# 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") # —––––––– Model loading + warm-up @st.cache_resource def load_pipelines(): # 1) Original BLIP-base for captions captioner = pipeline( "image-to-text", model="Salesforce/blip-image-captioning-base", device=0 # change to -1 if you only have CPU ) # 2) Small GPT-Neo for quick stories storyteller = pipeline( "text-generation", model="EleutherAI/gpt-neo-125M", device=0 ) # Warm up both so the first real call is faster dummy = Image.new("RGB", (384, 384), color=(128, 128, 128)) captioner(dummy) storyteller("Hello", max_new_tokens=1) return captioner, storyteller captioner, storyteller = load_pipelines() # —––––––– Main UI uploaded = st.file_uploader("Upload an image:", type=["jpg", "jpeg", "png"]) if uploaded: # 1) Load + resize for faster encoding image = Image.open(uploaded).convert("RGB") image = image.resize((384, 384), Image.LANCZOS) st.image(image, caption="Your image", use_container_width=True) # 2) Caption step with st.spinner("🔍 Generating caption..."): cap = captioner(image)[0]["generated_text"].strip() st.markdown(f"**Caption:** {cap}") # 3) Story generation (sampling + repetition control) prompt = ( f"Write an 80–100 word fun story for 3–10 year-old children " f"based on this description:\n\n“{cap}”\n\nStory: " ) with st.spinner("✍️ Generating story..."): out = storyteller( prompt, max_new_tokens=120, # room for ~100 words do_sample=True, # enable sampling temperature=0.8, # creativity top_p=0.9, # nucleus sampling top_k=50, # limit to top 50 tokens repetition_penalty=1.2, # discourage exact repeats no_repeat_ngram_size=3 # prevent 3-gram repeats ) # strip off the prompt so only the story remains story = out[0]["generated_text"][len(prompt):].strip() st.markdown("**Story:**") st.write(story) # 4) 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")