Update app.py

app.py

@@ -17,62 +17,62 @@ def load_pipelines():
     captioner = pipeline(
         "image-to-text",
         model="Salesforce/blip-image-captioning-base",
-        device=0  # change to -1 if you only have CPU
+        device=0
     )
-    # 2) Small GPT-Neo for quick stories
+    # 2) Instruction-tuned Flan-T5 small for stories
     storyteller = pipeline(
-        "text-generation",
-        model="EleutherAI/gpt-neo-125M",
+        "text2text-generation",
+        model="google/flan-t5-small",
         device=0
     )
 
-    # Warm up both so the first real call is faster
+    # Warm up so first real request is faster
     dummy = Image.new("RGB", (384, 384), color=(128, 128, 128))
     captioner(dummy)
-    storyteller("Hello", max_new_tokens=1)
+    storyteller("Tell me something", max_new_tokens=1)
 
     return captioner, storyteller
 
 captioner, storyteller = load_pipelines()
 
 # —––––––– Main UI
-uploaded = st.file_uploader("Upload an image:", type=["jpg", "jpeg", "png"])
+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)
+    # 1) Load + downsize
+    image = Image.open(uploaded).convert("RGB").resize((384, 384), Image.LANCZOS)
     st.image(image, caption="Your image", use_container_width=True)
 
-    # 2) Caption step
+    # 2) Caption
     with st.spinner("🔍 Generating caption..."):
         cap = captioner(image)[0]["generated_text"].strip()
     st.markdown(f"**Caption:** {cap}")
 
-    # 3) Story generation (sampling + repetition control)
+    # 3) Story generation
     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: "
+        f"Here is an image description: “{cap}”.\n"
+        "Write a playful, 80–100 word story for 3–10 year-olds\n\n"
+        "Story:"
     )
     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
+            max_new_tokens=150,
+            do_sample=True,
+            temperature=0.7,
+            top_p=0.9,
+            repetition_penalty=1.2,
+            no_repeat_ngram_size=3
         )
-        # strip off the prompt so only the story remains
-        story = out[0]["generated_text"][len(prompt):].strip()
+        # strip off the prompt so you only get the story
+        story = out[0]["generated_text"].split("Story:")[-1].strip()
+
     st.markdown("**Story:**")
     st.write(story)
 
-    # 4) Text-to-Speech via gTTS
+    # 4) Text-to-Speech
     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")
+    st.audio(tmp.name, format="audio/mp3")