update system prompt

app.py

@@ -421,19 +421,22 @@ def detect_objects(image, dino_proc, dino_mod, threshold=0.3) -> tuple:
 def fuse_captions(cap1: str, cap2: str, objects: str, qwen_tok, qwen_mod) -> str:
 
     system_prompt = (
-        "You are helping write image captions. "
-        "Write ONE short, simple, factual caption exactly as a person would "
-        "naturally describe this photo. Use plain everyday language. "
-        "Do not add any details that are not clearly visible. "
-        "Do not use dramatic or poetic language. "
-        "Return ONLY the caption, nothing else."
-    )
-    user_prompt = (
-        f"Caption A: {cap1}\n"
-        f"Caption B: {cap2}\n"
-        f"{objects}\n\n"
-        "Write a simple natural caption:"
-    )
+    "You write image captions. "
+    "Look at the two captions and detected objects provided. "
+    "Write ONE caption that covers: who is in the image, what they are doing, "
+    "what objects are around them, and where the scene is taking place. "
+    "Use simple, everyday words. Write 2 to 3 sentences. "
+    "Only describe what is clearly visible. "
+    "Do not guess, invent, or add dramatic language. "
+    "Return ONLY the caption, nothing else."
+)
+
+user_prompt = (
+    f"Caption A: {cap1}\n"
+    f"Caption B: {cap2}\n"
+    f"{objects}\n\n"
+    "Write a clear, natural caption covering the person, action, objects and setting:"
+)
 
     try:
         messages = [