app.py

# app.py

import streamlit as st
from PIL import Image
from io import BytesIO
from huggingface_hub import InferenceApi
from gtts import gTTS
import tempfile

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

# —––––––– Inference clients (cached)
@st.cache_resource
def load_clients():
    # read your HF token from Space secrets
    hf_token = st.secrets["HF_TOKEN"]

    # caption client: BLIP-base via HF Image-to-Text API
    caption_client = InferenceApi(
        repo_id="Salesforce/blip-image-captioning-base",
        task="image-to-text",
        token=hf_token
    )
    # story client: DeepSeek-R1-Distill via HF Text-Generation API
    story_client = InferenceApi(
        repo_id="deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B",
        task="text-generation",
        token=hf_token
    )
    return caption_client, story_client

caption_client, story_client = load_clients()

# —––––––– Main UI
uploaded = st.file_uploader("Upload an image:", type=["jpg", "jpeg", "png"])
if not uploaded:
    st.info("Please upload an image (JPG/PNG) to begin.")
else:
    # 1) Display the image
    img = Image.open(uploaded).convert("RGB")
    st.image(img, use_container_width=True)

    # 2) Caption via HF Inference API
    with st.spinner("🔍 Generating caption..."):
        buf = BytesIO()
        img.save(buf, format="PNG")
        caption_output = caption_client(data=buf.getvalue())
        # handle API return formats
        if isinstance(caption_output, dict):
            cap_text = caption_output.get("generated_text", "").strip()
        else:
            cap_text = str(caption_output).strip()
    st.markdown(f"**Caption:** {cap_text}")

    # 3) Build prompt
    prompt = (
        f"Here’s an image description: “{cap_text}”.\n\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) Story via HF Inference API
    with st.spinner("✍️ Generating story..."):
        story_output = story_client(
            inputs=prompt,
            params={
                "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
            }
        )
        # API returns list of generations or a dict
        if isinstance(story_output, list):
            story = story_output[0].get("generated_text", "").strip()
        else:
            story = story_output.get("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")