Update app.py

app.py

@@ -4,22 +4,22 @@ from transformers import BlipProcessor, BlipForConditionalGeneration, pipeline
 
 # Initialize BLIP for image captioning
 blip_processor = BlipProcessor.from_pretrained(
-    "Salesforce/blip-image-captioning-base"
+    "Salesforce/blip-image-captioning-base",
+    use_fast=True
 )
 blip_model = BlipForConditionalGeneration.from_pretrained(
     "Salesforce/blip-image-captioning-base"
 )
 
-# Instruction‑tuned HF model for ad improvement suggestions
-adv_generator = pipeline(
+# Lightweight instruction‑tuned HF model for suggestions
+advisor = pipeline(
     "text2text-generation",
-    model="google/flan-ul2",
-    tokenizer="google/flan-ul2",
-    max_new_tokens=150,
+    model="google/flan-t5-base",
+    tokenizer="google/flan-t5-base",
+    max_new_tokens=80,
     do_sample=True,
     temperature=0.7,
-    top_k=50,
-    top_p=0.95
+    top_p=0.9
 )
 
 
@@ -28,73 +28,60 @@ def get_recommendations():
         "https://i.imgur.com/InC88PP.jpeg",
         "https://i.imgur.com/7BHfv4T.png",
         "https://i.imgur.com/wp3Wzc4.jpeg",
-        # ... add more examples as needed
+        # add more examples if desired
     ]
 
 
 def generate_caption(image):
     inputs = blip_processor(images=image, return_tensors="pt")
-    out = blip_model.generate(**inputs)
-    caption = blip_processor.decode(out[0], skip_special_tokens=True)
+    outputs = blip_model.generate(**inputs)
+    caption = blip_processor.decode(outputs[0], skip_special_tokens=True)
     return caption
 
 
 def generate_advice_from_caption(caption):
-    """
-    Uses a Hugging Face instruction‑tuned model (Flan‑UL2) to turn the BLIP caption
-    into targeted, bullet‑pointed ad improvement suggestions.
-    """
     prompt = (
         f"Ad description: {caption}\n"
-        "Please suggest three specific improvements as bullet points, focusing on a clear call to action, key benefits, and visual appeal."
+        "Suggest three concise bullet‑point improvements: clear call to action, highlight key benefits, boost visual appeal."
     )
-    result = adv_generator(prompt)
-    # The model output may include the prompt, so extract after prompt
+    result = advisor(prompt)
     text = result[0]["generated_text"].replace(prompt, "").strip()
     return text
 
 
 def analyze_caption(caption):
-    keywords = ["product", "offer", "smile", "call to action", "brand"]
-    if any(k in caption.lower() for k in keywords):
-        return "👍 Likely effective for advertising."
-    return "👎 Consider clearer focus or stronger messaging."
+    keywords = ["product", "offer", "call to action", "brand"]
+    return (
+        "👍 Likely effective for advertising."
+        if any(k in caption.lower() for k in keywords)
+        else "👎 Consider clearer focus or stronger messaging."
+    )
 
 
 def process(image):
     caption = generate_caption(image)
-    suggestions = generate_advice_from_caption(caption)
+    advice = generate_advice_from_caption(caption)
     analysis = analyze_caption(caption)
-    recomms = get_recommendations()
-    return caption, suggestions, analysis, recomms
+    recs = get_recommendations()
+    return caption, advice, analysis, recs
 
 with gr.Blocks(theme=gr.themes.Default(primary_hue="blue")) as demo:
     gr.Markdown("## 📢 Smart Ad Analyzer")
     gr.Markdown(
-        "Upload an image ad. The app will caption it, suggest improvements with a local instruction‑tuned model, analyze focus, and show example ads."
+        "Upload an image ad. BLIP captions it. Flan‑T5‑Base suggests improvements. Then you get analysis and example ads."
     )
 
     with gr.Row():
         image_input = gr.Image(type="pil", label="Upload Ad Image")
         with gr.Column():
-            caption_out = gr.Textbox(
-                label="Generated Caption", interactive=False
-            )
-            suggestion_out = gr.Textbox(
-                label="Improvement Suggestions", interactive=False
-            )
-            analysis_out = gr.Textbox(
-                label="Ad Analysis", interactive=False
-            )
+            caption_out = gr.Textbox(label="Generated Caption", interactive=False)
+            suggestion_out = gr.Textbox(label="Improvement Suggestions", interactive=False)
+            analysis_out = gr.Textbox(label="Ad Analysis", interactive=False)
 
     btn = gr.Button("Analyze Ad")
-    recommendation_gallery = gr.Gallery(label="Recommended Example Ads")
+    gallery = gr.Gallery(label="Recommended Example Ads")
 
-    btn.click(
-        fn=process,
-        inputs=[image_input],
-        outputs=[caption_out, suggestion_out, analysis_out, recommendation_gallery]
-    )
+    btn.click(fn=process, inputs=[image_input], outputs=[caption_out, suggestion_out, analysis_out, gallery])
 
     gr.Markdown("Made by Simon Thalmay")