diff --git a/nodes/generator.py b/nodes/generator.py
index 32a02a9..5239f90 100644
--- a/nodes/generator.py
+++ b/nodes/generator.py
@@ -33,12 +33,17 @@ class OmniVoiceGenerate:
     def generate(self, model, text, mode, ref_audio=None, ref_text="", instruct="", speed=1.0, num_step=32):
         kwargs = {"text": text, "speed": speed, "num_step": num_step}
 
+        if mode == "voice_cloning" and ref_audio is None:
+            raise ValueError("voice_cloning mode requires ref_audio to be connected")
+        if mode == "voice_design" and not instruct:
+            raise ValueError("voice_design mode requires an instruct string (e.g. 'female, low pitch')")
+
         if mode == "voice_cloning" and ref_audio is not None:
             tmp = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
             tmp_path = tmp.name
             tmp.close()
             try:
-                waveform = ref_audio["waveform"].squeeze(0)  # (channels, samples)
+                waveform = ref_audio["waveform"].squeeze(0).cpu()  # (channels, samples)
                 torchaudio.save(tmp_path, waveform, ref_audio["sample_rate"])
                 kwargs["ref_audio"] = tmp_path
                 if ref_text:
diff --git a/tests/test_generator.py b/tests/test_generator.py
index eda3a12..6c1a306 100644
--- a/tests/test_generator.py
+++ b/tests/test_generator.py
@@ -1,5 +1,5 @@
 # tests/test_generator.py
-from unittest.mock import patch, MagicMock, call
+from unittest.mock import patch, MagicMock
 import torch
 import pytest
 from nodes.generator import OmniVoiceGenerate
@@ -92,6 +92,32 @@ def test_generate_voice_cloning():
     assert "ref_audio" in call_kwargs
 
 
+def test_voice_cloning_without_ref_audio_raises():
+    node = OmniVoiceGenerate()
+    mock_model = make_mock_model()
+    with pytest.raises(ValueError, match="ref_audio"):
+        node.generate(
+            model=mock_model,
+            text="Hello",
+            mode="voice_cloning",
+            speed=1.0,
+            num_step=32,
+        )
+
+
+def test_voice_design_without_instruct_raises():
+    node = OmniVoiceGenerate()
+    mock_model = make_mock_model()
+    with pytest.raises(ValueError, match="instruct"):
+        node.generate(
+            model=mock_model,
+            text="Hello",
+            mode="voice_design",
+            speed=1.0,
+            num_step=32,
+        )
+
+
 def test_output_waveform_shape():
     node = OmniVoiceGenerate()
     # Simulate two chunks returned by OmniVoice