diff --git a/vllm_omni/model_executor/models/qwen3_tts/qwen3_tts_code2wav.py b/vllm_omni/model_executor/models/qwen3_tts/qwen3_tts_code2wav.py
index 79f0f4a8def..ad9f30a2a9f 100644
--- a/vllm_omni/model_executor/models/qwen3_tts/qwen3_tts_code2wav.py
+++ b/vllm_omni/model_executor/models/qwen3_tts/qwen3_tts_code2wav.py
@@ -324,12 +324,18 @@ def forward(
             multimodal_outputs={"model_outputs": audios, "sr": srs},
         )
 
-    def make_omni_output(self, model_outputs: torch.Tensor | OmniOutput, **kwargs: Any) -> OmniOutput:
+    def make_omni_output(self, model_outputs: torch.Tensor | OmniOutput | tuple, **kwargs: Any) -> OmniOutput:
         if isinstance(model_outputs, OmniOutput):
             return model_outputs
 
+        if isinstance(model_outputs, tuple) and len(model_outputs) == len(OmniOutput._fields):
+            return OmniOutput(*model_outputs)
+
         if not (isinstance(model_outputs, tuple) and len(model_outputs) == 2):
-            raise TypeError(f"Qwen3TTSCode2Wav expected (audio_tensor, sr) outputs, got {type(model_outputs)}")
+            raise TypeError(
+                "Qwen3TTSCode2Wav expected OmniOutput, OmniOutput tuple, "
+                f"or (audio_tensor, sr) outputs, got {type(model_outputs)}"
+            )
 
         audio_tensor, sr = model_outputs
         return OmniOutput(