ComfyUI missing nodes: fix “node types were not found”

A ComfyUI workflow is a list of nodes, and every node has a type. The built-in types ship with ComfyUI; everything else comes from custom node packs that the workflow’s author had installed. When you load a workflow from Civitai, Reddit or a friend, ComfyUI looks up each type, and the ones it cannot find are listed in that dialog and drawn in red on the canvas.

Nothing is broken yet. The workflow simply cannot run until every red node has a pack behind it, and every model it names is on your disk.

ComfyUI-Manager reads the workflow and matches its missing types to packs it knows about. It is free and solves most cases in a minute.

1. Load the workflow so the red nodes are on the canvas.
2. Open Manager and choose Install Missing Custom Nodes.
3. Install every pack in the list, not only the first.
4. Restart ComfyUI completely, then refresh the browser tab.
5. Load the workflow again and check that nothing is red.

Some packs are not in Manager’s list, and some workflows name a node from an old version of a pack that renamed it. Copy the exact node type from the dialog and search for it on GitHub or in the Comfy Registry; the pack’s README usually says which version introduced it.

If the type is a built-in one, the fix is not a pack at all: your ComfyUI is older than the workflow. Update ComfyUI (Manager’s Update ComfyUI, a git pull, or the update script in the portable build) and reload.

Start ComfyUI from a terminal and read the lines it prints while loading custom nodes. A pack marked IMPORT FAILED is on disk but did not load, almost always because a Python dependency is missing or the wrong version.

Install the pack’s requirements with the same Python that runs ComfyUI. In the Windows portable build that is the embedded interpreter in the python_embeded folder, not the Python on your system; installing into the wrong one is the most common reason the error survives.

Nodes load but the run fails with “Value not in list” on a checkpoint, LoRA, VAE or other model input. The workflow names a file you do not have, or has it under another name.

Find the exact file the author used, usually on a Civitai version page or Hugging Face, and put it in the matching folder under models: checkpoints, loras, vae, and so on. Then pick it in the node’s dropdown. A model with the same name but a different version can load and still give very different results, so match the version, not only the name.

Every pack you add runs its own Python code and can change shared dependencies. Install packs for the workflows you actually use, not whole collections, and note which versions work together. When an update breaks things, disabling the newest pack in Manager is faster than reinstalling ComfyUI.

Workflows that need the most packs are face, detailer and video workflows, the ones an AI influencer uses daily; see consistent characters in ComfyUI and Wan 2.2 reels.

Musetta runs on your own ComfyUI. Its assistant, the Brain, reads a workflow before it runs, finds every missing node pack and model file, installs them, restarts ComfyUI, fixes settings your install lacks, and only says ready once a test run works. Try the demo above, or see how it compares with plain ComfyUI.

If you are weighing whether to keep a node graph at all, the ComfyUI alternatives guide compares the simpler front ends honestly.