Prompt changes

frigate/data_processing/post/review_descriptions.py

@@ -85,7 +85,7 @@ class ReviewDescriptionProcessor(PostProcessorApi):
                 return
             elif (
                 final_data["severity"] == "detection"
-                and not self.config.cameras[camera].review.detections
+                and not self.config.cameras[camera].review.genai.detections
             ):
                 self.tracked_review_items.pop(id)
                 return

frigate/data_processing/post/types.py

@@ -3,10 +3,10 @@ from pydantic import BaseModel, Field
 
 class ReviewMetadata(BaseModel):
     scene: str = Field(
-        description="A concise summary of the overall scene. This should be a single string of text."
+        description="A comprehensive description of the setting and entities, including relevant context and plausible inferences if supported by visual evidence."
     )
-    action: str = Field(
-        description="A concise description of the primary action or event happening in the scene. This should be a single string of text."
+    confidence: float = Field(
+        description="A float between 0 and 1 representing your overall confidence in this analysis."
     )
     potential_threat_level: int | None = Field(
         default=None,

frigate/genai/__init__.py

@@ -39,25 +39,34 @@ class GenAIClient:
     ) -> None:
         """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 and the objects interacting with it.
-        Focus solely on observable actions, visible entities, and the environment.
+        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.
+
+        When forming your description:
+        - Base all statements on visible details in the images.
+        - 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).
+        - Avoid implying hostility, criminal intent, or other strong judgments unless the visual evidence is unambiguous.
+        - Distinguish between what is certain (facts) and what appears likely (reasonable inference).
 
         Here is some information we already know:
-        - the following activity occurred at {review_data['timestamp'].strftime('%I:%M %p')}
-        - the following objects were detected: {review_data['objects']}
-        - the following recognized objects were detected: {review_data['recognized_objects']}
-        - the activity happened in the following zones: {review_data['zones']}
+        - The activity occurred at {review_data["timestamp"].strftime("%I:%M %p")}
+        - The following objects were detected: {review_data["objects"]}
+        - The following recognized objects were detected: {review_data["recognized_objects"]}
+        - The activity happened in the following zones: {review_data["zones"]}
 
         Your response **MUST** be a flat JSON object with the following fields:
-        - `scene` (string): A single, comprehensive description of the entire visual scene.
-        - `action` (string): A single description of any key actions or movements.
-        - `potential_threat_level` (integer, optional): An integer from 0 to 3. Only include if a clear security concern is visible. Omit if no threat.
+        - `scene` (string): A comprehensive description of the setting and entities, including relevant context and plausible inferences if supported by visual evidence.
+        - `confidence` (float): A float between 0 and 1 representing your overall confidence in this analysis.
+        - `potential_threat_level` (integer, optional): 1–3 scale. Only include if a clear security concern is visible:
+        - 1 = Unusual but not overtly threatening
+        - 2 = Suspicious or potentially harmful
+        - 3 = Clear and immediate threat
+        Omit this field entirely if there is no observable security concern.
 
-        **IMPORTANT:** The value for each field (e.g., "scene", "action") must be a plain string or integer, NOT another JSON object or a description of the field itself.
-
-        Provide the response in the exact JSON format specified by this schema:
-        {ReviewMetadata.model_json_schema()['properties']}
+        **IMPORTANT:**
+        - Values for each field must be plain strings or integers — no nested objects or explanatory text.
+        - The JSON must strictly match this structure:
+        {ReviewMetadata.model_json_schema()["properties"]}
         """
         logger.info(f"processing {review_data}")
         logger.info(f"Got GenAI review: {self._send(context_prompt, thumbnails)}")