Make audio processing nodes handle None -inputs (#13879)

comfy_extras/nodes_audio.py

@@ -82,6 +82,8 @@ class VAEEncodeAudio(IO.ComfyNode):
 
     @classmethod
     def execute(cls, vae, audio) -> IO.NodeOutput:
+        if audio is None:
+            raise ValueError("VAEEncodeAudio: input audio is None (source video may have no audio track).")
         sample_rate = audio["sample_rate"]
         vae_sample_rate = getattr(vae, "audio_sample_rate", 44100)
         if vae_sample_rate != sample_rate:
@@ -171,6 +173,8 @@ class SaveAudio(IO.ComfyNode):
 
     @classmethod
     def execute(cls, audio, filename_prefix="ComfyUI", format="flac") -> IO.NodeOutput:
+        if audio is None:
+            raise ValueError("SaveAudio: input audio is None (source video may have no audio track).")
         return IO.NodeOutput(
             ui=UI.AudioSaveHelper.get_save_audio_ui(audio, filename_prefix=filename_prefix, cls=cls, format=format)
         )
@@ -198,6 +202,8 @@ class SaveAudioMP3(IO.ComfyNode):
 
     @classmethod
     def execute(cls, audio, filename_prefix="ComfyUI", format="mp3", quality="128k") -> IO.NodeOutput:
+        if audio is None:
+            raise ValueError("SaveAudioMP3: input audio is None (source video may have no audio track).")
         return IO.NodeOutput(
             ui=UI.AudioSaveHelper.get_save_audio_ui(
                 audio, filename_prefix=filename_prefix, cls=cls, format=format, quality=quality
@@ -226,6 +232,8 @@ class SaveAudioOpus(IO.ComfyNode):
 
     @classmethod
     def execute(cls, audio, filename_prefix="ComfyUI", format="opus", quality="V3") -> IO.NodeOutput:
+        if audio is None:
+            raise ValueError("SaveAudioOpus: input audio is None (source video may have no audio track).")
         return IO.NodeOutput(
             ui=UI.AudioSaveHelper.get_save_audio_ui(
                 audio, filename_prefix=filename_prefix, cls=cls, format=format, quality=quality
@@ -252,6 +260,8 @@ class PreviewAudio(IO.ComfyNode):
 
     @classmethod
     def execute(cls, audio) -> IO.NodeOutput:
+        if audio is None:
+            raise ValueError("PreviewAudio: input audio is None (source video may have no audio track).")
         return IO.NodeOutput(ui=UI.PreviewAudio(audio, cls=cls))
 
     save_flac = execute  # TODO: remove
@@ -392,21 +402,26 @@ class TrimAudioDuration(IO.ComfyNode):
 
     @classmethod
     def execute(cls, audio, start_index, duration) -> IO.NodeOutput:
+        if audio is None:
+            return IO.NodeOutput(None)
         waveform = audio["waveform"]
         sample_rate = audio["sample_rate"]
         audio_length = waveform.shape[-1]
 
+        if audio_length == 0:
+            return IO.NodeOutput(audio)
+
         if start_index < 0:
             start_frame = audio_length + int(round(start_index * sample_rate))
         else:
             start_frame = int(round(start_index * sample_rate))
-        start_frame = max(0, min(start_frame, audio_length - 1))
+        start_frame = max(0, min(start_frame, audio_length))
 
         end_frame = start_frame + int(round(duration * sample_rate))
         end_frame = max(0, min(end_frame, audio_length))
 
         if start_frame >= end_frame:
-            raise ValueError("AudioTrim: Start time must be less than end time and be within the audio length.")
+            raise ValueError("TrimAudioDuration: Start time must be less than end time and be within the audio length.")
 
         return IO.NodeOutput({"waveform": waveform[..., start_frame:end_frame], "sample_rate": sample_rate})
 
@@ -433,11 +448,13 @@ class SplitAudioChannels(IO.ComfyNode):
 
     @classmethod
     def execute(cls, audio) -> IO.NodeOutput:
+        if audio is None:
+            return IO.NodeOutput(None, None)
         waveform = audio["waveform"]
         sample_rate = audio["sample_rate"]
 
         if waveform.shape[1] != 2:
-            raise ValueError("AudioSplit: Input audio has only one channel.")
+            raise ValueError(f"AudioSplit: Input audio must be stereo (2 channels), got {waveform.shape[1]} channel(s).")
 
         left_channel = waveform[..., 0:1, :]
         right_channel = waveform[..., 1:2, :]
@@ -465,6 +482,12 @@ class JoinAudioChannels(IO.ComfyNode):
 
     @classmethod
     def execute(cls, audio_left, audio_right) -> IO.NodeOutput:
+        if audio_left is None and audio_right is None:
+            return IO.NodeOutput(None)
+        if audio_left is None:
+            return IO.NodeOutput(audio_right)
+        if audio_right is None:
+            return IO.NodeOutput(audio_left)
         waveform_left = audio_left["waveform"]
         sample_rate_left = audio_left["sample_rate"]
         waveform_right = audio_right["waveform"]
@@ -538,6 +561,12 @@ class AudioConcat(IO.ComfyNode):
 
     @classmethod
     def execute(cls, audio1, audio2, direction) -> IO.NodeOutput:
+        if audio1 is None and audio2 is None:
+            return IO.NodeOutput(None)
+        if audio1 is None:
+            return IO.NodeOutput(audio2)
+        if audio2 is None:
+            return IO.NodeOutput(audio1)
         waveform_1 = audio1["waveform"]
         waveform_2 = audio2["waveform"]
         sample_rate_1 = audio1["sample_rate"]
@@ -585,6 +614,12 @@ class AudioMerge(IO.ComfyNode):
 
     @classmethod
     def execute(cls, audio1, audio2, merge_method) -> IO.NodeOutput:
+        if audio1 is None and audio2 is None:
+            return IO.NodeOutput(None)
+        if audio1 is None:
+            return IO.NodeOutput(audio2)
+        if audio2 is None:
+            return IO.NodeOutput(audio1)
         waveform_1 = audio1["waveform"]
         waveform_2 = audio2["waveform"]
         sample_rate_1 = audio1["sample_rate"]
@@ -595,6 +630,9 @@ class AudioMerge(IO.ComfyNode):
         length_1 = waveform_1.shape[-1]
         length_2 = waveform_2.shape[-1]
 
+        if length_1 == 0 or length_2 == 0:
+            return IO.NodeOutput({"waveform": waveform_1, "sample_rate": output_sample_rate})
+
         if length_2 > length_1:
             logging.info(f"AudioMerge: Trimming audio2 from {length_2} to {length_1} samples to match audio1 length.")
             waveform_2 = waveform_2[..., :length_1]
@@ -646,6 +684,8 @@ class AudioAdjustVolume(IO.ComfyNode):
 
     @classmethod
     def execute(cls, audio, volume) -> IO.NodeOutput:
+        if audio is None:
+            return IO.NodeOutput(None)
         if volume == 0:
             return IO.NodeOutput(audio)
         waveform = audio["waveform"]
@@ -729,8 +769,14 @@ class AudioEqualizer3Band(IO.ComfyNode):
 
     @classmethod
     def execute(cls, audio, low_gain_dB, low_freq, mid_gain_dB, mid_freq, mid_q, high_gain_dB, high_freq) -> IO.NodeOutput:
+        if audio is None:
+            return IO.NodeOutput(None)
         waveform = audio["waveform"]
         sample_rate = audio["sample_rate"]
+
+        if waveform.shape[-1] == 0:
+            return IO.NodeOutput(audio)
+
         eq_waveform = waveform.clone()
 
         # 1. Apply Low Shelf (Bass)

tests-unit/comfy_test/model_detection_test.py

@@ -1,9 +1,23 @@
+from collections import defaultdict
+
 import torch
 
 from comfy.model_detection import detect_unet_config, model_config_from_unet_config
 import comfy.supported_models
 
 
+def _freeze(value):
+    """Recursively convert a value to a hashable form so configs can be
+    compared/used as dict keys or set members."""
+    if isinstance(value, dict):
+        return frozenset((k, _freeze(v)) for k, v in value.items())
+    if isinstance(value, (list, tuple)):
+        return tuple(_freeze(v) for v in value)
+    if isinstance(value, set):
+        return frozenset(_freeze(v) for v in value)
+    return value
+
+
 def _make_longcat_comfyui_sd():
     """Minimal ComfyUI-format state dict for pre-converted LongCat-Image weights."""
     sd = {}
@@ -110,3 +124,21 @@ class TestModelDetection:
         model_config = model_config_from_unet_config(unet_config, sd)
         assert model_config is not None
         assert type(model_config).__name__ == "FluxSchnell"
+
+    def test_unet_config_and_required_keys_combination_is_unique(self):
+        """Each model in the registry must have a unique combination of
+        ``unet_config`` and ``required_keys``. If two models share the same
+        combination, ``BASE.matches`` cannot disambiguate between them and the
+        first one in the list will always win."""
+        models = comfy.supported_models.models
+        groups = defaultdict(list)
+        for model in models:
+            key = (_freeze(model.unet_config), _freeze(model.required_keys))
+            groups[key].append(model.__name__)
+
+        duplicates = {k: names for k, names in groups.items() if len(names) > 1}
+        assert not duplicates, (
+            "Found models sharing the same (unet_config, required_keys) "
+            "combination, which makes detection ambiguous: "
+            + "; ".join(", ".join(names) for names in duplicates.values())
+        )