Random crashes using Latent2RGB preview method

[LukeG89]

Expected Behavior

Generating without crashes using Latent2RGB preview.

Actual Behavior

After some generations, ComfyUI crashes when is about to preview the latest image in the Preview Image node.
It can happen after only 5-6 images or up to 20 or more, pretty random behavior.
The crash also makes reboot my external SSD drive, where my ComfyUI portable is installed.

After some testing, I tried switching to TAESD method and the issue was gone, so I guess Latent2RGB is the problem.

Steps to Reproduce

Select Latent2RGB as preview method and start generating.

Debug Logs

Exception in thread Thread-9 (prompt_worker):
Traceback (most recent call last):
  File "E:\python_embeded\Lib\site-packages\PIL\ImageFile.py", line 547, in _save
AttributeError: '_idat' object has no attribute 'fileno'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "E:\ComfyUI\execution.py", line 323, in execute
  File "E:\ComfyUI\execution.py", line 198, in get_output_data
  File "E:\ComfyUI\execution.py", line 169, in _map_node_over_list
  File "E:\ComfyUI\execution.py", line 158, in process_inputs
  File "E:\ComfyUI\nodes.py", line 1511, in save_images
  File "E:\python_embeded\Lib\site-packages\PIL\Image.py", line 2568, in save
  File "E:\python_embeded\Lib\site-packages\PIL\PngImagePlugin.py", line 1431, in _save
  File "E:\python_embeded\Lib\site-packages\PIL\ImageFile.py", line 551, in _save
  File "E:\python_embeded\Lib\site-packages\PIL\ImageFile.py", line 571, in _encode_tile
  File "E:\python_embeded\Lib\site-packages\PIL\PngImagePlugin.py", line 1101, in write
  File "E:\python_embeded\Lib\site-packages\PIL\PngImagePlugin.py", line 1088, in putchunk
OSError: [Errno 22] Invalid argument

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "threading.py", line 1045, in _bootstrap_inner
  File "threading.py", line 982, in run
  File "E:\ComfyUI\main.py", line 125, in prompt_worker
    e.execute(item[2], prompt_id, item[3], item[4])
  File "E:\ComfyUI\custom_nodes\rgthree-comfy\__init__.py", line 217, in rgthree_execute
  File "E:\ComfyUI\execution.py", line 500, in execute
  File "E:\ComfyUI\execution.py", line 390, in execute
  File "E:\ComfyUI\execution.py", line 390, in <listcomp>
  File "E:\ComfyUI\execution.py", line 242, in format_value
  File "E:\python_embeded\Lib\site-packages\torch\_tensor.py", line 523, in __repr__
  File "E:\python_embeded\Lib\site-packages\torch\_tensor_str.py", line 706, in _str
  File "contextlib.py", line 137, in __enter__
  File "E:\python_embeded\Lib\site-packages\torch\utils\_python_dispatch.py", line 235, in _disable_current_modes
  File "<frozen importlib._bootstrap>", line 1176, in _find_and_load
  File "<frozen importlib._bootstrap>", line 1147, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 690, in _load_unlocked
  File "<frozen importlib._bootstrap_external>", line 936, in exec_module
  File "<frozen importlib._bootstrap_external>", line 1073, in get_code
  File "<frozen importlib._bootstrap_external>", line 1131, in get_data
OSError: [Errno 22] Invalid argument

Other

No response

[ltdrdata]

For devevloper:
One of the cases that causes this error is when the image size is incorrect, such as (0, 100).
This is an aspect worth checking when investigating.
python-pillow/Pillow#3208

[ltdrdata]

The current log is incomplete.
Please disable rgthree, retry to reproduce the situation, and upload the complete log.

[LukeG89]

Tested all this morning, tried with both rgthree enabled and disabled, gradually removed rgthree nodes and bookmarks, but the problem still persist.

I didn't get any issue generating with the default workflow, so maybe there is another custom node in my workflow that conflicts.

Here's the workflow I use:
custom_workflow.json

In my tests I start queueing 30 batches.

Here's the latest debug logs (with rgthree disabled):
debug_logs.txt

[ltdrdata]

What is your pillow version?
If pillow is not version 10.4.0, try upgrading it.

And the issue is still persist.

ComfyUI/nodes.py

Line 1515 in 6ad0ddb

img.save(os.path.join(full_output_folder, file), pnginfo=metadata, compress_level=self.compress_level)

Please modify line 1515 of ComfyUI/nodes.py as follows and record the log again.

            ### edit like this -->
            try:
                img.save(os.path.join(full_output_folder, file), pnginfo=metadata, compress_level=self.compress_level)
            except Exception as e:
                print(f"[DEBUG] img: {img.size}")
                print(f"[DEBUG] path: {os.path.join(full_output_folder, file)}")
                print(f"[DEBUG] metadata: {metadata}")
                print(f"[DEBUG] compress_level: {self.compress_level}")
            ### <--

[LukeG89]

What is your pillow version?
If pillow is not version 10.4.0, try upgrading it.

Yes, Pillow is 10.4.0.

Please modify line 1515 of ComfyUI/nodes.py as follows and record the log again.

I don't know if I did it right because I edited nodes.py using notepad.
Here's the new debug logs:
Debug_logs_nodes-py_mod.txt

[ltdrdata]

How about this?

            ### edit like this -->
            try:
                img.save(os.path.join(full_output_folder, file), pnginfo=metadata, compress_level=self.compress_level)
            except Exception as e:
                print(f"[DEBUG] RETRY")
                img.save(os.path.join(full_output_folder, file), pnginfo=metadata, compress_level=self.compress_level)
                print(f"[DEBUG] RETRY - SUCCESS")
            ### <--

[LukeG89]

How about this?

        ### edit like this -->
        try:
            img.save(os.path.join(full_output_folder, file), pnginfo=metadata, compress_level=self.compress_level)
        except Exception as e:
            print(f"[DEBUG] RETRY")
            img.save(os.path.join(full_output_folder, file), pnginfo=metadata, compress_level=self.compress_level)
            print(f"[DEBUG] RETRY - SUCCESS")
        ### <--

Here's the result:
Debug_logs_nodes-py_mod-2.txt

[LukeG89]

I think I copied the code wrong in the previous test, I fixed it and tested it again.
However, the error is still the same:
Debug_logs_nodes-py_mod-2-2.txt

[LukeG89]

@ltdrdata I found the cause! It's the Manager!

• First, I simplified my workflow to only have ComfyUI nodes inside: Still crashing. 😞
• Then, I disabled one by one my custom nodes, remaining only with the Manager: Still crashing. 😞
• And finally, I disabled the Manager and used --preview-method latent2rgb argument: No crashing! 🎉

Here's the debug logs with only the Manager enabled:
Debug_logs_only_manager_enabled.txt

And this is the simplified workflow that I used:
custom_workflow_simplified.json

[LukeG89]

I think my crashing problem has been resolved somehow, so I am closing the issue for now.

Thanks @ltdrdata for your assistance! ❤️