Update app.py

app.py

@@ -118,12 +118,12 @@ class TextImageRequest(BaseModel):
     voice: str = "af_heart"  # Default voice that we know exists
     speed: float = 1.0
 
-    # Annotated as a ClassVar so Pydantic doesn't treat it as a model field.
+    # Annotate as a ClassVar so Pydantic ignores it as a field.
     AVAILABLE_VOICES: ClassVar[List[str]] = ["af_heart"]
 
     def validate_voice(self):
         if self.voice not in self.AVAILABLE_VOICES:
-            return "af_heart"  # Default to a known available voice
+            return "af_heart"
         return self.voice
 
 class AudioResponse(BaseModel):
@@ -158,7 +158,7 @@ def llm_chat_response(text, image_base64=None):
         
         logger.info("Initializing InferenceClient...")
         client = InferenceClient(
-            provider="hf-inference",  # Using the correct provider
+            provider="hf-inference",  # Using the correct provider as per sample
             api_key=HF_TOKEN
         )
         
@@ -178,12 +178,8 @@ def llm_chat_response(text, image_base64=None):
             base_url = os.getenv("BASE_URL", "http://localhost:8000")
             image_url = f"{base_url}/static/{filename}"
             prompt = text if text else "Describe this image in one sentence."
-            # Include a system message to provide conversation context
+            # Construct message exactly as in the reference
             messages = [
-                {
-                    "role": "system",
-                    "content": "You are a helpful assistant that describes images and answers questions about them."
-                },
                 {
                     "role": "user",
                     "content": [