Symptom

- When starting ComfyUI, multiple lines of
(IMPORT FAILED)appear in the terminal, and specific custom nodes cannot be used or do not appear in the node list.
Timing of occurrence
- Immediately after starting ComfyUI.
- When installed custom nodes do not appear in the node list.
Cause
- Essential libraries are missing, or an incompatible Python / PyTorch version is used.
- If you scroll up from
(IMPORT FAILED)in the terminal, you will seeModuleNotFoundError: No module named 'facenet_pytorch', etc., indicating that the library required for that custom node cannot be found.
- If you scroll up from
- Path or OS dependent issues.
- Example: Cases like
UnicodeDecodeError('cp932' codec can't decode byte 0x87...)where having non-ASCII characters in the folder path causes errors.
- Example: Cases like
- Conflict with other custom nodes.
- Sometimes nodes with old frontend implementations or unique dependencies crash because they are incompatible with the new ComfyUI.
Solution
-
If libraries are missing
- If you are familiar with Python, you can
pip installyourself, but otherwise, first open the GitHub of that custom node and check if the installation method is correct. - Currently, if it is via ComfyUI Manager, many nodes install necessary libraries all together. Unless there is a special reason, please install from Manager first.
- Even for nodes where special procedures like "Please run this script manually" are written in README, if you don't understand well, please do not touch it in the first place.
- If you are familiar with Python, you can
-
If conflict with other custom nodes is suspected
- If possible, evacuate custom nodes other than the one outputting
(IMPORT FAILED)fromcustom_nodes, and check if ComfyUI starts with only that node remaining. - After that, return them little by little in combinations of "problem node + one other node" to isolate which node causes the crash when combined.
- If you don't need to use it that much, uninstall that node (delete the folder itself).
- If possible, evacuate custom nodes other than the one outputting