app.py

# app.py

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

# —––––––– Page config
st.set_page_config(page_title="Storyteller for Kids", layout="centered")
st.title("🖼️ ➡️ 📖 Interactive Storyteller")

# —––––––– Load & warm pipelines
@st.cache_resource
def load_pipelines():
    # 1) BLIP-base for captions
    captioner = pipeline(
        "image-to-text",
        model="Salesforce/blip-image-captioning-base",
        device=0  # set to -1 if you only have CPU
    )
    # 2) DeepSeek-R1-Distill (Qwen-1.5B) for stories
    ds_storyteller = pipeline(
        "text-generation",
        model="deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B",
        trust_remote_code=True,
        device=0
    )

    # Warm-up both so the first real request is faster
    dummy = Image.new("RGB", (384, 384), color=(128, 128, 128))
    captioner(dummy)
    ds_storyteller("Warm up", max_new_tokens=1)

    return captioner, ds_storyteller

captioner, ds_storyteller = load_pipelines()

# —––––––– Main UI
uploaded = st.file_uploader("Upload an image:", type=["jpg", "jpeg", "png"])
if uploaded:
    # 1) Preprocess & display
    image = Image.open(uploaded).convert("RGB")
    image = image.resize((384, 384), Image.LANCZOS)
    st.image(image, caption="Your image", use_container_width=True)

    # 2) Generate caption
    with st.spinner("🔍 Generating caption..."):
        cap = captioner(image)[0]["generated_text"].strip()
    st.markdown(f"**Caption:** {cap}")

    # 3) Build prompt
    prompt = (
        f"Here is an image description: “{cap}”.\n"
        "Write an 80–100 word playful story for 3–10 year-old children that:\n"
        "1) Describes the scene and main subject.\n"
        "2) Explains what it’s doing and how it feels.\n"
        "3) Concludes with a fun, imaginative ending.\n\n"
        "Story:"
    )

    # 4) Generate story via DeepSeek
    with st.spinner("✍️ Generating story with DeepSeek..."):
        out = ds_storyteller(
            prompt,
            max_new_tokens=120,
            do_sample=True,
            temperature=0.7,
            top_p=0.9,
            top_k=50,
            repetition_penalty=1.2,
            no_repeat_ngram_size=3
        )
        story = out[0]["generated_text"].strip()

    st.markdown("**Story:**")
    st.write(story)

    # 5) Text-to-Speech via gTTS
    with st.spinner("🔊 Converting to speech..."):
        tts = gTTS(text=story, lang="en")
        tmp = tempfile.NamedTemporaryFile(suffix=".mp3", delete=False)
        tts.write_to_fp(tmp)
        tmp.flush()
    st.audio(tmp.name, format="audio/mp3")