Update app.py

app.py

@@ -1,44 +1,10 @@
+import os
 import gradio as gr
-from PIL import Image
 from transformers import pipeline
-import tempfile
 
-# Initialize pipeline (Gemma 3, image-text-to-text)
-pipe = pipeline("image-text-to-text", model="google/gemma-3-4b-it")
-
-def process(image):
-    # Save image to temp file and use local path for Gemma
-    with tempfile.NamedTemporaryFile(suffix=".png", delete=True) as tmp:
-        image.save(tmp, format="PNG")
-        tmp.flush()
-        img_path = tmp.name
-
-        # Build multimodal message for Gemma
-        messages = [
-            {
-                "role": "user",
-                "content": [
-                    {"type": "image", "path": img_path},
-                    {"type": "text", "text": "Analyze this ad image in detail. What is the product or service? Who is the target audience? Suggest five unique improvements to the ad. Output exactly four fields separated by |||: (1) Category (2) Five-sentence analysis (3) Five bullet suggestions (4) Short punchy headline for the ad."}
-                ]
-            }
-        ]
-
-        # Get model output
-        out = pipe(text=messages, max_new_tokens=512)
-        output = out[0]["generated_text"][-1]["content"] if "generated_text" in out[0] else out[0]["content"]
-
-    # Parse results by "|||"
-    if "|||" in output:
-        cat, analysis, suggestions, headline = [x.strip() for x in output.split("|||")]
-    else:
-        cat = "N/A"
-        analysis = output.strip()
-        suggestions = ""
-        headline = ""
-
-    # Also provide example ads (unchanged)
-    gallery = [
+def get_recommendations():
+    # Returns list of 10 example ad image URLs
+    return [
         "https://i.imgur.com/InC88PP.jpeg",
         "https://i.imgur.com/7BHfv4T.png",
         "https://i.imgur.com/wp3Wzc4.jpeg",
@@ -51,32 +17,68 @@ def process(image):
         "https://i.imgur.com/Xj92Cjv.jpeg",
     ]
 
-    return cat, analysis, suggestions, headline, gallery
+# BLIP for image captioning (always on CPU, runs fast)
+captioner = pipeline("image-to-text", model="Salesforce/blip-image-captioning-base", device="cpu")
+
+# Gemma 1B for text generation
+gemma = pipeline("text-generation", model="google/gemma-1.1-1b-it", device="cpu")
+
+def process(image):
+    if image is None:
+        return "", "", "", get_recommendations()
+
+    # 1. Caption image
+    cap_result = captioner(image)
+    caption = cap_result[0]['generated_text'] if cap_result else "No caption generated."
+
+    # 2. Compose a prompt for Gemma (category, analysis, suggestions)
+    prompt = (
+        f"Here is a description of an ad image: {caption}\n"
+        "1. Assign a concise ad category label (e.g., 'Food', 'Fitness').\n"
+        "2. Write exactly five sentences analyzing what this ad communicates and its emotional impact.\n"
+        "3. Suggest five specific ways to improve this ad, each as a short, practical sentence.\n"
+        "Answer in three parts clearly marked as Category, Analysis, and Suggestions."
+    )
+
+    gemma_out = gemma(prompt, max_new_tokens=256)[0]['generated_text']
+    # Split results
+    lines = gemma_out.split('\n')
+    cat, analysis, suggestions = "", "", ""
+    for i, line in enumerate(lines):
+        if "category" in line.lower():
+            cat = line.split(":", 1)[-1].strip()
+        elif "analysis" in line.lower():
+            analysis = "\n".join(lines[i+1:i+6])
+        elif "suggestions" in line.lower():
+            suggestions = "\n".join(lines[i+1:i+6])
+    # Fallback if Gemma output is not perfectly formatted
+    if not cat: cat = lines[0][:80]
+    if not analysis: analysis = "\n".join(lines[1:6])
+    if not suggestions: suggestions = "\n".join(lines[6:11])
+
+    return cat.strip(), analysis.strip(), suggestions.strip(), get_recommendations()
 
 def main():
-    with gr.Blocks(title="Smart Ad Analyzer (Gemma 3)") as demo:
-        gr.Markdown("## 📢 Smart Ad Analyzer (Gemma 3)")
-        gr.Markdown("""
-        Upload your ad image and get:
-        - **Category**
-        - **Analysis**
-        - **5 Suggestions**
-        - **Headline idea**
-        - **Example ad gallery**
-        """)
+    with gr.Blocks(title="Smart Ad Analyzer (BLIP+Gemma Edition)") as demo:
+        gr.Markdown("## 📢 Smart Ad Analyzer (BLIP+Gemma Edition)")
+        gr.Markdown(
+            """
+            Upload your ad image below and instantly get expert feedback.
+            Category, analysis, improvement suggestions—and example ads for inspiration.
+            """
+        )
         with gr.Row():
             inp = gr.Image(type='pil', label='Upload Ad Image')
             with gr.Column():
-                cat_out = gr.Textbox(label='Ad Category')
-                ana_out = gr.Textbox(label='Analysis', lines=5)
-                sug_out = gr.Textbox(label='Improvement Suggestions', lines=5)
-                head_out = gr.Textbox(label='Headline Suggestion')
+                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)
                 btn = gr.Button('Analyze Ad', variant='primary')
         gallery = gr.Gallery(label='Example Ads')
         btn.click(
             fn=process,
             inputs=[inp],
-            outputs=[cat_out, ana_out, sug_out, head_out, gallery],
+            outputs=[cat_out, ana_out, sug_out, gallery],
         )
         gr.Markdown('Made by Simon Thalmay')
     return demo