Fixing "The Following Node Types Were Not Found" in LTX 2.3 ComfyUI Workflows

You load an LTX 2.3 workflow into ComfyUI and instead of your graph, you get a popup: "When loading the graph, the following node types were not found." Underneath it is a list of node class names, and the corresponding nodes in your workflow are highlighted red. The workflow won't run until every one of those nodes resolves correctly.

This is not an LTX-specific bug. It's ComfyUI's generic way of saying "a node in this workflow's JSON doesn't exist in your current install." For LTX 2.3 workflows specifically, it almost always means the custom node pack that ships the LTX loaders, samplers, and VAE nodes isn't installed, isn't fully installed, or failed to import.

Why this happens

A ComfyUI workflow file just stores node type names as strings. When you load it, ComfyUI tries to match each name against the nodes it currently has registered — built-in nodes plus whatever custom node packs are in custom_nodes/. If a name doesn't match anything registered, it can't be drawn properly, so ComfyUI flags it as "not found" and shows it in red.

For LTX 2.3, that usually points to one of three situations:

1. The node pack isn't installed at all. LTX 2.3 workflows depend on custom nodes from the official Lightricks/ComfyUI-LTXVideo repo — things like LTX-specific loaders, the LTX VAE decode/encode nodes, and the Gemma 3 text encoder loader used for prompt conditioning. If that repo was never added to custom_nodes/, every one of those node types will be missing.
2. It's installed but never registered, because ComfyUI hasn't been restarted since the install. Refreshing the browser tab doesn't reload Python modules on the backend — only a full ComfyUI server restart does.
3. It's installed but failed to import. This is the case people miss most often. If the node pack's Python code throws an exception on import (a missing dependency, a version conflict), ComfyUI silently drops those nodes from the registry and they show up in the exact same "not found" popup — even though the folder is sitting right there in custom_nodes/. A known real-world example: some ComfyUI-LTXVideo installs have failed to import due to a kornia version mismatch (ImportError: cannot import name 'pad' from 'kornia.geometry.transform.pyramid'), which is a dependency conflict, not a missing-node problem — see ComfyUI-LTXVideo issue #517.

How to fix it

1. Try ComfyUI-Manager's "Install Missing Custom Nodes" first. If you have ComfyUI-Manager installed, open it and use the "Install Missing Custom Nodes" button — it scans the loaded workflow, matches the missing node names against its node database, and installs whatever it recognizes automatically. For LTX 2.3 workflows this should pull in ComfyUI-LTXVideo directly. If Manager's node database is stale, search for "LTXVideo" manually in the Manager's node browser and install from there instead.

2. If Manager can't find it, clone the repo manually. From your ComfyUI install directory:

cd ComfyUI/custom_nodes/
git clone https://github.com/Lightricks/ComfyUI-LTXVideo.git
cd ComfyUI-LTXVideo
pip install -r requirements.txt

If you're on a portable/embedded ComfyUI build, use that build's embedded Python executable to run pip install instead of your system Python, or the dependencies will install in the wrong environment. Full setup details, including which model files to pair with the node pack, are in our ComfyUI setup guide and the models hub if you still need to grab the LTX 2.3 checkpoints.

3. Fully restart ComfyUI — don't just reload the browser. Stop the ComfyUI process (close the terminal window or kill the process) and start it again. A browser refresh only reloads the frontend; it does not reload Python node code on the backend. This step trips people up constantly — they install the nodes, refresh the tab, still see the same error, and assume the install failed.

4. Check the ComfyUI console log, not just the browser popup. After restarting, look at the terminal/console where ComfyUI is running. If ComfyUI-LTXVideo (or any custom node) fails to import, the console prints the real Python traceback right there during startup — a ModuleNotFoundError, an ImportError, a version conflict, whatever it actually is. The browser's "not found" popup is generic by design; it doesn't tell you why a node is missing, only that it is. The console log is where you find the actual root cause.

Still not working?

If the nodes still don't show up after a clean restart:

• Check folder permissions. Make sure the ComfyUI-LTXVideo folder inside custom_nodes/ is actually readable by the user running ComfyUI, especially if you cloned it with sudo or copied it from another machine.
• Check for Python package conflicts. Requirements from different custom node packs can clash (this is exactly what happened with the kornia version issue above). If the console shows an import error naming a specific package, try upgrading or pinning that package version and restarting again.
• Confirm you're reading the backend console, not just browser dev tools. The browser console shows frontend JS errors; it will never show you a Python import traceback. You need the terminal window (or log file) where you launched ComfyUI.

If all of this feels like more setup than you want to deal with, you can skip the local install entirely and run LTX 2.3 workflows through our hosted generator instead — no custom nodes, no Python environment to babysit.

Sources

• Lightricks/ComfyUI-LTXVideo GitHub repository
• ComfyUI-LTXVideo issue #517: Installation/Import Fails (kornia import error)
• ComfyUI forum: Missing Node Types — "When loading the graph, the following node types were not found"
• comfyanonymous/ComfyUI issue #3780: node types not found (UltimateSDUpscale example)