[Errno 22] Invalid argument: 'avcodec_send_frame()'; last error log: [aac] Input contains (near) NaN/+-Inf

[lebakerino]

Hi, I keep having this issue with LTX 2.3 right at the end of the video at 'save video' in comfyui.

I have tried about 4 workflows including my own that I use for LTX2.0 (that works fine) and only one workflow worked (Kijais) I've tried replicating the different parts and components he uses in one of the other workflows but no luck, the same error or a very similar error code appears. Kijai uses simply videcombine at the end instead of the 'save video' but even replacing save video in my(and the other non-working) workflow gives the error

My comfyui is up to date, my nodes are up to date.

I have switched between both FP4 and GGUF LTX models, I've tried GGUF and FP8 gemma models, I've tried using the LTX2 Audio decode.

I've tried it all.

Need help as others don't seem to be having this issue much.

[raffetazarius]

https://huggingface.co/RuneXX has workflows that work well for LTX-2.0 and 2.3.
Lightricks official workflows also work well: https://github.com/Lightricks/ComfyUI-LTXVideo/tree/master/example_workflows
These might be a good starting point, since there are plenty of folks using them, so we can help each other better.

If you're expecting someone to give you more direct support on your specific issue, you'd need to include more information, we're flying blind with the info you provided.

At a minimum, you could provide:

1. Your workflow JSON file
2. Details of the specific node that's throwing the error
3. Logs from ComfyUI

I'd also recommend trying to use AI to diagnose ComfyUI errors. Gemini can read a JSON file and spot problematic configurations.... and it's even smarter if you tell it your exact computer configuration. I'd expect ChatGPT or Claude can too.

Your issue may be nothing to do with LTX, and thus not related to the https://github.com/Lightricks/ComfyUI-LTXVideo/ node pack you've posted the issue in. The 'Save Video' node isn't part of this Node Pack... there are a few different ones, the Video Helper Suite (a completely different GitHub repo) and the built-in one, provided from ComfyUI themselves.

Hope this helps! And good luck out there in the wild west of AI video gen! 🤠

[lebakerino]

https://huggingface.co/RuneXX has workflows that work well for LTX-2.0 and 2.3. Lightricks official workflows also work well: https://github.com/Lightricks/ComfyUI-LTXVideo/tree/master/example_workflows These might be a good starting point, since there are plenty of folks using them, so we can help each other better.

If you're expecting someone to give you more direct support on your specific issue, you'd need to include more information, we're flying blind with the info you provided.

At a minimum, you could provide:

1. Your workflow JSON file
2. Details of the specific node that's throwing the error
3. Logs from ComfyUI

I'd also recommend trying to use AI to diagnose ComfyUI errors. Gemini can read a JSON file and spot problematic configurations.... and it's even smarter if you tell it your exact computer configuration. I'd expect ChatGPT or Claude can too.

Your issue may be nothing to do with LTX, and thus not related to the https://github.com/Lightricks/ComfyUI-LTXVideo/ node pack you've posted the issue in. The 'Save Video' node isn't part of this Node Pack... there are a few different ones, the Video Helper Suite (a completely different GitHub repo) and the built-in one, provided from ComfyUI themselves.

Hope this helps! And good luck out there in the wild west of AI video gen! 🤠

I was tired and sending a general question regarding this error as I have seen people have it during searching and solution was shared.

To go into more detail, I've been discussing and trying to solve this all weekend.

The official workflows don't work, Runes doesn't work, my own workflow that works fine with LTX2 ,modified for 2.3 with its relevant models doesn't work. Only Kijais has completed a video generation without this error or another similar error related to the same 'aac' thing.

That's why I'm being vague as I believe the solution isn't some little thing.

There is no 'node' other than 'Save video', that is where the error is, the error is what I have shared.

And yes I've spent the weekend several times trying to use Gemini, Chatgpt and claude by even going as far as uploading my workflow, kijais and runes and asking it to explain why Kijai's works and it one way or another hasn't given a solution other than something to do with frame count (of which I then mimicked Kijai's nodes and routing in my own workflow and yet it still doesn't work even though I've copied Kijai's workflow)

LTX 2 works fine, WAN works fine so I'm not sure Comfy is the issue, something has changed between 2.0 and 2.3 LTX so I'd gather it has something to do with LTX models or how LTX is expecting to receive data.

Here is a more elaborate error log from the end of the video generation comfyui, do recall the video exports fine with kijai's workflow though it seems to output 2 streams, whenever this error happens though in my/the other workflows, it outputs just a black video. I've updated ffmpeg and made sure it is set in enviroment variables to the correct path etc, no difference made.

"Model VideoVAE prepared for dynamic VRAM loading. 1384MB Staged. 0 patches attached.
D:\SM\Data\Packages\Comf\custom_nodes\comfyui-videohelpersuite\videohelpersuite\nodes.py:130: RuntimeWarning: invalid value encountered in cast
return tensor_to_int(tensor, 8).astype(np.uint8)

!!! Exception during processing !!! An error occured in the ffmpeg subprocess:

Traceback (most recent call last):
File "D:\SM\Data\Packages\Comf\custom_nodes\comfyui-videohelpersuite\videohelpersuite\nodes.py", line 597, in combine_video
res = subprocess.run(mux_args, input=audio_data,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\SM\Data\Assets\Python\cpython-3.12.12-windows-x86_64-none\Lib\subprocess.py", line 571, in run
raise CalledProcessError(retcode, process.args,
subprocess.CalledProcessError: Command '['.\ffmpeg.EXE', '-v', 'error', '-n', '-i', 'g:\comfyui\output\AnimateDiff_00016.mp4', '-ar', '48000', '-ac', '2', '-f', 'f32le', '-i', '-', '-c:v', 'copy', '-c:a', 'aac', '-af', 'apad=whole_dur=3.375', '-shortest', 'g:\comfyui\output\AnimateDiff_00016-audio.mp4']' returned non-zero exit status 4294967274.

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "D:\SM\Data\Packages\Comf\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:\SM\Data\Packages\Comf\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:\SM\Data\Packages\Comf\execution.py", line 307, in _async_map_node_over_list
await process_inputs(input_dict, i)
File "D:\SM\Data\Packages\Comf\execution.py", line 295, in process_inputs
result = f(**inputs)
^^^^^^^^^^^
File "D:\SM\Data\Packages\Comf\custom_nodes\comfyui-videohelpersuite\videohelpersuite\nodes.py", line 600, in combine_video
raise Exception("An error occured in the ffmpeg subprocess:\n"
Exception: An error occured in the ffmpeg subprocess: "

I've added my workflow.

LTX-2_01064_.json

[raffetazarius]

https://g.co/gemini/share/c61bd4a13e4a

[lebakerino]

https://g.co/gemini/share/c61bd4a13e4a

Like I said I already did that with Gemini, Claude, ChatGPT multiple times, it didn't have the answer as to what was causing it.

I did eventually figure it out after breaking down one of the workflows that wasn't working to match kijai's that did and it turns out it wasn't the workflow that was special. In Kijai's workflow I had accidently left the 19b model distilled LORA in the lora loader meanwhile in the nonworking workflows I had replaced that with the distilled lora for 22b BUT it was FP8 format (I had also converted and used fp8 lora with 19b) but for some reason FP8 distilled LORA wont work with 22b and that was causing the error, once removed it all works fine.

So yeah a mistake on my part for having different LORA's but also something fundamentally has changed with the Distilled FP8 lora functionality in 22b. Even when I went back to the page where I downloaded the fp8 lora the guy had deleted it and then I downloaded Kijai's and converted it myself and it didn't work either.

Anyway case closed, was going to delete this but I think it will potentially prove valuable for someone else... be wary of them FP8 Loras!

Cheers.

[digitfold]

https://g.co/gemini/share/c61bd4a13e4a

Like I said I already did that with Gemini, Claude, ChatGPT multiple times, it didn't have the answer as to what was causing it.

I did eventually figure it out after breaking down one of the workflows that wasn't working to match kijai's that did and it turns out it wasn't the workflow that was special. In Kijai's workflow I had accidently left the 19b model distilled LORA in the lora loader meanwhile in the nonworking workflows I had replaced that with the distilled lora for 22b BUT it was FP8 format (I had also converted and used fp8 lora with 19b) but for some reason FP8 distilled LORA wont work with 22b and that was causing the error, once removed it all works fine.

So yeah a mistake on my part for having different LORA's but also something fundamentally has changed with the Distilled FP8 lora functionality in 22b. Even when I went back to the page where I downloaded the fp8 lora the guy had deleted it and then I downloaded Kijai's and converted it myself and it didn't work either.

Anyway case closed, was going to delete this but I think it will potentially prove valuable for someone else... be wary of them FP8 Loras!

Cheers.

I actually tried removing the LoRA node entirely and running only the base model, but the problem still appears on some machines. On my test machine the same workflow runs fine, but a few users still get the NaN/Inf AAC error. So it seems like it might be something related to the system environment rather than only the LoRA.

[pappagheno]

I had the same issues, hte problem is the sigma values. check your sigma node if hte values are random or go from a little value and then a high value this is goig nto give you ssies specially if you ahve two passes of generations

[Rusoloco73]

ltx and comfyui ..what a piece of shit combo