Update app.py

app.py

@@ -1,19 +1,44 @@
 import gradio as gr
 from PIL import Image
-import io
-import os
-from huggingface_hub import InferenceClient
-
-# You may need to set your Hugging Face token as an environment variable:
-# os.environ["HF_TOKEN"] = "your-hf-token"
-client = InferenceClient(
-    model="google/gemma-3-4b-it",
-    token=os.environ.get("HF_TOKEN")  # Add your token if needed for Spaces
-)
-
-def get_recommendations():
-    # Returns list of 10 example ad image URLs
-    return [
+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 = [
         "https://i.imgur.com/InC88PP.jpeg",
         "https://i.imgur.com/7BHfv4T.png",
         "https://i.imgur.com/wp3Wzc4.jpeg",
@@ -26,104 +51,32 @@ def get_recommendations():
         "https://i.imgur.com/Xj92Cjv.jpeg",
     ]
 
-def gemma_image_analysis(image: Image.Image):
-    buf = io.BytesIO()
-    image.save(buf, format="PNG")
-    buf.seek(0)
-    img_bytes = buf.read()
-
-    messages = [
-        {
-            "role": "system",
-            "content": [
-                {"type": "text", "text": (
-                    "You are an expert ad analyst AI. Analyze the given ad and provide answers for three sections:"
-                    " 1. Category (one word),"
-                    " 2. Analysis (five sentences),"
-                    " 3. Five unique, actionable improvement suggestions as a list starting with '- '."
-                    " Output must be in three sections with clear headings."
-                )}
-            ]
-        },
-        {
-            "role": "user",
-            "content": [
-                {"type": "image", "image": img_bytes},
-                {"type": "text", "text": (
-                    "Please respond with:\n\nCategory:\n[category]\n\nAnalysis:\n[5 sentences]\n\nImprovement Suggestions:\n"
-                    "- [suggestion 1]\n- [suggestion 2]\n- [suggestion 3]\n- [suggestion 4]\n- [suggestion 5]\n\n"
-                    "Each suggestion must be unique and actionable. Do not repeat suggestions. If you don't know, say 'not detected'."
-                )}
-            ]
-        }
-    ]
-
-    response = client.chat.completions.create(
-        model="google/gemma-3-4b-it",
-        messages=messages,
-        max_tokens=512,
-    )
-
-    return response.choices[0].message["content"]
-
-def process(image: Image.Image):
-    if image is None:
-        return "", "", "", get_recommendations()
-
-    # Run multimodal analysis
-    full_output = gemma_image_analysis(image)
-
-    # Parse the output into sections (simple but robust splitting)
-    category = ""
-    analysis = ""
-    suggestions = ""
-    # Find the sections in model output (robust to little formatting errors)
-    try:
-        lower = full_output.lower()
-        cat_idx = lower.find("category")
-        ana_idx = lower.find("analysis")
-        sug_idx = lower.find("improvement suggestions")
-
-        if cat_idx != -1 and ana_idx != -1:
-            category = full_output[cat_idx + 8 : ana_idx].strip().strip(":")
-        if ana_idx != -1 and sug_idx != -1:
-            analysis = full_output[ana_idx + 8 : sug_idx].strip().strip(":")
-        if sug_idx != -1:
-            suggestions = full_output[sug_idx + len("improvement suggestions"):].strip()
-    except Exception:
-        # Fallback if parsing fails
-        return "", "Analysis parsing failed", "Suggestion parsing failed", get_recommendations()
-
-    return category.strip(), analysis.strip(), suggestions.strip(), get_recommendations()
+    return cat, analysis, suggestions, headline, gallery
 
 def main():
-    with gr.Blocks(title="Smart Ad Analyzer") as demo:
-        gr.Markdown("## 📢 Smart Ad Analyzer (Gemma-3 Model)")
-        gr.Markdown(
-            """
-            **Upload your ad image below and instantly get expert feedback.**
-
-            This AI tool will analyze your ad and provide:
-            - 📂 **Category** — What type of ad is this?
-            - 📊 **In-depth Analysis** — Five detailed sentences covering message, visuals, emotional impact, and more.
-            - 🚀 **Improvement Suggestions** — Five actionable, unique ways to make your ad better.
-            - 📸 **Inspiration Gallery** — See other effective ads for ideas.
-
-            Perfect for marketers, founders, designers, and anyone looking to boost ad performance with actionable insights!
-            """
-        )
+    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.Row():
             inp = gr.Image(type='pil', label='Upload Ad Image')
             with gr.Column():
-                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)
+                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')
                 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, gallery],
+            outputs=[cat_out, ana_out, sug_out, head_out, gallery],
         )
         gr.Markdown('Made by Simon Thalmay')
     return demo
@@ -131,4 +84,3 @@ def main():
 if __name__ == "__main__":
     demo = main()
     demo.launch()
-