TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.

[junwoo7626]

Hello,

I encountered a TypeError while using the Video Combine node (VHS_VideoCombine). It seems like the input tensor is remaining on the GPU (cuda:0) when the code attempts to convert it to numpy, causing the crash.

I am currently testing a workflow involving the LTXV model, and the error occurs at the video generation step.

Thank you for your great work!

[PGCRT]

/videohelpersuite/nodes.py

Line 603. Find this: python.numpy().tobytes()

Change it to: python.cpu().numpy().tobytes()

Maybe our numpy version is not supported yet? Idk, but this should fix the issue

[PATATAJEC]

I have the same problem. this line was just .numpy().tobytes() without python. Changin it to python.cpu().numpy().tobytes() failed loading the node to comfy, and changin it to .cpu().numpy().tobytes() gives this error:

Traceback (most recent call last): File "F:\ComfyUI_2D\ComfyUI\execution.py", line 518, 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:\ComfyUI_2D\ComfyUI\execution.py", line 329, 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:\ComfyUI_2D\ComfyUI\execution.py", line 303, in _async_map_node_over_list await process_inputs(input_dict, i) File "F:\ComfyUI_2D\ComfyUI\execution.py", line 291, in process_inputs result = f(**inputs) File "F:\ComfyUI_2D\ComfyUI\custom_nodes\ComfyUI-VideoHelperSuite\videohelpersuite\nodes.py", line 609, in combine_video raise Exception("An error occured in the ffmpeg subprocess:\n" \ + e.stderr.decode(*ENCODE_ARGS)) Exception: An error occured in the ffmpeg subprocess: [pcm_f32le @ 0000027c68c5e740] Too many channels: 230640 [pcm_f32le @ 0000027c68c5e740] Decoder requires channel layout to be set Last message repeated 1 times [pcm_f32le @ 0000027c68c3c8c0] Decoder requires channel layout to be set [aist#1:0/pcm_f32le @ 0000027c68c5eb00] [dec:pcm_f32le @ 0000027c68c0b540] Error while opening decoder: Invalid argument Error opening output file F:\ComfyUI_2D\ComfyUI\output\2026-01-20\LTX2\LTX2-KJ-I2V_KeyFrame__00006-audio.mp4. Error opening output files: Invalid argument

[PATATAJEC]

hmmm... I think I've found the culprit. I was using this standard node for audio decode instead of LTX one:

After changin it it works with this line: .cpu().numpy().tobytes() ... i'll try to change it to the original one.

[PATATAJEC]

yes, that was the issue. now after changing to original line of code and changin to ltxv audio vae decode from common vae decode it works. @PGCRT check this out and close the issue, if it's working for you.