Update app.py

app.py

@@ -4,7 +4,6 @@ import gradio as gr
 from dotenv import load_dotenv
 import io
 from PIL import Image
-import pytesseract  # Using Tesseract OCR to extract text from the image
 
 # Load environment variables (where your OpenAI key will be stored)
 load_dotenv()
@@ -12,42 +11,57 @@ load_dotenv()
 # Load the OpenAI API key from environment variables and strip any trailing newlines or spaces
 openai.api_key = os.getenv("OPENAI_API_KEY").strip()
 
-# Function to analyze the ad image by first extracting the text with pytesseract
+# Function to analyze the ad image using GPT-4 Vision's multimodal capabilities
 def analyze_ad(image):
-    # Extract text from the image using Tesseract OCR
-    ad_copy = pytesseract.image_to_string(image)
-
-    if not ad_copy.strip():  # If OCR doesn't extract text, return an error message
-        return "No text was detected in the image. Please upload a clearer ad image."
+    # Convert the PIL image to bytes for GPT-4 Vision input
+    image_bytes = io.BytesIO()
+    image.save(image_bytes, format='PNG')
+    image_bytes = image_bytes.getvalue()
 
     # Prompt for the marketing persona and scoring rubric
-    prompt = f"""
-    Analyze the following ad copy and generate a marketing persona. Then, provide a score (out of 10) for each of the following:
-    
+    prompt = """
+    Analyze this advertisement image and extract any text present in the image. 
+    Then, generate a marketing persona based on the ad. Provide a score (out of 10) for each of the following:
+
     1. Relevance to Target Audience: Is the ad appealing to the intended demographic?
     2. Emotional Engagement: Does the ad evoke the right emotional response?
     3. Brand Consistency: Does the ad align with the brand’s voice and values?
     4. Creativity: How unique or innovative is the ad's design and text approach?
     5. Persuasiveness: Does the ad motivate action, such as clicking or purchasing?
 
-    Ad Copy: {ad_copy}
-    
     Provide the persona description and the scores in table form with a final score.
     """
 
-    # Send the prompt to GPT-4-turbo for analysis
+    # Send the image and prompt to GPT-4-turbo for multimodal analysis
     response = openai.ChatCompletion.create(
-        model="gpt-4-turbo",
+        model="gpt-4-turbo",  # Use the GPT-4 Vision-enabled model
         messages=[
             {"role": "system", "content": "You are a marketing expert analyzing an advertisement."},
             {"role": "user", "content": prompt}
         ],
+        functions=[
+            {
+                "name": "analyze_image",
+                "description": "Analyze an image and generate marketing insights",
+                "parameters": {
+                    "type": "image",
+                    "properties": {
+                        "image": {
+                            "type": "string",
+                            "description": "The input advertisement image for analysis"
+                        }
+                    },
+                    "required": ["image"]
+                }
+            }
+        ],
+        function_call={"name": "analyze_image", "arguments": {"image": image_bytes}},  # Sending the image as input
         temperature=0.7,
-        max_tokens=400
+        max_tokens=500
     )
 
     # Extract the response text from the API output
-    result = response['choices'][0]['message']['content']
+    result = response['choices'][0]['message']['content'].strip()
 
     # Return the result for display
     return result