RuntimeError: cuFFT error: CUFFT_INTERNAL_ERROR

[myprivacygithub]

Custom Node Testing

• I have tried disabling custom nodes and the issue persists (see how to disable custom nodes if you need help)

Expected Behavior

RuntimeError: cuFFT error: CUFFT_INTERNAL_ERROR, since last comfyui update earlier today

Actual Behavior

when use audio seperation node / Mel-Band RoFormer Sampler in any infinitetalk/multitalk workflow, its happening

Steps to Reproduce

when use audio seperation node / Mel-Band RoFormer Sampler in any infinitetalk/multitalk workflow, its happening

Debug Logs

Traceback (most recent call last):
File "D:\ComfyUI\execution.py", line 524, in execute
output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\ComfyUI\execution.py", line 333, in get_output_data
return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\ComfyUI\execution.py", line 307, in _async_map_node_over_list
await process_inputs(input_dict, i)
File "D:\ComfyUI\execution.py", line 295, in process_inputs
result = f(**inputs)
^^^^^^^^^^^
File "D:\ComfyUI\custom_nodes\ComfyUI-MelBandRoFormer\nodes.py", line 139, in process
x = model(part.unsqueeze(0))[0]
^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\anaconda3\envs\comfyui_env\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\anaconda3\envs\comfyui_env\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\ComfyUI\custom_nodes\ComfyUI-MelBandRoFormer\model\mel_band_roformer.py", line 427, in forward
stft_repr = torch.stft(raw_audio, **self.stft_kwargs, window=stft_window, return_complex=True)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\anaconda3\envs\comfyui_env\Lib\site-packages\torch\functional.py", line 681, in stft
return _VF.stft( # type: ignore[attr-defined]
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: cuFFT error: CUFFT_INTERNAL_ERROR

Other

No response

[orzechowy3334-rgb]

Same here! Pytorch 2.9.0+cu130, python 3.13.11, RTX 5090, ComfyUI 0.15.1, after last update

PS. Actually, any audio vae encode fails :(

[rattus128]

Can you post more log?

Everything on startup until the "listening on ..." And everything from "got prompt" up to and including the crash?

[jdorpitan-code]

Hi Rattus128,
here are some logs (conf : python 3.13.9 (python embedded from the portable comfyui, torch and torchaudio both at 2.9.1+cu130)

got prompt
VAE load device: cuda:0, offload device: cpu, dtype: torch.bfloat16
I:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\functional.py:681: UserWarning: A window was not provided. A rectangular window will be applied,which is known to cause spectral leakage. Other windows such as torch.hann_window or torch.hamming_window are recommended to reduce spectral leakage.To suppress this warning and use a rectangular window, explicitly set window=torch.ones(n_fft, device=<device>). (Triggered internally at C:\actions-runner_work\pytorch\pytorch\pytorch\aten\src\ATen\native\SpectralOps.cpp:842.)
return _VF.stft( # type: ignore[attr-defined]
Processing chunks: 0%| | 0/6 [00:00<?, ?it/s]
!!! Exception during processing !!! cuFFT error: CUFFT_INTERNAL_ERROR
Traceback (most recent call last):
File "I:\ComfyUI_windows_portable\ComfyUI\execution.py", line 524, in execute
output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "I:\ComfyUI_windows_portable\ComfyUI\execution.py", line 333, in get_output_data
return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "I:\ComfyUI_windows_portable\ComfyUI\execution.py", line 307, in _async_map_node_over_list
await process_inputs(input_dict, i)
File "I:\ComfyUI_windows_portable\ComfyUI\execution.py", line 295, in process_inputs
result = f(**inputs)
File "I:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-MelBandRoFormer\nodes.py", line 139, in process
x = model(part.unsqueeze(0))[0]
~~~~~^^^^^^^^^^^^^^^^^^^
File "I:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1775, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "I:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1786, in _call_impl
return forward_call(*args, **kwargs)
File "I:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-MelBandRoFormer\model\mel_band_roformer.py", line 427, in forward
stft_repr = torch.stft(raw_audio, **self.stft_kwargs, window=stft_window, return_complex=True)
File "I:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\functional.py", line 681, in stft
return _VF.stft( # type: ignore[attr-defined]
~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
input,
^^^^^^
...<7 lines>...
align_to_window,
^^^^^^^^^^^^^^^^
)
^
RuntimeError: cuFFT error: CUFFT_INTERNAL_ERROR

Prompt executed in 3.40 seconds

[rattus128]

My Linux 5090 works on 2.9.1+cu128

WARNING: You need pytorch with cu130 or higher to use optimized CUDA operations.
Found comfy_kitchen backend eager: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'scaled_mm_nvfp4']}
Found comfy_kitchen backend triton: {'available': True, 'disabled': True, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8']}
Found comfy_kitchen backend cuda: {'available': True, 'disabled': True, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'scaled_mm_nvfp4']}
Checkpoint files will always be loaded safely.
Total VRAM 32087 MB, total RAM 96108 MB
pytorch version: 2.9.1+cu128
Set vram state to: NORMAL_VRAM
Device: cuda:0 NVIDIA GeForce RTX 5090 : cudaMallocAsync
Using async weight offloading with 2 streams
Enabled pinned memory 91302.0
working around nvidia conv3d memory bug.
Using pytorch attention
aimdo: src/control.c:65:INFO:comfy-aimdo inited for GPU: NVIDIA GeForce RTX 5090 (VRAM: 32086 MB)
DynamicVRAM support detected and enabled
Python version: 3.12.3 (main, Nov  6 2025, 13:44:16) [GCC 13.3.0]
ComfyUI version: 0.15.1

got prompt
/home/rattus/venv/lib/python3.12/site-packages/torch/functional.py:681: UserWarning: A window was not provided. A rectangular window will be applied,which is known to cause spectral leakage. Other windows such as torch.hann_window or torch.hamming_window are recommended to reduce spectral leakage.To suppress this warning and use a rectangular window, explicitly set `window=torch.ones(n_fft, device=<device>)`. (Triggered internally at /pytorch/aten/src/ATen/native/SpectralOps.cpp:836.)
  return _VF.stft(  # type: ignore[attr-defined]
Resampling input 24000 to 44100
Processing chunks: 100%|██████████| 2/2 [00:00<00:00,  2.35it/s]
Prompt executed in 2.92 seconds
got prompt
Prompt executed in 0.01 seconds

(venv) (BEAST)rattus@rattus-box:~$ pip show torchaudio
Name: torchaudio
Version: 2.9.1
Summary: An audio package for PyTorch
Home-page: https://github.com/pytorch/audio
Author: Soumith Chintala, David Pollack, Sean Naren, Peter Goldsborough, Moto Hira, Caroline Chen, Jeff Hwang, Zhaoheng Ni, Xiaohui Zhang
Author-email: soumith@pytorch.org
License: 
Location: /home/rattus/venv/lib/python3.12/site-packages
Requires: torch
Required-by: 

[solss-bit]

import torchaudio; print(torch.version, torchaudio.version, torch.version.cuda)" 2.9.1+cu128 2.9.1+cu128 12.8.

I was using an image+audio to video for LTX2 workflow. Did an update to test out the new RAM handling. Was working fine prior to the update, and immediately after:

Requested to load AudioVAE
loaded completely; 21744.67 MB usable, 415.20 MB loaded, full load: True
!!! Exception during processing !!! cuFFT error: CUFFT_INTERNAL_ERROR
Traceback (most recent call last):
File "F:\Data\Packages\ComfyUI\execution.py", line 524, in execute
output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "F:\Data\Packages\ComfyUI\execution.py", line 333, in get_output_data
return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "F:\Data\Packages\ComfyUI\custom_nodes\comfyui-lora-manager\py\metadata_collector\metadata_hook.py", line 168, in async_map_node_over_list_with_metadata
results = await original_map_node_over_list(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "F:\Data\Packages\ComfyUI\execution.py", line 307, in async_map_node_over_list
await process_inputs(input_dict, i)
File "F:\Data\Packages\ComfyUI\execution.py", line 295, in process_inputs
result = f(**inputs)
^^^^^^^^^^^
File "F:\Data\Packages\ComfyUI\comfy_api\internal_init.py", line 149, in wrapped_func
return method(locked_class, **inputs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "F:\Data\Packages\ComfyUI\comfy_api\latest_io.py", line 1751, in EXECUTE_NORMALIZED
to_return = cls.execute(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "F:\Data\Packages\ComfyUI\comfy_extras\nodes_lt_audio.py", line 54, in execute
audio_latents = audio_vae.encode(audio)
^^^^^^^^^^^^^^^^^^^^^^^
File "F:\Data\Packages\ComfyUI\comfy\ldm\lightricks\vae\audio_vae.py", line 189, in encode
mel_spec = self.preprocessor.waveform_to_mel(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "F:\Data\Packages\ComfyUI\comfy\ldm\lightricks\vae\audio_vae.py", line 127, in waveform_to_mel
mel = mel_transform(waveform)
^^^^^^^^^^^^^^^^^^^^^^^
File "F:\Data\Packages\ComfyUI\venv\Lib\site-packages\torch\nn\modules\module.py", line 1775, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "F:\Data\Packages\ComfyUI\venv\Lib\site-packages\torch\nn\modules\module.py", line 1786, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "F:\Data\Packages\ComfyUI\venv\Lib\site-packages\torchaudio\transforms_transforms.py", line 620, in forward
specgram = self.spectrogram(waveform)
^^^^^^^^^^^^^^^^^^^^^^^^^^
File "F:\Data\Packages\ComfyUI\venv\Lib\site-packages\torch\nn\modules\module.py", line 1775, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "F:\Data\Packages\ComfyUI\venv\Lib\site-packages\torch\nn\modules\module.py", line 1786, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "F:\Data\Packages\ComfyUI\venv\Lib\site-packages\torchaudio\transforms_transforms.py", line 111, in forward
return F.spectrogram(
^^^^^^^^^^^^^^
File "F:\Data\Packages\ComfyUI\venv\Lib\site-packages\torchaudio\functional\functional.py", line 124, in spectrogram
spec_f = torch.stft(
^^^^^^^^^^^
File "F:\Data\Packages\ComfyUI\venv\Lib\site-packages\torch\functional.py", line 681, in stft
return _VF.stft( # type: ignore[attr-defined]
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: cuFFT error: CUFFT_INTERNAL_ERROR

[rattus128]

Im good on pyt2.12dev+cu13 too

Checkpoint files will always be loaded safely.
Total VRAM 32087 MB, total RAM 96108 MB
pytorch version: 2.12.0.dev20260224+cu130
Set vram state to: NORMAL_VRAM
Device: cuda:0 NVIDIA GeForce RTX 5090 : cudaMallocAsync
Using async weight offloading with 2 streams
Enabled pinned memory 91302.0
Using pytorch attention
aimdo: src/control.c:64:INFO:comfy-aimdo inited for GPU: NVIDIA GeForce RTX 5090 (VRAM: 32086 MB)
DynamicVRAM support detected and enabled
Python version: 3.12.3 (main, Aug 14 2025, 17:47:21) [GCC 13.3.0]
ComfyUI version: 0.15.1

To see the GUI go to: http://0.0.0.0:8188
got prompt
/home/rattus/venv-pyt210-cu13/lib/python3.12/site-packages/torch/functional.py:681: UserWarning: A window was not provided. A rectangular window will be applied,which is known to cause spectral leakage. Other windows such as torch.hann_window or torch.hamming_window are recommended to reduce spectral leakage.To suppress this warning and use a rectangular window, explicitly set `window=torch.ones(n_fft, device=<device>)`. (Triggered internally at /pytorch/aten/src/ATen/native/SpectralOps.cpp:835.)
  return _VF.stft(  # type: ignore[attr-defined]
Resampling input 24000 to 44100
Processing chunks: 100%|████████████████████████████████████████████████████████████| 2/2 [00:00<00:00,  2.53it/s]
Prompt executed in 3.03 seconds

You should report to ComfyUI-MelBandRoFormer. (maintained by @kijai )

[rattus128]

import torchaudio; print(torch.version, torchaudio.version, torch.version.cuda)" 2.9.1+cu128 2.9.1+cu128 12.8.

I was using an image+audio to video for LTX2 workflow. Did an update to test out the new RAM handling. Was working fine prior to the update, and immediately after:

Ok native reproducer. Thx.

Can I get the top of your log? Everything from startup to "listening on ..."

[solss-bit]

I don't have a 'listening on', but going from your previous post, is this what you want to see? If not, I'll post my full startup console.

WARNING: You need pytorch with cu130 or higher to use optimized CUDA operations.
Found comfy_kitchen backend eager: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'scaled_mm_nvfp4']}
Found comfy_kitchen backend triton: {'available': True, 'disabled': True, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8']}
Found comfy_kitchen backend cuda: {'available': True, 'disabled': True, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8']}
Checkpoint files will always be loaded safely.
Total VRAM 24576 MB, total RAM 65277 MB
pytorch version: 2.9.1+cu128
xformers version: 0.0.33+aa7bc366.d20251112
Set vram state to: NORMAL_VRAM
Device: cuda:0 NVIDIA GeForce RTX 3090 : cudaMallocAsync
Using async weight offloading with 2 streams
Enabled pinned memory 29374.0
working around nvidia conv3d memory bug.
Using xformers attention
aimdo: src-win/cuda-detour.c:77:INFO:aimdo_setup_hooks: found driver at 00007FFD8F020000, installing 4 hooks
aimdo: src-win/cuda-detour.c:61:DEBUG:install_hook_entrys: hooks successfully installed
aimdo: src/control.c:66:INFO:comfy-aimdo inited for GPU: NVIDIA GeForce RTX 3090 (VRAM: 24575 MB)
DynamicVRAM support detected and enabled
Python version: 3.11.9 (tags/v3.11.9:de54cf5, Apr 2 2024, 10:12:12) [MSC v.1938 64 bit (AMD64)]
ComfyUI version: 0.15.1
ComfyUI frontend version: 1.39.19

[rattus128]

I don't have a 'listening on', but going from your previous post, is this what you want to see? If not, I'll post my full startup console.

WARNING: You need pytorch with cu130 or higher to use optimized CUDA operations. Found comfy_kitchen backend eager: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'scaled_mm_nvfp4']} Found comfy_kitchen backend triton: {'available': True, 'disabled': True, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8']} Found comfy_kitchen backend cuda: {'available': True, 'disabled': True, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8']} Checkpoint files will always be loaded safely. Total VRAM 24576 MB, total RAM 65277 MB pytorch version: 2.9.1+cu128 xformers version: 0.0.33+aa7bc366.d20251112 Set vram state to: NORMAL_VRAM Device: cuda:0 NVIDIA GeForce RTX 3090 : cudaMallocAsync Using async weight offloading with 2 streams Enabled pinned memory 29374.0 working around nvidia conv3d memory bug. Using xformers attention aimdo: src-win/cuda-detour.c:77:INFO:aimdo_setup_hooks: found driver at 00007FFD8F020000, installing 4 hooks aimdo: src-win/cuda-detour.c:61:DEBUG:install_hook_entrys: hooks successfully installed aimdo: src/control.c:66:INFO:comfy-aimdo inited for GPU: NVIDIA GeForce RTX 3090 (VRAM: 24575 MB) DynamicVRAM support detected and enabled Python version: 3.11.9 (tags/v3.11.9:de54cf5, Apr 2 2024, 10:12:12) [MSC v.1938 64 bit (AMD64)] ComfyUI version: 0.15.1 ComfyUI frontend version: 1.39.19

Perfect thx.

[jdorpitan-code]

Not sure it's on melband roformer side.
I got the same error on LTXVAudioVAEEncode.
The same workflow that worked well last week stopped working.
The logs shows others part of code crash, but yet the same CUFFT_INTERNAL_ERROR :

got prompt
VAE load device: cuda:0, offload device: cpu, dtype: torch.bfloat16
Requested to load AudioVAE
loaded completely; 5851.64 MB usable, 415.20 MB loaded, full load: True
!!! Exception during processing !!! cuFFT error: CUFFT_INTERNAL_ERROR
Traceback (most recent call last):
File "I:\ComfyUI_windows_portable\ComfyUI\execution.py", line 524, in execute
output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "I:\ComfyUI_windows_portable\ComfyUI\execution.py", line 333, in get_output_data
return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "I:\ComfyUI_windows_portable\ComfyUI\execution.py", line 307, in async_map_node_over_list
await process_inputs(input_dict, i)
File "I:\ComfyUI_windows_portable\ComfyUI\execution.py", line 295, in process_inputs
result = f(**inputs)
File "I:\ComfyUI_windows_portable\ComfyUI\comfy_api\internal_init.py", line 149, in wrapped_func
return method(locked_class, **inputs)
File "I:\ComfyUI_windows_portable\ComfyUI\comfy_api\latest_io.py", line 1751, in EXECUTE_NORMALIZED
to_return = cls.execute(*args, **kwargs)
File "I:\ComfyUI_windows_portable\ComfyUI\comfy_extras\nodes_lt_audio.py", line 54, in execute
audio_latents = audio_vae.encode(audio)
File "I:\ComfyUI_windows_portable\ComfyUI\comfy\ldm\lightricks\vae\audio_vae.py", line 189, in encode
mel_spec = self.preprocessor.waveform_to_mel(
waveform, waveform_sample_rate, device=self.device_manager.load_device
)
File "I:\ComfyUI_windows_portable\ComfyUI\comfy\ldm\lightricks\vae\audio_vae.py", line 127, in waveform_to_mel
mel = mel_transform(waveform)
File "I:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1775, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "I:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1786, in _call_impl
return forward_call(*args, **kwargs)
File "I:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torchaudio\transforms_transforms.py", line 620, in forward
specgram = self.spectrogram(waveform)
File "I:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1775, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "I:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1786, in _call_impl
return forward_call(*args, **kwargs)
File "I:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torchaudio\transforms_transforms.py", line 111, in forward
return F.spectrogram(
~~~~~~~~~~~~~^
waveform,
^^^^^^^^^
...<9 lines>...
self.onesided,
^^^^^^^^^^^^^^
)
^
File "I:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torchaudio\functional\functional.py", line 124, in spectrogram
spec_f = torch.stft(
input=waveform,
...<8 lines>...
return_complex=True,
)
File "I:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\functional.py", line 681, in stft
return _VF.stft( # type: ignore[attr-defined]
~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
input,
^^^^^^
...<7 lines>...
align_to_window,
^^^^^^^^^^^^^^^^
)
^
RuntimeError: cuFFT error: CUFFT_INTERNAL_ERROR

Prompt executed in 2.67 seconds

[rattus128]

I have reproduced this (2.10, Windows, 5060, cu130):

Found comfy_kitchen backend cuda: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'scaled_mm_nvfp4']}
Found comfy_kitchen backend eager: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'scaled_mm_nvfp4']}
Found comfy_kitchen backend triton: {'available': False, 'disabled': True, 'unavailable_reason': "ImportError: No module named 'triton'", 'capabilities': []}
Checkpoint files will always be loaded safely.
Total VRAM 8150 MB, total RAM 32691 MB
pytorch version: 2.10.0+cu130
Set vram state to: NORMAL_VRAM
Device: cuda:0 NVIDIA GeForce RTX 5060 : cudaMallocAsync
Using async weight offloading with 2 streams
working around nvidia conv3d memory bug.
packages\torchvision\_C.pyd
packages\torchvision\image.pyd
Using pytorch attention
packages\torchaudio\lib\libtorchaudio.pyd
aimdo: src-win/cuda-detour.c:77:INFO:aimdo_setup_hooks: found driver at 00007FFBAD8C0000, installing 4 hooks
aimdo: src-win/cuda-detour.c:61:DEBUG:install_hook_entrys: hooks successfully installed
aimdo: src/control.c:66:INFO:comfy-aimdo inited for GPU: NVIDIA GeForce RTX 5060 (VRAM: 8150 MB)
DynamicVRAM support detected and enabled
Python version: 3.13.11 (tags/v3.13.11:6278944, Dec  5 2025, 16:26:58) [MSC v.1944 64 bit (AMD64)]
ComfyUI version: 0.15.1

  File "C:\Users\rattus\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\ldm\lightricks\vae\audio_vae.py", line 127, in waveform_to_mel
    mel = mel_transform(waveform)
  File "C:\Users\rattus\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
  File "C:\Users\rattus\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
    return forward_call(*args, **kwargs)
  File "C:\Users\rattus\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torchaudio\transforms\_transforms.py", line 620, in forward
    specgram = self.spectrogram(waveform)
  File "C:\Users\rattus\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
  File "C:\Users\rattus\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
    return forward_call(*args, **kwargs)
  File "C:\Users\rattus\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torchaudio\transforms\_transforms.py", line 111, in forward
    return F.spectrogram(
           ~~~~~~~~~~~~~^
        waveform,
        ^^^^^^^^^
    ...<9 lines>...
        self.onesided,
        ^^^^^^^^^^^^^^
    )
    ^
  File "C:\Users\rattus\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torchaudio\functional\functional.py", line 123, in spectrogram
    spec_f = torch.stft(
        input=waveform,
    ...<8 lines>...
        return_complex=True,
    )
  File "C:\Users\rattus\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\functional.py", line 681, in stft
    return _VF.stft(  # type: ignore[attr-defined]
           ~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
        input,
        ^^^^^^
    ...<7 lines>...
        align_to_window,
        ^^^^^^^^^^^^^^^^
    )
    ^
RuntimeError: cuFFT error: CUFFT_INTERNAL_ERROR

Prompt executed in 0.95 seconds

[rattus128]

Not sure it's on melband roformer side. I got the same error on LTXVAudioVAEEncode. The same workflow that worked well last week stopped working. The logs shows others part of code crash, but yet the same CUFFT_INTERNAL_ERROR :

got prompt VAE load device: cuda:0, offload device: cpu, dtype: torch.bfloat16 Requested to load AudioVAE loaded completely; 5851.64 MB usable, 415.20 MB loaded, full load: True !!! Exception during processing !!! cuFFT error: CUFFT_INTERNAL_ERROR Traceback (most recent call last): File "I:\ComfyUI_windows_portable\ComfyUI\execution.py", line 524, in execute output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "I:\ComfyUI_windows_portable\ComfyUI\execution.py", line 333, in get_output_data return_values = await async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "I:\ComfyUI_windows_portable\ComfyUI\execution.py", line 307, in async_map_node_over_list await process_inputs(input_dict, i) File "I:\ComfyUI_windows_portable\ComfyUI\execution.py", line 295, in process_inputs result = f(**inputs) File "I:\ComfyUI_windows_portable\ComfyUI\comfy_api\internal__init.py", line 149, in wrapped_func return method(locked_class, **inputs) File "I:\ComfyUI_windows_portable\ComfyUI\comfy_api\latest_io.py", line 1751, in EXECUTE_NORMALIZED to_return = cls.execute(*args, **kwargs) File "I:\ComfyUI_windows_portable\ComfyUI\comfy_extras\nodes_lt_audio.py", line 54, in execute audio_latents = audio_vae.encode(audio) File "I:\ComfyUI_windows_portable\ComfyUI\comfy\ldm\lightricks\vae\audio_vae.py", line 189, in encode mel_spec = self.preprocessor.waveform_to_mel( waveform, waveform_sample_rate, device=self.device_manager.load_device ) File "I:\ComfyUI_windows_portable\ComfyUI\comfy\ldm\lightricks\vae\audio_vae.py", line 127, in waveform_to_mel mel = mel_transform(waveform) File "I:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1775, in _wrapped_call_impl return self._call_impl(*args, **kwargs) ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "I:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1786, in _call_impl return forward_call(*args, **kwargs) File "I:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torchaudio\transforms_transforms.py", line 620, in forward specgram = self.spectrogram(waveform) File "I:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1775, in _wrapped_call_impl return self._call_impl(*args, **kwargs) ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "I:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1786, in _call_impl return forward_call(*args, **kwargs) File "I:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torchaudio\transforms_transforms.py", line 111, in forward return F.spectrogram( ~~~~~~~~~~~~~^ waveform, ^^^^^^^^^ ...<9 lines>... self.onesided, ^^^^^^^^^^^^^^ ) ^ File "I:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torchaudio\functional\functional.py", line 124, in spectrogram spec_f = torch.stft( input=waveform, ...<8 lines>... return_complex=True, ) File "I:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\functional.py", line 681, in stft return _VF.stft( # type: ignore[attr-defined] ~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ input, ^^^^^^ ...<7 lines>... align_to_window, ^^^^^^^^^^^^^^^^ ) ^ RuntimeError: cuFFT error: CUFFT_INTERNAL_ERROR

Prompt executed in 2.67 seconds

my results are inconsistent with you as I can only reproduce it with dynamic-vram mode, wheras you appear to not have that yet (its pre-stable).

Can I get the startup log?