From 665f9f90ab532ccd3904abb01ce70f50ecfeef44 Mon Sep 17 00:00:00 2001
From: RyanOnTheInside <7623207+ryanontheinside@users.noreply.github.com>
Date: Mon, 9 Feb 2026 09:15:34 -0500
Subject: [PATCH] fix: localize annoying imports

todo: replace librosa with numpy and torch native functions
---
 nodes/audio/audio_nodes.py            |  5 ++--
 nodes/audio/audio_nodes_utility.py    |  4 ++-
 nodes/audio/audio_processor_legacy.py | 40 ++++++++++++++-------------
 nodes/audio/audio_utils.py            |  2 +-
 nodes/audio/midi_nodes.py             |  5 +++-
 nodes/flex/feature_extractors_midi.py |  1 -
 nodes/flex/features_audio.py          | 15 +++++++---
 7 files changed, 43 insertions(+), 29 deletions(-)

diff --git a/nodes/audio/audio_nodes.py b/nodes/audio/audio_nodes.py
index 99adc59..e054dd5 100644
--- a/nodes/audio/audio_nodes.py
+++ b/nodes/audio/audio_nodes.py
@@ -1,7 +1,7 @@
 import torch
 import numpy as np
 from scipy.ndimage import zoom
-from .audio_processor_legacy import AudioVisualizer, PygameAudioVisualizer
+
 from scipy import signal
 import comfy.model_management as mm
 import folder_paths
@@ -421,7 +421,8 @@ class AudioFeatureVisualizer(AudioNodeBase):
 
     def visualize_audio_feature(self, audio, video_frames, visualization_type, frame_rate, x_axis, y_axis, cmap, visualizer):
         num_frames, height, width, _ = video_frames.shape
-        if visualizer == "pygame":  
+        from .audio_processor_legacy import AudioVisualizer, PygameAudioVisualizer
+        if visualizer == "pygame":
             visualizer = PygameAudioVisualizer(audio, num_frames, height, width, frame_rate)
         elif visualizer == "matplotlib":
             visualizer = AudioVisualizer(audio, num_frames, height, width, frame_rate, x_axis, y_axis, cmap)
diff --git a/nodes/audio/audio_nodes_utility.py b/nodes/audio/audio_nodes_utility.py
index 9c59ff6..b181244 100644
--- a/nodes/audio/audio_nodes_utility.py
+++ b/nodes/audio/audio_nodes_utility.py
@@ -11,7 +11,6 @@ from .audio_utils import (
 )
 import torch
 from ...tooltips import apply_tooltips
-import librosa
 import numpy as np
 
 class AudioUtility(AudioNodeBase):
@@ -232,6 +231,7 @@ class AudioInfo(AudioUtility):
 
     def _detect_key(self, audio_mono, sample_rate):
         """Detect musical key using chromagram and Krumhansl-Kessler key profiles."""
+        import librosa
         # Extract chromagram using Constant-Q Transform (better for pitch detection)
         chromagram = librosa.feature.chroma_cqt(y=audio_mono, sr=sample_rate)
 
@@ -269,6 +269,7 @@ class AudioInfo(AudioUtility):
         return best_key
 
     def get_audio_info(self, audio, frame_rate):
+        import librosa
         # Get basic audio info
         waveform = audio['waveform']
         sample_rate = audio['sample_rate']
@@ -399,6 +400,7 @@ class Knob(AudioUtility):
     FUNCTION = "enhance_audio"
 
     def enhance_audio(self, audio, knob, other_knob):
+        import librosa
         waveform, sample_rate = audio['waveform'], audio['sample_rate']
         
         # Make a copy of the input waveform
diff --git a/nodes/audio/audio_processor_legacy.py b/nodes/audio/audio_processor_legacy.py
index cb2b3df..d822973 100644
--- a/nodes/audio/audio_processor_legacy.py
+++ b/nodes/audio/audio_processor_legacy.py
@@ -1,14 +1,8 @@
 import numpy as np
 import torch
-import librosa
 import cv2
 import matplotlib.pyplot as plt
 from comfy.utils import ProgressBar
-import numpy as np
-import librosa
-import torch
-import cv2
-import matplotlib.pyplot as plt
 from ... import ProgressMixin
 
 
@@ -42,13 +36,7 @@ class BaseAudioProcessor(ProgressMixin):
         end_sample = int(end_time * self.sample_rate)
         return self.audio[start_sample:end_sample]
 
-import librosa
-import librosa.display
-import matplotlib.pyplot as plt
-import numpy as np
-import torch
-
-class AudioVisualizer(BaseAudioProcessor):  
+class AudioVisualizer(BaseAudioProcessor):
 
     def __init__(self, audio, num_frames, height, width, frame_rate, x_axis, y_axis, cmap):
         super().__init__(audio, num_frames, height, width, frame_rate)
@@ -58,6 +46,8 @@ class AudioVisualizer(BaseAudioProcessor):
         self.cmap = cmap
 
     def _generate_frame(self, data):
+        import librosa
+        import librosa.display
         fig, ax = plt.subplots(figsize=(self.width / 100, self.height / 100), dpi=100)
         librosa.display.specshow(data, sr=self.sample_rate, x_axis=self.x_axis, y_axis=self.y_axis, cmap=self.cmap, ax=ax)
         if self.x_axis == 'off':
@@ -81,6 +71,7 @@ class AudioVisualizer(BaseAudioProcessor):
         return frame
 
     def create_spectrogram(self):
+        import librosa
         frames = []
         self.start_progress(self.num_frames, "Creating Spectrogram")
         for i in range(self.num_frames):
@@ -97,6 +88,8 @@ class AudioVisualizer(BaseAudioProcessor):
         return torch.from_numpy(frames).float() / 255.0
 
     def create_waveform(self):
+        import librosa
+        import librosa.display
         frames = []
         self.start_progress(self.num_frames, "Creating Waveform")
         for i in range(self.num_frames):
@@ -130,6 +123,7 @@ class AudioVisualizer(BaseAudioProcessor):
         return torch.from_numpy(frames).float() / 255.0
 
     def create_mfcc(self):
+        import librosa
         frames = []
         self.start_progress(self.num_frames, "Creating MFCC")
         for i in range(self.num_frames):
@@ -144,6 +138,7 @@ class AudioVisualizer(BaseAudioProcessor):
         return torch.from_numpy(frames).float() / 255.0
 
     def create_chroma(self):
+        import librosa
         frames = []
         self.start_progress(self.num_frames, "Creating Chroma")
         for i in range(self.num_frames):
@@ -158,6 +153,7 @@ class AudioVisualizer(BaseAudioProcessor):
         return torch.from_numpy(frames).float() / 255.0
 
     def create_tonnetz(self):
+        import librosa
         frames = []
         self.start_progress(self.num_frames, "Creating Tonnetz")
         for i in range(self.num_frames):
@@ -172,6 +168,7 @@ class AudioVisualizer(BaseAudioProcessor):
         return torch.from_numpy(frames).float() / 255.0
 
     def create_spectral_centroid(self):
+        import librosa
         frames = []
         self.start_progress(self.num_frames, "Creating Spectral Centroid")
         for i in range(self.num_frames):
@@ -214,22 +211,20 @@ class AudioFeatureExtractor(BaseAudioProcessor):
         return np.array([np.sqrt(np.mean(self._get_audio_frame(i)**2)) for i in range(self.num_frames)])
 
     def _spectral_centroid(self):
+        import librosa
         return np.array([np.mean(librosa.feature.spectral_centroid(y=self._get_audio_frame(i), sr=self.sample_rate)[0]) for i in range(self.num_frames)])
 
     def _onset_detection(self):
+        import librosa
         return np.array([np.mean(librosa.onset.onset_strength(y=self._get_audio_frame(i), sr=self.sample_rate)) for i in range(self.num_frames)])
 
     def _chroma_features(self):
+        import librosa
         return np.array([np.mean(librosa.feature.chroma_stft(y=self._get_audio_frame(i), sr=self.sample_rate), axis=1) for i in range(self.num_frames)])
 
-import numpy as np
-import torch
-import pygame
-import librosa
-from comfy.utils import ProgressBar
-
 class PygameAudioVisualizer(BaseAudioProcessor):
     def __init__(self, audio, num_frames, height, width, frame_rate, scroll_direction='left'):
+        import pygame
         super().__init__(audio, num_frames, height, width, frame_rate)
         self.height = height
         self.width = width
@@ -244,6 +239,7 @@ class PygameAudioVisualizer(BaseAudioProcessor):
         self.total_samples = len(self.full_waveform)
 
     def _surface_to_array(self, surface):
+        import pygame
         # Convert Pygame surface to NumPy array
         frame_array = pygame.surfarray.array3d(surface)
         # Transpose to (H, W, C)
@@ -310,6 +306,7 @@ class PygameAudioVisualizer(BaseAudioProcessor):
         return torch.from_numpy(frames)
 
     def _generate_spectrogram_frame(self, audio_frame):
+        import librosa
         self.screen.fill((0, 0, 0))  # Clear the screen
         # Compute spectrogram
         n_fft = min(2048, len(audio_frame))
@@ -345,6 +342,7 @@ class PygameAudioVisualizer(BaseAudioProcessor):
         return torch.from_numpy(frames)
 
     def _generate_mfcc_frame(self, audio_frame):
+        import librosa
         self.screen.fill((0, 0, 0))  # Clear the screen
         # Compute MFCCs
         mfccs = librosa.feature.mfcc(y=audio_frame, sr=self.sample_rate, n_mfcc=20)
@@ -377,6 +375,7 @@ class PygameAudioVisualizer(BaseAudioProcessor):
         return torch.from_numpy(frames)
 
     def _generate_chroma_frame(self, audio_frame):
+        import librosa
         self.screen.fill((0, 0, 0))  # Clear the screen
         # Compute Chroma features
         chromagram = librosa.feature.chroma_stft(y=audio_frame, sr=self.sample_rate)
@@ -409,6 +408,7 @@ class PygameAudioVisualizer(BaseAudioProcessor):
         return torch.from_numpy(frames)
 
     def _generate_tonnetz_frame(self, audio_frame):
+        import librosa
         self.screen.fill((0, 0, 0))  # Clear the screen
         # Compute Tonnetz
         tonnetz = librosa.feature.tonnetz(y=audio_frame, sr=self.sample_rate)
@@ -441,6 +441,7 @@ class PygameAudioVisualizer(BaseAudioProcessor):
         return torch.from_numpy(frames)
 
     def _generate_spectral_centroid_frame(self, audio_frame):
+        import librosa
         self.screen.fill((0, 0, 0))  # Clear the screen
         # Compute Spectral Centroid
         centroid = librosa.feature.spectral_centroid(y=audio_frame, sr=self.sample_rate)
@@ -459,4 +460,5 @@ class PygameAudioVisualizer(BaseAudioProcessor):
         return self.screen.copy()
 
     def __del__(self):
+        import pygame
         pygame.quit()
\ No newline at end of file
diff --git a/nodes/audio/audio_utils.py b/nodes/audio/audio_utils.py
index 7262ee0..723fba3 100644
--- a/nodes/audio/audio_utils.py
+++ b/nodes/audio/audio_utils.py
@@ -2,7 +2,6 @@ import torch
 import torchaudio
 import torchaudio.transforms as T
 import torchaudio.functional as F
-import librosa
 import numpy as np
 import torch.nn.functional as nnf
 
@@ -85,6 +84,7 @@ def time_stretch(waveform, rate):
     device = waveform.device
     waveform = waveform.cpu()
 
+    import librosa
     # Convert to numpy for librosa processing
     waveform_np = waveform.numpy()
 
diff --git a/nodes/audio/midi_nodes.py b/nodes/audio/midi_nodes.py
index e6029d1..0083ab6 100644
--- a/nodes/audio/midi_nodes.py
+++ b/nodes/audio/midi_nodes.py
@@ -1,6 +1,5 @@
 import torch
 import numpy as np
-import mido
 import os
 import folder_paths
 import math
@@ -350,6 +349,7 @@ class MIDILoader:
     CATEGORY = "RyanOnTheInside/Audio/MIDI"
     
     def load_midi(self, midi_file, track_selection, start_measure=1, start_beat=1, end_measure=0, end_beat=1):
+        import mido
         try:
             midi_path = folder_paths.get_full_path("midi_files", midi_file)
             if not midi_path or not os.path.exists(midi_path):
@@ -491,6 +491,7 @@ class MIDILoader:
     
     @classmethod
     def analyze_midi(cls, midi_path, start_measure=1, start_beat=1, end_measure=0, end_beat=1):
+        import mido
         midi_data = mido.MidiFile(midi_path)
         
         # Get the total measures
@@ -546,6 +547,7 @@ class MIDILoader:
     
     @classmethod
     def VALIDATE_INPUTS(cls, midi_file, track_selection, start_measure=1, start_beat=1, end_measure=0, end_beat=1):
+        import mido
         midi_path = folder_paths.get_full_path("midi_files", midi_file)
         if not midi_path or not os.path.isfile(midi_path):
             return f"MIDI file not found: {midi_file}"
@@ -642,6 +644,7 @@ async def upload_midi(request):
     
 @routes.post('/refresh_midi_data')
 async def refresh_midi_data(request):
+    import mido
     data = await request.json()
     midi_file = data.get('midi_file')
     track_selection = data.get('track_selection')
diff --git a/nodes/flex/feature_extractors_midi.py b/nodes/flex/feature_extractors_midi.py
index 7d9c1fb..e7a4df7 100644
--- a/nodes/flex/feature_extractors_midi.py
+++ b/nodes/flex/feature_extractors_midi.py
@@ -1,4 +1,3 @@
-import mido
 import os
 import folder_paths
 from .features_midi import MIDIFeature
diff --git a/nodes/flex/features_audio.py b/nodes/flex/features_audio.py
index 2baf3cf..3acd633 100644
--- a/nodes/flex/features_audio.py
+++ b/nodes/flex/features_audio.py
@@ -1,12 +1,10 @@
 from .features import BaseFeature
-import librosa
 import numpy as np
 from scipy.signal import hilbert
 from scipy import signal
 import functools
 from abc  import ABC, abstractmethod
 import numpy as np
-import librosa
 from scipy.signal import medfilt, hilbert
 from scipy import signal
 import functools
@@ -96,6 +94,7 @@ class AudioFeature(BaseAudioFeature):
         return self
 
     def _calculate_feature(self, frame):
+        import librosa
         if frame.size == 0:
             return 0.0
         if self.feature_name == 'amplitude_envelope':
@@ -125,7 +124,6 @@ class AudioFeature(BaseAudioFeature):
         self.features[self.feature_name] = normalized.tolist()
 
 import numpy as np
-import librosa
 
 class RhythmFeature(BaseAudioFeature):
     def __init__(self, feature_name, audio, frame_count, frame_rate, width, height, feature_type='beat_locations', time_signature=4):
@@ -150,8 +148,9 @@ class RhythmFeature(BaseAudioFeature):
         ]
 
     def extract(self):
+        import librosa
         self.features = {}
-        
+
         # Extract basic rhythm features
         tempo, beat_frames = librosa.beat.beat_track(y=self.audio_array, sr=self.sample_rate)
         onset_env = librosa.onset.onset_strength(y=self.audio_array, sr=self.sample_rate)
@@ -178,6 +177,7 @@ class RhythmFeature(BaseAudioFeature):
         return self
 
     def _extract_beat_locations(self, beat_frames):
+        import librosa
         beat_times = librosa.frames_to_time(beat_frames, sr=self.sample_rate)
         beat_sequence = np.zeros(self.frame_count)
         for beat_time in beat_times:
@@ -200,6 +200,7 @@ class RhythmFeature(BaseAudioFeature):
         self.features[self.feature_name] = resampled_onset.tolist()
 
     def _extract_beat_emphasis(self, beat_frames, onset_env):
+        import librosa
         beat_emphasis = np.zeros(self.frame_count)
         for beat in beat_frames:
             if beat < len(onset_env):
@@ -224,12 +225,14 @@ class RhythmFeature(BaseAudioFeature):
         self.features[self.feature_name] = resampled_syncopation.tolist()
 
     def _extract_rhythm_regularity(self, beat_frames):
+        import librosa
         # Measure regularity by calculating the standard deviation of inter-beat intervals
         ibi = np.diff(librosa.frames_to_time(beat_frames, sr=self.sample_rate))
         regularity = 1 / (1 + np.std(ibi))  # Invert so that higher values mean more regular
         self.features[self.feature_name] = [regularity] * self.frame_count
 
     def _extract_beat_types(self, beat_frames):
+        import librosa
         beat_times = librosa.frames_to_time(beat_frames, sr=self.sample_rate)
         down_beats = np.zeros(self.frame_count)
         up_beats = np.zeros(self.frame_count)
@@ -327,11 +330,13 @@ class PitchFeature(BaseAudioFeature):
     
     @classmethod
     def pitch_to_note(cls, pitch):
+        import librosa
         if pitch == 0:
             return "N/A"
         return librosa.hz_to_note(pitch)
 
     def _calculate_pitch_sequence(self):
+        import librosa
         # Try to use CREPE model if available
         if self.crepe_model != "none":
             try:
@@ -385,6 +390,7 @@ class PitchFeature(BaseAudioFeature):
         self.features[self.feature_name + '_confidence'] = confidences.tolist()
 
     def _fallback_pitch_estimation(self):
+        import librosa
         pitches, magnitudes = librosa.piptrack(
             y=self.audio_array,
             sr=self.sample_rate,
@@ -429,6 +435,7 @@ class PitchFeature(BaseAudioFeature):
         self.features[self.feature_name] = feature_values.tolist()
 
     def _extract_semitone(self):
+        import librosa
         pitches = np.array(self.features[self.feature_name + '_pitch'])
         midi_notes = librosa.hz_to_midi(pitches)
         midi_notes = np.round(midi_notes)