nodes/kijai-ComfyUI-WanVideoWrapper
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Squashed commit of the following:

Browse Source 
commit fd32b14fdc
Author: kijai <40791699+kijai@users.noreply.github.com>
Date:   Tue Dec 23 02:02:31 2025 +0200

    Clean prints

commit 1776695e26
Author: kijai <40791699+kijai@users.noreply.github.com>
Date:   Tue Dec 23 01:48:12 2025 +0200

    Update nodes_model_loading.py

commit ef36204fa8
Author: kijai <40791699+kijai@users.noreply.github.com>
Date:   Tue Dec 23 01:35:08 2025 +0200

    Reduce peak VRAM use

commit c6f32c1424
Author: kijai <40791699+kijai@users.noreply.github.com>
Date:   Mon Dec 22 23:53:41 2025 +0200

    Norm dtype

commit 6d4a0f6e53
Merge: e7e0006 3e45021
Author: kijai <40791699+kijai@users.noreply.github.com>
Date:   Mon Dec 22 22:11:38 2025 +0200

    Merge branch 'main' into longcat_avatar

commit e7e00061e5
Author: kijai <40791699+kijai@users.noreply.github.com>
Date:   Mon Dec 22 00:43:01 2025 +0200

    Update nodes_sampler.py

commit eb5ec262a0
Merge: 7c0ba84 fed3b22
Author: kijai <40791699+kijai@users.noreply.github.com>
Date:   Mon Dec 22 00:42:53 2025 +0200

    Merge branch 'main' into longcat_avatar

commit 7c0ba84a26
Author: kijai <40791699+kijai@users.noreply.github.com>
Date:   Sun Dec 21 23:00:43 2025 +0200

    remove prints

commit 06a86923e7
Author: kijai <40791699+kijai@users.noreply.github.com>
Date:   Sun Dec 21 22:53:25 2025 +0200

    Fix ref latent

    oops

commit dca3106f10
Author: kijai <40791699+kijai@users.noreply.github.com>
Date:   Sat Dec 20 18:46:32 2025 +0200

    Expose more options, make vid2vid easier

commit 175418b8d2
Author: kijai <40791699+kijai@users.noreply.github.com>
Date:   Sat Dec 20 03:15:24 2025 +0200

    Create LongCatAvatar_testing_wip.json

commit 4a6e2d3c6c
Author: kijai <40791699+kijai@users.noreply.github.com>
Date:   Sat Dec 20 03:14:49 2025 +0200

    Init
This commit is contained in:
kijai
2025-12-23 12:17:07 +02:00
parent 3e45021422
commit 5360eeb345
11 changed files with 6238 additions and 144 deletions
Download Patch File Download Diff File Expand all files Collapse all files
multitalk/nodes.py
+19 -1
View File
@@ -7,6 +7,8 @@ from ..utils import log, set_module_tensor_to_device
import os
import json
import datetime
import scipy.signal as ss
import numpy as np
script_directory = os.path.dirname(os.path.abspath(__file__))
folder_paths.add_model_folder_path("wav2vec2", os.path.join(folder_paths.models_dir, "wav2vec2"))
@@ -134,6 +136,15 @@ def loudness_norm(audio_array, sr=16000, lufs=-23):
return audio_array
normalized_audio = pyloudnorm.normalize.loudness(audio_array, loudness, lufs)
return normalized_audio
def _add_noise_floor(audio, noise_db=-45):
noise_amp = 10 ** (noise_db / 20)
noise = np.random.randn(len(audio)) * noise_amp
return audio + noise
def _smooth_transients(audio, sr=16000):
b, a = ss.butter(3, 3000 / (sr/2))
return ss.lfilter(b, a, audio)
class MultiTalkWav2VecEmbeds:
@classmethod
@@ -153,6 +164,8 @@ class MultiTalkWav2VecEmbeds:
"audio_3": ("AUDIO",),
"audio_4": ("AUDIO",),
"ref_target_masks": ("MASK", {"tooltip": "Per-speaker semantic mask(s) in pixel space. Supply one mask per speaker (plus optional background) to guide mouth assignment"}),
"add_noise_floor": ("BOOLEAN", {"default": False, "tooltip": "Add a low-level noise floor to the audio to reduce silent gaps"}),
"smooth_transients": ("BOOLEAN", {"default": False, "tooltip": "Apply a low-pass filter to the audio to smooth out transients"}),
}
}
@@ -161,7 +174,8 @@ class MultiTalkWav2VecEmbeds:
FUNCTION = "process"
CATEGORY = "WanVideoWrapper"
def process(self, wav2vec_model, normalize_loudness, fps, num_frames, audio_1, audio_scale, audio_cfg_scale, multi_audio_type, audio_2=None, audio_3=None, audio_4=None, ref_target_masks=None):
def process(self, wav2vec_model, normalize_loudness, fps, num_frames, audio_1, audio_scale, audio_cfg_scale, multi_audio_type, audio_2=None, audio_3=None, audio_4=None,
ref_target_masks=None, add_noise_floor=False, smooth_transients=False):
model_type = wav2vec_model["model_type"]
if not "tencent" in model_type.lower():
raise ValueError("Only tencent wav2vec2 models supported by MultiTalk")
@@ -207,6 +221,10 @@ class MultiTalkWav2VecEmbeds:
if normalize_loudness:
audio_segment = loudness_norm(audio_segment, sr=sr)
if add_noise_floor:
audio_segment = _add_noise_floor(audio_segment, noise_db=-45)
if smooth_transients:
audio_segment = _smooth_transients(audio_segment, sr=sr)
audio_feature = np.squeeze(
wav2vec2_feature_extractor(audio_segment, sampling_rate=sr).input_values