feat: add better prompt

src/video_processor/processor.py

@@ -34,14 +34,25 @@ class VideoAnalyzer:
     def process_video(self, video_path: str, frame_interval: int = 30) -> List[Dict]:
         logger.info(f"Processing video: {video_path} with frame_interval={frame_interval}")
         try:
-            # Create message for model
-            messages = [{
-                "role": "user",
-                "content": [
-                    {"type": "video", "path": video_path},
-                    {"type": "text", "text": "Describe this video in detail - with all the timestamps and the actions happening in the video. I should be able to understand the video by reading the description, and search for it later."}
-                ]
-            }]
+            # Create message for model with detailed system prompt
+            messages = [
+                {
+                    "role": "system",
+                    "content": [
+                        {
+                            "type": "text", 
+                            "text": "You are a detailed video analysis assistant that can understand videos. Your task is to provide comprehensive descriptions including all events, actions, and important details with their timestamps. Focus on being specific and thorough."
+                        }
+                    ]
+                },
+                {
+                    "role": "user",
+                    "content": [
+                        {"type": "video", "path": video_path},
+                        {"type": "text", "text": "Please provide a detailed analysis of this video. Include:\n1. All significant actions and events\n2. Temporal information and timestamps\n3. Important visual details and context\n4. Any text or speech content if present\n5. Scene transitions and changes\nBe thorough and specific so the description can be used for detailed searching later."}
+                    ]
+                }
+            ]
             
             # Process video using chat template
             inputs = self.processor.apply_chat_template(
@@ -52,11 +63,12 @@ class VideoAnalyzer:
                 return_tensors="pt"
             ).to(self.model.device)
             
-            # Generate description
+            # Generate description with increased token limit
             generated_ids = self.model.generate(
                 **inputs, 
-                do_sample=False,
-                max_new_tokens=100
+                do_sample=True,
+                temperature=0.7,
+                max_new_tokens=512  # Increased from 100 to get more detailed descriptions
             )
             description = self.processor.batch_decode(
                 generated_ids,
@@ -64,7 +76,7 @@ class VideoAnalyzer:
             )[0]
             
             return [{
-                "description": description
+                "description": description.split("Assistant: ")[-1]  # Remove assistant prefix if present
             }]
             
         except Exception as e: