Adjust prompt

frigate/genai/__init__.py

@@ -41,31 +41,34 @@ class GenAIClient:
         """Generate a description for the review item activity."""
         context_prompt = f"""
         Please analyze the image(s), which are in chronological order, strictly from the perspective of the {review_data["camera"].replace("_", " ")} security camera.
-        Your task is to provide a **neutral, factual, and objective description** of the scene, while also including **reasonable, evidence-based inferences** about the likely context or activity — but do not make unfounded assumptions.
+
+        Your task is to provide a **neutral, factual, and objective description** of the scene, while also:
+        - Clearly stating **what is happening** based on observable actions and movements.
+        - Including **reasonable, evidence-based inferences** about the likely activity or context, but only if directly supported by visible details.
 
         When forming your description:
-        - Base all statements on visible details in the images.
+        - **Facts first**: Describe the physical setting, people, and objects exactly as seen.
-        - You may deduce plausible intent or context only if supported by clear, observable evidence (e.g., someone carrying tools toward a car may indicate vehicle maintenance).
+        - **Then context**: Briefly note plausible purposes or activities (e.g., “appears to be delivering a package” if carrying a box to a door).
-        - Avoid implying hostility, criminal intent, or other strong judgments unless the visual evidence is unambiguous.
+        - Clearly separate certain facts (“A person is holding a ladder”) from reasonable inferences (“likely performing maintenance”).
-        - Distinguish between what is certain (facts) and what appears likely (reasonable inference).
+        - Do not speculate beyond what is visible, and do not imply hostility, criminal intent, or other strong judgments unless there is unambiguous visual evidence.
 
-        Here is some information we already know:
+        Here is information already known:
-        - The activity occurred at {review_data["timestamp"].strftime("%I:%M %p")}
+        - Activity occurred at {review_data["timestamp"].strftime("%I:%M %p")}
-        - The following objects were detected: {review_data["objects"]}
+        - Detected objects: {review_data["objects"]}
-        - The following recognized objects were detected: {review_data["recognized_objects"]}
+        - Recognized objects: {review_data["recognized_objects"]}
-        - The activity happened in the following zones: {review_data["zones"]}
+        - Zones involved: {review_data["zones"]}
 
-        Your response **MUST** be a flat JSON object with the following fields:
+        Your response **MUST** be a flat JSON object with:
-        - `scene` (string): A comprehensive description of the setting and entities, including relevant context and plausible inferences if supported by visual evidence.
+        - `scene` (string): A full description including setting, entities, actions, and any plausible supported inferences.
-        - `confidence` (float): A float between 0 and 1 representing your overall confidence in this analysis.
+        - `confidence` (float): A number 0–1 for overall confidence in the analysis.
-        - `potential_threat_level` (integer, optional): 1–3 scale. Only include if a clear security concern is visible:
+        - `potential_threat_level` (integer, optional): Include only if there is a clear, observable security concern:
-        - 1 = Unusual but not overtly threatening
+            - 1 = Unusual but not overtly threatening
-        - 2 = Suspicious or potentially harmful
+            - 2 = Suspicious or potentially harmful
-        - 3 = Clear and immediate threat
+            - 3 = Clear and immediate threat
-        Omit this field entirely if there is no observable security concern.
+            Omit this field if no concern is evident.
 
         **IMPORTANT:**
-        Values for each field must be plain strings or integers — no nested objects or explanatory text.
+        - Values must be plain strings, floats, or integers — no nested objects, no extra commentary.
         """
         response = self._send(context_prompt, thumbnails)