Updated script

alt-text-generator.py

@@ -36,6 +36,7 @@ def validate_anthropic_key(selected_model):
     "Alt text appears to be meaningless. Replace it with descriptive content.",
     "Alt text seems too short. Consider providing more context.",
     "Consider simplifying the text.",
+    "Alt text is too short. Provide more context."
 ]
 
 # Set up logging for debugging
@@ -108,54 +109,78 @@ def generate_with_anthropic(prompt):
 
 
 def generate_with_ollama(prompt, model_name="llama3.1:latest"):
-    """Generate concise alt text using a hosted Ollama model."""
+    """Generate text using a hosted Ollama model with improved response handling."""
     try:
-        # Refine the prompt for clarity and specificity
-        refined_prompt = (
-            f"Generate alt text for the provided image. Respond ONLY with the text that should go inside the alt "
-            f"attribute of an img tag. Do not include explanations, context, or prefaces. Just provide the alt text.\n\n"
-            f"Image description: {prompt}"
+        # Create a more specific prompt that explicitly requests just the alt text
+        formatted_prompt = (
+            "Generate alt text for an image. Respond ONLY with the text that should go inside "
+            "the alt attribute of an img tag. Do not include 'Alt text:', explanations, quotes, "
+            "or any other text. Keep the description concise and factual.\n\n"
+            f"Image details: {prompt}"
         )
-
-        # Send the refined prompt to the Ollama API
-        payload = {"model": model_name, "prompt": refined_prompt}
+
+        payload = {
+            "model": model_name,
+            "prompt": formatted_prompt,
+            "stream": False,
+            "system": "You are a helpful assistant that generates alt text for images. Respond only with the alt text itself, without any explanations, disclaimers, or meta-commentary."
+        }
+
         response = requests.post(OLLAMA_API_URL, json=payload, timeout=30)
-
-        # Log the full response text for debugging
-        logging.debug(f"Ollama API raw response: {response.text}")
-
-        # Check for HTTP errors
         response.raise_for_status()
-
-        # Parse the JSON response
-        response_data = response.json()
-
-        # Post-process the response to extract only the alt text
-        alt_text = response_data.get("response", "").strip()
-
-        # Remove any unwanted prefixes or extraneous text
-        prefixes_to_remove = [
-            "Based on the image, I will generate an alt text that is concise, yet descriptive.",
-            "Here is a concise and descriptive alt text for the image:",
-            "Alt Text:",
-            "Unfortunately, I cannot verify the image, but here is a suggestion:",
-        ]
-        for prefix in prefixes_to_remove:
-            if alt_text.startswith(prefix):
-                alt_text = alt_text[len(prefix):].strip()
-
-        # Ensure the text is concise and clean
-        return alt_text.strip()
-
+
+        try:
+            response_data = response.json()
+            generated_text = ""
+
+            if "response" in response_data:
+                generated_text = response_data["response"].strip()
+            elif "text" in response_data:
+                generated_text = response_data["text"].strip()
+            else:
+                return "Error: Unexpected response structure from Ollama"
+
+            # Clean up the response
+            # Remove common prefixes
+            prefixes_to_remove = [
+                "Alt text:",
+                "Here is",
+                "The alt text is",
+                "I suggest",
+                "Based on the image,",
+                "A concise alt text would be",
+                "Here's a descriptive alt text:",
+                "The appropriate alt text is",
+            ]
+
+            for prefix in prefixes_to_remove:
+                if generated_text.lower().startswith(prefix.lower()):
+                    generated_text = generated_text[len(prefix):].strip()
+
+            # Remove quotes and leading/trailing punctuation
+            generated_text = generated_text.strip('"\'".,: ')
+
+            # Remove any explanatory text after the main description
+            if "\n" in generated_text:
+                generated_text = generated_text.split("\n")[0].strip()
+
+            # Ensure proper sentence format
+            generated_text = generated_text.strip(".")
+            if generated_text:
+                generated_text = generated_text[0].upper() + generated_text[1:] + "."
+
+            return generated_text
+
+        except ValueError as val_err:
+            logging.error(f"JSON decode error with Ollama API: {val_err}")
+            return "Error: Malformed JSON from Ollama API"
+
     except requests.exceptions.HTTPError as http_err:
-        logging.error(f"Ollama API HTTP error: {http_err} - Response: {http_err.response.text}")
-        return f"Error using Ollama API: {http_err.response.text}"
-    except ValueError as val_err:
-        logging.error(f"JSON decode error with Ollama API: {val_err}")
-        return "Error: Malformed JSON from Ollama API"
+        logging.error(f"Ollama API HTTP error: {http_err}")
+        return f"Error using Ollama API: {http_err}"
     except Exception as e:
         logging.error(f"Error using Ollama API: {e}")
-        return f"Error generating text with Ollama API: {e}"
+        return f"Error generating text with Ollama API: {str(e)}"
 
 
 def check_image_exists(image_url):