import streamlit as st
from transformers import pipeline
from PIL import Image
import io
from gtts import gTTS

st.title("🖼️ → 📖 Image-to-Story Demo")
st.write("Upload an image and watch as it’s captioned, turned into a short story, and even read aloud!")

@st.cache_resource
def load_captioner():
    return pipeline("image-captioning", model="nlpconnect/vit-gpt2-image-captioning")

@st.cache_resource
def load_story_gen():
    return pipeline("text-generation", model="gpt2", tokenizer="gpt2")

captioner = load_captioner()
story_gen = load_story_gen()

# 1) Upload
uploaded = st.file_uploader("Upload an image", type=["png","jpg","jpeg"], key="image")
if uploaded:
    img = Image.open(uploaded)
    st.image(img, use_column_width=True)

    # 2) Caption
    if "caption" not in st.session_state:
        with st.spinner("Generating caption…"):
            st.session_state.caption = captioner(img)[0]["generated_text"]
    st.write("**Caption:**", st.session_state.caption)

    # 3) Story
    if "story" not in st.session_state:
        with st.spinner("Spinning up a story…"):
            out = story_gen(
                st.session_state.caption,
                max_length=200,
                num_return_sequences=1,
                do_sample=True,
                top_p=0.9
            )
            st.session_state.story = out[0]["generated_text"]
    st.write("**Story:**", st.session_state.story)

    # 4) Pre-generate raw MP3 bytes
    if "audio_bytes" not in st.session_state:
        with st.spinner("Generating audio…"):
            tts = gTTS(text=st.session_state.story, lang="en")
            buf = io.BytesIO()
            tts.write_to_fp(buf)
            st.session_state.audio_bytes = buf.getvalue()

    # 5) Play on demand
    if st.button("🔊 Play Story Audio"):
        audio_buffer = io.BytesIO(st.session_state.audio_bytes)
        st.audio(audio_buffer, format="audio/mp3")