Custom Node Troubleshooting: Disable First, Then Bisect

Startup errors, nodes turning red, or freezing right after installing new nodes? Don't rush to reinstall ComfyUI—most ComfyUI custom node errors can be located in ten minutes: first disable all nodes to confirm the scope, then use bisection to pinpoint which one it is.

Symptoms

If any of these occur, suspect custom nodes first:

• At startup, the console floods with red import failure errors (like Failed to import and Cannot import … module for custom nodes), and nodes are missing from the node menu.
• After loading a workflow, some nodes appear in red on the canvas.
• Right after installing a new node package, ComfyUI gets stuck on the startup screen or crashes outright.

First reactions:

Startup logs show errors like Failed to import

First, launch once with --disable-all-custom-nodes to separate "node problems" from "issues with ComfyUI itself".

Nodes appear in red on the canvas

Confirm the node package is installed and enabled; if it is still red after refreshing, the package failed to import—go back to the startup logs to find the cause.

Startup hangs or crashes after installing new nodes

First disable all nodes to restore startup, then use bisection to find the problematic node.

Quick Check (60 seconds)

• Recently installed or updated a node package? Roll it back or uninstall it first.
• Using the portable version? You'll need to modify the startup .bat to add parameters when troubleshooting.
• Some nodes download models from the internet on first launch—check your network and proxy first.
• Dependencies must be installed into the Python environment that ComfyUI actually uses, not the system Python.

Three-Step Troubleshooting

Step 1: Disable All Custom Nodes and Start

Portable version: open the startup .bat you normally double-click with Notepad (run_nvidia_gpu.bat for NVIDIA GPUs, run_amd_gpu.bat for AMD, run_cpu.bat for CPU), append --disable-all-custom-nodes to the end of the startup command, save, and restart.

Source or venv installation: add the same parameter after the startup command.

If it starts normally, the problem is in the nodes—go to step 2; if errors persist, the problem is in ComfyUI itself or the environment—go back to the Troubleshooting Overview first.

Step 2: Use Bisection to Locate the Problematic Node

Split the node packages under the custom_nodes directory into two equal halves, temporarily move one half out of the directory (cut it elsewhere as a backup), and start a test with only the other half remaining:

• Errors: the problem is in the half that stayed—keep splitting it in half.
• Normal: the problem is in the half that was moved out—swap it back and split again.

Repeating this for two or three rounds will pinpoint a single node package: 10 packages take at most 4 rounds, much faster than testing one by one.

Step 3: Fix or Replace the Problematic Node

• Update to the latest version first: most errors happen because nodes haven't kept up with the new ComfyUI interface.
• Install missing dependencies per the README: go into the node directory and run pip install -r requirements.txt@@I18N_KEEP_000022@@ using the Python that ComfyUI runs on.
• Still failing: file an issue at the node's repository with the full startup log, or delete it and switch to an alternative node with the same functionality.

When to Escalate

• Still won't start even after disabling all nodes: it's not a node problem—go to Model Loading Failures.
• Log errors point to torch, CUDA, or VRAM rather than imports: it's an environment problem—don't waste time on nodes.
• Still erroring after updating and reinstalling dependencies: seek help in the node repository's issue section, attaching the full logs and the ComfyUI version number.

---

Related Reading: Troubleshooting Overview · Model Loading Failures · Template Library: Load Official Workflows in One Click