[Bug] Crash on startup due to ImportError from transformers library

[dpapperini]

Hi, the custom node fails to load and causes an error on ComfyUI startup, likely due to an incompatibility with a recent version of the transformers library.

Full Traceback

[ComfyUI] Traceback (most recent call last):
[ComfyUI] File "/src/ComfyUI/nodes.py", line 2124, in load_custom_node
[ComfyUI] module_spec.loader.exec_module(module)
[ComfyUI] File "<frozen importlib._bootstrap_external>", line 999, in exec_module
[ComfyUI] File "<frozen importlib._bootstrap>", line 488, in _call_with_frames_removed
[ComfyUI] File "/src/ComfyUI/custom_nodes/ComfyUI-HunyuanVideoWrapper/__init__.py", line 1, in <module>
[ComfyUI] from .nodes import NODE_CLASS_MAPPINGS as NODES_CLASS, NODE_DISPLAY_NAME_MAPPINGS as NODES_DISPLAY
[ComfyUI] File "/src/ComfyUI/custom_nodes/ComfyUI-HunyuanVideoWrapper/nodes.py", line 11, in <module>
[ComfyUI] from .hyvideo.text_encoder import TextEncoder
[ComfyUI] File "/src/ComfyUI/custom_nodes/ComfyUI-HunyuanVideoWrapper/hyvideo/text_encoder/__init__.py", line 9, in <module>
[ComfyUI] from .processing_llava import LlavaProcessor
[ComfyUI] File "/src/ComfyUI/custom_nodes/ComfyUI-HunyuanVideoWrapper/hyvideo/text_encoder/processing_llava.py", line 23, in <module>
[ComfyUI] from transformers.processing_utils import ProcessingKwargs, ProcessorMixin, Unpack, _validate_images_text_input_order
[ComfyUI] ImportError: cannot import name '_validate_images_text_input_order' from 'transformers.processing_utils' (/root/.pyenv/versions/3.12.11/lib/python3.12/site-packages/transformers/processing_utils.py)
[ComfyUI] Cannot import /src/ComfyUI/custom_nodes/ComfyUI-HunyuanVideoWrapper module for custom nodes: cannot import name '_validate_images_text_input_order' from 'transformers.processing_utils' (/root/.pyenv/versions/3.12.11/lib/python3.12/site-packages/transformers/processing_utils.py)

Environment

Please provide the following information to help diagnose the issue:

• OS: Ubuntu 22.04
• Python Version: 3.12.11
• Transformers Version: 4.54.1
• PyTorch Version: 2.5.1+cu121
• xformers Version: 0.0.29.post1
• CUDA Version: 12.1
• Device Version: NVIDIA GeForce RTX 3070
• ComfyUI Version: 0.3.46 Last Commit 5d4cc3ba1b412b9acacd37fd23d59e0e1654f83c
• ComfyUI-HunyuanVideoWrapper Version: Last Commit 3ce9640497139b14462910b0dbf2be1df855c1d6

Thanks!

[mertkemal1131]

having the same issue

[Haiyuan-Zhang]

having the same issue

same issue, downgrade transformers==4.53.3 , solved

[mertkemal1131]

in where because still having the issue

[Spnaba]

Have anyone solved this? upgraded transformers to 4.53.3 and still having the same issue

[dpapperini]

Hi @Spnaba,

Have anyone solved this? upgraded transformers to 4.53.3 and still having the same issue

@Haiyuan-Zhang verified that downgrading to transformers==4.53.3 fixes the problem. If it didn't work for you, make sure you've completely uninstalled the previous version before installing the correct one. Sometimes pip install doesn't properly update the library unless you first uninstall the existing version.

You can try running this command:

pip uninstall transformers && pip install transformers==4.53.3

If the issue persists, it would be helpful to get more information on environment and a full trace log of the error.

Let us know if this solves it for you!

[dpapperini]

Hi @mertkemal1131,

having the same issue

in where because still having the issue

To fix this, you have a couple of options.

Option 1: Use the Command Line

You can run this command in your ComfyUI environment to downgrade the transformers library to a compatible version:

pip uninstall transformers && pip install transformers==4.53.3

---

Option 2: Edit the requirements.txt File

Alternatively, you can edit the requirements.txt file, which is located in the root folder of the ComfyUI-HunyuanVideoWrapper project.

Open the file and find the line that looks like this:

transformers>=4.49.0

Change it to:

transformers==4.53.3

This tells the system to use a version of transformers that is equal to 4.53.3.

After you save the file, run pip install -r requirements.txt in your project folder to force the installation.

Let us know if this solves it for you!

[Carrott2016]

Hi,
I installed https://github.com/Enemyx-net/VibeVoice-ComfyUI/tree/main, which requires transformers==4.51.3.

Got the same error, the installer updated from transformers=4.49.0 to 4.51.3 even though, Hunyuan Video Wrapper requirement is transformers >= 4.49.0 and the insatlled transformers==4.51.3 is lower version than transformers==4.53.3m that is recommended above. I fixed the Wrapper issue by uninstalling VibeVoice and downgrading to transformers==4.49.0 again.

So, how to prevent conflicts between different node requirements? I think we can only have one Transformer version installed in ComfyUI_portable\python_embeded\Lib\site-packages.

[dpapperini]

Hi @Carrott2016,

Hi, I installed https://github.com/Enemyx-net/VibeVoice-ComfyUI/tree/main, which requires transformers==4.51.3.

Got the same error, the installer updated from transformers=4.49.0 to 4.51.3 even though, Hunyuan Video Wrapper requirement is transformers >= 4.49.0 and the insatlled transformers==4.51.3 is lower version than transformers==4.53.3m that is recommended above. I fixed the Wrapper issue by uninstalling VibeVoice and downgrading to transformers==4.49.0 again.

So, how to prevent conflicts between different node requirements? I think we can only have one Transformer version installed in ComfyUI_portable\python_embeded\Lib\site-packages.

This is a common issue when managing different libraries in a single Python environment, especially in a context like ComfyUI where each node can have specific requirements.

Tips to Prevent Library Conflicts

1. Virtual Environments: The most effective way to prevent these conflicts is to use a virtual environment for each project or for each set of nodes that have conflicting requirements. This way, each environment will have its own installed dependencies without interfering with others.
2. Specify Versions: Maintain a dedicated requirements.txt for each environment so you can easily reinstall exact dependencies as needed.
3. Manual Management: If conflicts persist, the solution is manual version management, as you've already done. Uninstall the problematic version (pip uninstall transformers) and install the version required by the other nodes (eg: pip install transformers==4.51.3).

The best solution is to isolate dependencies using virtual environments.

This is a great idea. Using different virtual environments is the most reliable way to avoid conflicts.
Here are the commands you can use to set up and manage them.

Managing Library Conflicts with Virtual Environments

A virtual environment creates an isolated space for your Python project. This means you can install a specific version of a library like transformers for one set of nodes without affecting another set.

Here's how to create and use one:

1. Create a New Virtual Environment:
   Open your terminal or command prompt, navigate to a directory where you want to store your environments (e.g., inside your ComfyUI folder), and run this command. This will create a new folder named venv with a fresh, isolated Python installation.

   python -m venv venv

2. Activate the Virtual Environment:
   After creating it, you need to activate it. The prompt in your terminal will change, indicating you are now working inside this isolated environment.

   • On Windows:

     venv\Scripts\activate

   • On macOS/Linux:

     source venv/bin/activate

3. Install Required Libraries:
   Now that the environment is active, you can install the specific library versions needed for your nodes (e.g., the VibeVoice node) without affecting anything else.

   pip install transformers==4.51.3
   pip install -r requirements.txt

4. Deactivate the Environment:
   When you're finished or want to switch back to your main environment, simply run this command.

   deactivate

By following these steps, you can create separate environments for different nodes that have conflicting transformers requirements. Just create a new venv for each set of nodes and activate the correct one before running ComfyUI. This ensures that each node's dependencies are completely isolated.

If this is not possible, manual version management remains the only alternative to resolve conflicts.

[elkay]

Same issue, 4.53.3 fixes the issue for me, too, and so far doesn't break anything else. We're on borrowed time until it does, though, so hopefully this can be fixed soon.