️ Audio conversion failed. Text version still available.")

# Footer
st.markdown("---")
st.caption("Made with ♥ by The Story Wizard • [Report Issues](https://example.com)")