Fix TypeError: forward_orig() got an unexpected keyword argument 'attn_mask' in ComfyUI

TypeError: forward_orig() got an unexpected keyword argument 'attn_mask'

How to Fix It

Solution 1: Downgrade ComfyUI to v0.3.4 (Community-Reported Fix)

This is the most straightforward fix, reported by the community member dileepajay in the GitHub issue. It involves reverting ComfyUI to version v0.3.4, which still supports the attn_mask argument.

Steps:

1. Navigate to your ComfyUI workspace directory. This is typically the folder where ComfyUI is installed. If you're using a manager like Stability Matrix, the path might be something like F:\StabilityMatrix\Data\Packages\ComfyUI on Windows or /workspace/ on Linux.

   cd /workspace/
   

2. Download the v0.3.4 release archive:

   wget https://github.com/comfyanonymous/ComfyUI/archive/refs/tags/v0.3.4.zip
   

   On Windows, you can download the ZIP file manually from the GitHub releases page and extract it to a temporary location.

3. Unzip the downloaded archive:

   unzip v0.3.4.zip
   

   This creates a folder named ComfyUI-0.3.4.

4. Replace your current ComfyUI files with the older version. The --force flag ensures existing files are overwritten without prompting:

   cp -r ComfyUI-0.3.4/* ComfyUI/ --force
   

   On Windows, you can use File Explorer to copy the contents of the ComfyUI-0.3.4 folder into your existing ComfyUI folder, overwriting all files when prompted.

5. Restart your ComfyUI server. If you're running it as a systemd service on Linux:

   systemctl restart comfyui
   

   If you're running it manually, stop the process (Ctrl+C in the terminal) and start it again with your usual command, for example:

   python main.py --preview-method auto
   

What to expect: After downgrading, the forward_orig() method in the Flux model should accept the attn_mask argument again, and your workflow should run without the error.

Important caveat: This is a temporary fix. The community member notes: "This is just a temporary fix to keep things working while the team sorts out the issue in the next update." You may miss out on new features or bug fixes introduced in v0.3.5, v0.3.6, and v0.3.7.

Solution 2: Update Custom Nodes (If Available)

Before downgrading, check if the custom nodes involved in the error have been updated to be compatible with ComfyUI v0.3.6/v0.3.7. The error involves ComfyUI_smZNodes, which may have a newer version that no longer passes attn_mask.

Steps:

1. Open ComfyUI Manager (if installed) by clicking the "Manager" button in the ComfyUI interface.
2. Go to the "Install Custom Nodes" tab.
3. Search for "smZNodes" and check if an update is available.
4. If an update exists, click "Install" or "Update" to get the latest version.
5. Restart ComfyUI and test your workflow.

If you don't have ComfyUI Manager, you can manually update the custom node by navigating to its folder (e.g., custom_nodes/ComfyUI_smZNodes) and running:

git pull

Or, if it was installed from a ZIP, download the latest release from the node's GitHub repository and replace the folder.

What to expect: If the custom node developer has released a fix, the updated node will not pass attn_mask to the Flux model, and the error will be resolved without downgrading ComfyUI.

Solution 3: Modify the Workflow to Remove attention_mask

If you cannot downgrade or update custom nodes, you can modify the workflow JSON to stop passing the attention_mask that causes the error. This requires editing the workflow file manually.

Steps:

1. Open the workflow JSON file (e.g., "FLUX Redux + Pulid Image to Image Face Swap.json") in a text editor.

2. Search for "attention_mask" in the file. It may appear in the inputs of a node or in the model options.

3. Remove or comment out the lines that set attention_mask. For example, if you find something like:

   "attention_mask": {
     "link": 123,
     "type": "MASK"
   }
   

   Delete that entire block or set it to null.

4. Save the file and reload it in ComfyUI.

5. Run the workflow again.

What to expect: Removing the attention_mask may affect the behavior of PuLID or other face-swapping features, but it should prevent the TypeError from occurring. The generated images may have slightly different quality or face preservation characteristics.

Solution 4: Use a Different Sampler Node

The error occurs specifically with the SamplerCustomAdvanced node (Node ID 13). Try replacing it with a different sampler node, such as:

• KSampler (the standard sampler)
• SamplerCustom (if available)
• Efficient Sampler (from Efficiency Nodes)

Steps:

1. In your workflow, right-click on the SamplerCustomAdvanced node and select "Replace Node."
2. Search for "KSampler" and select it.
3. Reconnect the inputs (model, positive/negative conditioning, latent image, etc.) to the new sampler node.
4. Adjust the sampler settings (steps, CFG, sampler name, scheduler) to match your desired output.
5. Run the workflow.

What to expect: Different sampler nodes may use different internal code paths that do not pass attn_mask to the Flux model. This can bypass the error entirely, though the sampling behavior may differ slightly.

Solution 5: Reinstall ComfyUI from Scratch

If none of the above solutions work, a clean installation of ComfyUI may resolve underlying file corruption or partial update issues.

Steps:

1. Back up your custom nodes, models, and workflows. Copy the custom_nodes, models, and output folders to a safe location.

2. Delete the entire ComfyUI folder.

3. Clone the latest stable version from the official repository:

   git clone https://github.com/comfyanonymous/ComfyUI.git
   

4. Install dependencies:

   cd ComfyUI
   pip install -r requirements.txt
   

5. Restore your custom nodes, models, and workflows from the backup.

6. Start ComfyUI and test the workflow.

What to expect: A clean installation ensures you have a fresh, uncorrupted version of ComfyUI. If the error persists after this, it is almost certainly a compatibility issue with the custom nodes or workflow itself.

If Nothing Works

If you have tried all the above solutions and the error still occurs, consider the following escalation paths:

Report the Issue on GitHub

The original issue is tracked at https://github.com/Comfy-Org/ComfyUI/issues/6077. You can add a comment with your specific workflow, custom node versions, and ComfyUI version. Include the full error report and system information, as shown in the source material. The ComfyUI development team monitors these issues and may provide an official fix in a future release.

Check for Updates Regularly

The community member who provided the downgrade fix advises: "Keep an eye out for new releases!" The ComfyUI team may release v0.3.8 or later that resolves the attn_mask incompatibility. You can watch the GitHub repository or join the ComfyUI Discord server for announcements.

Use an Alternative Face Swap Workflow

If the error is specific to the "FLUX Redux + PuLID Face Swap" workflow, consider using a different face swap approach that does not rely on PuLID with Flux. Alternatives include:

• ReActor Node: A popular face swap node that works with SD1.5 and SDXL models.
• IP-Adapter Face ID: Uses IP-Adapter with face ID embeddings.
• InstantID: Another face ID preservation method that may be compatible with newer ComfyUI versions.

Disable Problematic Custom Nodes

Temporarily disable the ComfyUI_smZNodes custom node by renaming its folder (e.g., add _disabled to the folder name) and restarting ComfyUI. If the error disappears, you have identified the culprit. You can then either wait for an update or use an alternative node that provides similar functionality.

How to Prevent It

Pin ComfyUI to a Stable Version

If you rely on workflows that use custom nodes with specific API expectations, consider pinning your ComfyUI installation to a version that is known to work. For example, if v0.3.4 works for your setup, avoid updating to v0.3.6 or v0.3.7 until you have confirmed compatibility with your custom nodes.

Test Updates in a Separate Environment

Before updating ComfyUI in your production environment, create a separate test installation. Copy your custom nodes and workflows to the test environment, apply the update, and run your critical workflows. This allows you to catch errors like this without disrupting your main setup.

Keep Custom Nodes Updated

Regularly check for updates to your custom nodes, especially after a ComfyUI update. Node developers often release compatibility patches shortly after a new ComfyUI version. Using ComfyUI Manager can automate this process.

Review Workflow JSON for Deprecated Parameters

If you create or download workflows that use Flux models, inspect the JSON file for any references to attention_mask or attn_mask. These parameters may be deprecated in newer model versions. Removing them proactively can prevent errors.

Use Official Nodes When Possible

Whenever feasible, use the built-in nodes provided by ComfyUI rather than custom nodes. Official nodes are more likely to be maintained and compatible with the latest ComfyUI version. For example, use the standard KSampler instead of SamplerCustomAdvanced if you don't need its advanced features.

Summary

The TypeError: forward_orig() got an unexpected keyword argument 'attn_mask' error in ComfyUI is caused by a version mismatch between ComfyUI v0.3.6/v0.3.7 and custom nodes or workflows that pass an attn_mask argument to the Flux model's forward_orig() method. The most reliable fix is to downgrade to ComfyUI v0.3.4, as reported by the community. Alternative solutions include updating custom nodes, modifying the workflow to remove the attention_mask, using a different sampler node, or performing a clean reinstall. To prevent this error in the future, pin your ComfyUI version, test updates in a separate environment, and keep custom nodes up to date.