How to Fix KeyError: CLIPTextEncode in ComfyUI Workflows

The Error Message

You're loading a workflow in ComfyUI — then this hits:

KeyError: 'CLIPTextEncode'

The backend tried to look up CLIPTextEncode in its node registry and came up empty. Since this is one of ComfyUI's core nodes, a missing definition almost always points to a broken install, an outdated build, or a corrupted workflow file.

Why This Happens

Four causes account for most cases:

• Version mismatch: You're running a workflow built on a newer ComfyUI commit. Your local copy doesn't know about that node's updated implementation yet.
• Corrupted installation: Files inside comfy/ or nodes.py have been modified, partially deleted, or overwritten by a bad update.
• Mangled JSON: Workflows copied from web snippets, pastebin, or chat messages often get garbled in transit. The type field can end up as CLIPTextEncod — one missing character is enough to trigger the error.
• Custom node conflict: Some third-party nodes try to override core nodes. If such a node crashes on load, it can silently unregister CLIPTextEncode along with itself.

Fix 1: Update ComfyUI Core

Version mismatch is the most common culprit. Pull the latest changes before trying anything else.

Open a terminal in your ComfyUI root folder and run:

git pull
pip install -r requirements.txt

On the portable Windows build? Run update_comfyui.bat from the update/ folder instead. Restart the server once it finishes.

Fix 2: Use ComfyUI-Manager to Restore Missing Nodes

Some workflows rely on a custom CLIP encoder — an SDXL-specific conditioning node, for example — that your system doesn't have installed. ComfyUI-Manager can scan for gaps and fill them in.

• Open ComfyUI in your browser.
• Click Manager in the side panel.
• Select Install Missing Custom Nodes.
• If CLIPTextEncode or any conditioning node appears in the list, install it and restart.

Fix 3: Repair the Workflow JSON

Sometimes the problem isn't your install — it's the file. Workflows shared as Base64 strings or pasted through chat often lose characters in transit. A single misquote or stripped bracket is enough to make the node type unreadable.

Decode the string manually and look for "type": "CLIPTextEncode" inside the nodes array. I use this Base64 tool to decode and inspect the raw JSON. If the workflow came in as a URL-encoded string, run it through a URL decoder first — special characters like + and %22 can corrupt field names silently.

Fix 4: Delete and Reconnect the Node

Workflow opens but shows a red box where the prompt node should be? The node ID is probably orphaned. Here's what to do:

• Note which wires connect to the broken node (CLIP input, conditioning output).
• Delete the red CLIPTextEncode node.
• Right-click → Add Node → conditioning → CLIP Text Encode (Prompt).
• Reconnect the wires.
• Save the workflow as a new JSON file.

Prevention

A few habits that will keep this from coming back:

• Pull weekly: ComfyUI ships node definition changes frequently. A weekly git pull keeps you in sync and avoids version drift.
• Back up stable workflows: Don't rely on last_workflow.json. Save working setups as named JSON files in a dedicated folder.
• Use a virtual environment: A dedicated venv or Conda environment ensures other Python projects can't quietly break ComfyUI's dependencies.

Verifying the Fix

Restart ComfyUI, then hard-refresh your browser with Ctrl+F5. Load the workflow. A healthy CLIPTextEncode node shows a green or violet header — no red outline. Queue a prompt. If the console prints something like Prompt executed in 3.2 seconds, you're done.