Update app.py

app.py

@@ -1,23 +1,25 @@
-import os
 import gradio as gr
 from PIL import Image
-from transformers import BlipProcessor, BlipForConditionalGeneration, pipeline as hf_pipeline
-import openai
-
-# Read OpenAI API key
-openai.api_key = os.getenv("OPENAI_API_KEY")
-has_api = bool(openai.api_key)
-
-# Load BLIP for image captioning
-blip_processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-base")
-blip_model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-base")
-
-# Local fallback generator (EleutherAI GPT-Neo 2.7B)
-local_generator = hf_pipeline(
-    "text-generation",
-    model="EleutherAI/gpt-neo-2.7B",
-    tokenizer="EleutherAI/gpt-neo-2.7B",
-    framework="pt"
+from transformers import BlipProcessor, BlipForConditionalGeneration, pipeline
+
+# Initialize BLIP for image captioning
+blip_processor = BlipProcessor.from_pretrained(
+    "Salesforce/blip-image-captioning-base"
+)
+blip_model = BlipForConditionalGeneration.from_pretrained(
+    "Salesforce/blip-image-captioning-base"
+)
+
+# Instruction‑tuned HF model for ad improvement suggestions
+adv_generator = pipeline(
+    "text2text-generation",
+    model="google/flan-ul2",
+    tokenizer="google/flan-ul2",
+    max_new_tokens=150,
+    do_sample=True,
+    temperature=0.7,
+    top_k=50,
+    top_p=0.95
 )
 
 
@@ -26,83 +28,64 @@ def get_recommendations():
         "https://i.imgur.com/InC88PP.jpeg",
         "https://i.imgur.com/7BHfv4T.png",
         "https://i.imgur.com/wp3Wzc4.jpeg",
-        # ... other URLs ...
+        # ... add more examples as needed
     ]
 
 
 def generate_caption(image):
     inputs = blip_processor(images=image, return_tensors="pt")
     out = blip_model.generate(**inputs)
-    return blip_processor.decode(out[0], skip_special_tokens=True)
+    caption = blip_processor.decode(out[0], skip_special_tokens=True)
+    return caption
 
 
 def generate_advice_from_caption(caption):
-    if has_api:
-        try:
-            messages = [
-                {"role": "system", "content": "You are an expert ad copywriter."},
-                {"role": "user", "content": (
-                    f"Ad description: {caption}\n"
-                    "Suggest three specific improvements to this ad. "
-                    "Focus on a clear call to action, highlight key benefits, and boost visual appeal. "
-                    "Return concise bullet points."
-                )}
-            ]
-            resp = openai.ChatCompletion.create(
-                model="gpt-3.5-turbo",
-                messages=messages,
-                temperature=0.7,
-                max_tokens=200
-            )
-            return resp.choices[0].message.content.strip()
-        except Exception as e:
-            return f"Error generating suggestions: {e}"
-    else:
-        prompt = (
-            f"Ad description: {caption}\n"
-            "Suggest three specific improvements as bullet points."
-        )
-        result = local_generator(
-            prompt,
-            max_new_tokens=150,
-            do_sample=True,
-            temperature=0.8,
-            top_k=40,
-            top_p=0.9
-        )
-        # remove prompt from output
-        text = result[0]["generated_text"].replace(prompt, "").strip()
-        return text
+    """
+    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."
+    )
+    result = adv_generator(prompt)
+    # The model output may include the prompt, so extract after prompt
+    text = result[0]["generated_text"].replace(prompt, "").strip()
+    return text
 
 
 def analyze_caption(caption):
     keywords = ["product", "offer", "smile", "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."
-    )
+    if any(k in caption.lower() for k in keywords):
+        return "👍 Likely effective for advertising."
+    return "👎 Consider clearer focus or stronger messaging."
 
 
 def process(image):
     caption = generate_caption(image)
     suggestions = generate_advice_from_caption(caption)
     analysis = analyze_caption(caption)
-    recs = get_recommendations()
-    return caption, suggestions, analysis, recs
+    recomms = get_recommendations()
+    return caption, suggestions, analysis, recomms
 
 with gr.Blocks(theme=gr.themes.Default(primary_hue="blue")) as demo:
     gr.Markdown("## 📢 Smart Ad Analyzer")
     gr.Markdown(
-        "Upload an image ad. The model will caption it, suggest improvements, analyze focus, and show examples."
+        "Upload an image ad. The app will caption it, suggest improvements with a local instruction‑tuned model, analyze focus, and show 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")
@@ -115,4 +98,5 @@ with gr.Blocks(theme=gr.themes.Default(primary_hue="blue")) as demo:
 
     gr.Markdown("Made by Simon Thalmay")
 
-if __name__ == "__main__":
+if __name__ == "__main__":
+    demo.launch()