Update app.py

app.py

@@ -1,21 +1,17 @@
 import re
 import gradio as gr
 from PIL import Image
-from transformers import (
-    AutoProcessor,
-    AutoModelForVision2Seq,
-    pipeline,
-)
+from transformers import AutoProcessor, AutoModelForVision2Seq, pipeline
 
-# Load processor & model
+# 1) Load BLIP-2 processor & model
 processor = AutoProcessor.from_pretrained("Salesforce/blip2-flan-t5-xl")
 model     = AutoModelForVision2Seq.from_pretrained("Salesforce/blip2-flan-t5-xl")
 
-# Create the image-text-to-text pipeline
+# 2) Build the multimodal pipeline correctly
 pipe = pipeline(
-    task="image-text-to-text",
+    "image-text-to-text",
     model=model,
-    feature_extractor=processor.feature_extractor,
+    feature_extractor=processor.image_processor,   # BLIP2Processor uses .image_processor
     tokenizer=processor.tokenizer,
     max_new_tokens=500,
     do_sample=True,
@@ -39,6 +35,7 @@ def get_recommendations():
     ]
 
 def process(image: Image):
+    # A single prompt that asks BLIP-2+Flan-T5 to emit exactly three sections
     prompt = (
         "You are an expert ad critic. Given the image below, output exactly three sections:\n\n"
         "Category: <one concise label>\n\n"
@@ -51,25 +48,27 @@ def process(image: Image):
         "- <bullet 5>\n"
     )
 
+    # Run the pipeline
     out = pipe(image, prompt=prompt)[0]["generated_text"]
 
-    # pull out each section via regex
-    cat = re.search(r"Category:(.*?)Analysis:", out, re.S)
-    ana = re.search(r"Analysis:(.*?)Suggestions:", out, re.S)
-    sug = re.search(r"Suggestions:(.*)", out, re.S)
+    # Regex-extract each section
+    cat_match = re.search(r"Category:(.*?)Analysis:", out, re.S)
+    ana_match = re.search(r"Analysis:(.*?)Suggestions:", out, re.S)
+    sug_match = re.search(r"Suggestions:(.*)", out, re.S)
 
-    category    = cat.group(1).strip() if cat else ""
-    analysis    = ana.group(1).strip() if ana else ""
-    suggestions = sug.group(1).strip() if sug else ""
+    category    = cat_match.group(1).strip() if cat_match else ""
+    analysis    = ana_match.group(1).strip() if ana_match else ""
+    suggestions = sug_match.group(1).strip() if sug_match else ""
 
-    # ensure exactly 5 bullets
-    bullets = [l for l in suggestions.splitlines() if l.startswith("-")]
+    # Ensure exactly 5 bullets
+    bullets = [line for line in suggestions.splitlines() if line.startswith("-")]
     if len(bullets) < 5:
         bullets += ["- (no suggestion)"] * (5 - len(bullets))
     suggestions = "\n".join(bullets[:5])
 
     return category, analysis, suggestions, get_recommendations()
 
+# --- Gradio UI ---
 with gr.Blocks(theme=gr.themes.Default(primary_hue="blue")) as demo:
     gr.Markdown("## 📢 Smart Ad Analyzer")
     gr.Markdown(