Update app.py

app.py

@@ -10,10 +10,9 @@ from transformers import (
     AutoModelForSeq2SeqLM,
 )
 
-# Auto-detect CPU/GPU
 DEVICE = 0 if torch.cuda.is_available() else -1
 
-# 1) BLIP captioner
+# BLIP captioner setup
 processor = AutoProcessor.from_pretrained("Salesforce/blip-image-captioning-large")
 blip_model = AutoModelForVision2Seq.from_pretrained("Salesforce/blip-image-captioning-large")
 caption_pipe = pipeline(
@@ -24,7 +23,7 @@ caption_pipe = pipeline(
     device=DEVICE,
 )
 
-# 2) FLAN-T5 for text-to-text
+# FLAN-T5 setup
 FLAN_MODEL = "google/flan-t5-large"
 flan_tokenizer = AutoTokenizer.from_pretrained(FLAN_MODEL)
 flan_model = AutoModelForSeq2SeqLM.from_pretrained(FLAN_MODEL)
@@ -38,7 +37,6 @@ category_pipe = pipeline(
     do_sample=True,
     temperature=1.0,
 )
-
 analysis_pipe = pipeline(
     "text2text-generation",
     model=flan_model,
@@ -48,7 +46,6 @@ analysis_pipe = pipeline(
     do_sample=True,
     temperature=1.0,
 )
-
 suggestion_pipe = pipeline(
     "text2text-generation",
     model=flan_model,
@@ -58,7 +55,6 @@ suggestion_pipe = pipeline(
     do_sample=True,
     temperature=1.0,
 )
-
 expansion_pipe = pipeline(
     "text2text-generation",
     model=flan_model,
@@ -68,7 +64,6 @@ expansion_pipe = pipeline(
     do_sample=False,
 )
 
-# Example gallery helper
 def get_recommendations():
     return [
         "https://i.imgur.com/InC88PP.jpeg",
@@ -83,74 +78,71 @@ def get_recommendations():
         "https://i.imgur.com/Xj92Cjv.jpeg",
     ]
 
-# Main processing function
-
 def process(image: Image):
     if image is None:
-        return "", "", "", "", get_recommendations()
+        return "", "", "", get_recommendations()
 
-    # 1) BLIP caption
+    # Captioning with BLIP
     caption_res = caption_pipe(image, max_new_tokens=64)
     raw_caption = caption_res[0]["generated_text"].strip()
-
-    # 1a) Expand if too short
     if len(raw_caption.split()) < 3:
-        exp = expansion_pipe(f"Expand into a detailed description: {raw_caption}")
-        desc = exp[0]["generated_text"].strip()
+        desc = expansion_pipe(f"Expand into a detailed description: {raw_caption}")[0]["generated_text"].strip()
     else:
         desc = raw_caption
 
-    # 2) Category
+    # Ad Category
     cat_prompt = (
         f"Description: {desc}\n\n"
-        "Provide a concise category label for this ad (e.g. 'Food', 'Fitness'):"
+        "Provide a concise, one or two-word category label for this ad (examples: 'Food', 'Fitness', 'Fashion', 'Tech'):"
     )
     cat_out = category_pipe(cat_prompt)[0]["generated_text"].splitlines()[0].strip()
 
-    # 3) Five-sentence analysis
+    # Analysis
     ana_prompt = (
         f"Description: {desc}\n\n"
-        "Write exactly five sentences explaining what this ad communicates and its emotional impact."
+        "Write exactly five unique and meaningful sentences explaining what this ad communicates, its visual style, the target audience, the marketing message, and emotional appeal."
     )
     ana_raw = analysis_pipe(ana_prompt)[0]["generated_text"].strip()
     sentences = re.split(r'(?<=[.!?])\s+', ana_raw)
     analysis = " ".join(sentences[:5])
 
-    # 4) Five bullet-point suggestions
+    # Bullet Suggestions (enforced non-repetitive)
     sug_prompt = (
         f"Description: {desc}\n\n"
-        "Provide five distinct improvement suggestions, each starting with '- '."
+        "Suggest five actionable and distinct ways this ad could be improved. Each improvement must start with '- ' and each one should address a different aspect such as clarity, visual design, call-to-action, message, or emotional engagement. No repeats or generic phrases."
     )
     sug_raw = suggestion_pipe(sug_prompt)[0]["generated_text"].strip()
-    bullets = [l for l in sug_raw.splitlines() if l.startswith("-")]
-    if len(bullets) < 5:
-        for line in sug_raw.splitlines():
-            if len(bullets) >= 5:
-                break
-            if line and not line.startswith("-"):
-                bullets.append("- " + line)
-    while len(bullets) < 5:
-        bullets.append("- Improve visual appeal and clarity.")
-    suggestions = "\n".join(bullets[:5])
+    bullets = [l for l in sug_raw.splitlines() if l.strip().startswith("- ")]
+    seen = set()
+    cleaned = []
+    for b in bullets:
+        text = b.strip().lower()
+        if text not in seen and len(cleaned) < 5:
+            cleaned.append(b)
+            seen.add(text)
+    while len(cleaned) < 5:
+        cleaned.append("- Add more product context or a stronger call to action.")
+
+    suggestions = "\n".join(cleaned[:5])
 
-    return raw_caption, cat_out, analysis, suggestions, get_recommendations()
+    return cat_out, analysis, suggestions, get_recommendations()
 
-# Gradio UI
 def main():
     with gr.Blocks(title="Smart Ad Analyzer") as demo:
         gr.Markdown("## 📢 Smart Ad Analyzer")
         gr.Markdown(
-            "Upload an ad image to get:\n"
-            "- 🔍 **BLIP Caption**\n"
-            "- 📂 **Ad Category**\n"
-            "- 📊 **Five-sentence Analysis**\n"
-            "- 🚀 **Five Improvement Suggestions**\n"
-            "- 📸 **Example Ads**"
+            "Welcome to Smart Ad Analyzer! \n\n"
+            "Upload any ad image and instantly receive:\n"
+            "- 📂 **Ad Category:** How would an AI classify this ad?\n"
+            "- 📊 **In-depth Analysis:** Five unique sentences covering message, design, audience, and emotion.\n"
+            "- 🚀 **Improvement Suggestions:** Five non-repetitive, actionable ideas to make your ad stronger.\n"
+            "- 📸 **Example Ads Gallery:** For creative inspiration.\n\n"
+            "Designed for marketers, designers, students, or anyone curious about effective advertising. No account or API needed."
         )
         with gr.Row():
             inp = gr.Image(type='pil', label='Upload Ad Image')
             with gr.Column():
-                cap_out = gr.Textbox(label='🔍 BLIP Caption', interactive=False)
+                # BLIP caption hidden from UI
                 cat_out = gr.Textbox(label='📂 Ad Category', interactive=False)
                 ana_out = gr.Textbox(label='📊 Ad Analysis', lines=5, interactive=False)
                 sug_out = gr.Textbox(label='🚀 Improvement Suggestions', lines=5, interactive=False)
@@ -159,7 +151,7 @@ def main():
         btn.click(
             fn=process,
             inputs=[inp],
-            outputs=[cap_out, cat_out, ana_out, sug_out, gallery],
+            outputs=[cat_out, ana_out, sug_out, gallery],
         )
         gr.Markdown('Made by Simon Thalmay')
     return demo