CLIP Text Encode (Auto-Tags)

CLIP Text Encode (Auto-Tags) (CLIPTextEncodeWithTags) is a drop-in replacement for ComfyUI's built-in CLIP Text Encode that does exactly one extra thing: it appends a tags string to your prompt before encoding. Wire the TAGS output from the LoraTags loader into it and your saved trigger words appear in every generation automatically. Forget them, and nothing does.

It's the second half of the ComfyUI-LoraTags story. The loader remembers your trigger words and hands them to you as a string; this node is what turns that string into actual conditioning. Standalone it's just a text encoder with a spare input - fine, but boring. The magic is that pair.

How it works

Mechanically, it's embarrassingly straightforward. It takes your text, takes the incoming tags, and joins them with ", ". If your prompt is empty it uses the tags alone. Then it runs the standard CLIP pipeline - tokenize, encode - and returns the conditioning exactly like the core node. Because it delegates to whatever CLIP you feed in, it works with any architecture ComfyUI wraps: SD 1.5, SDXL, and the LLM-encoded models alike. There's no voodoo, no LoRA loading, nothing to break.

One quirk to know: it prints the combined prompt to the ComfyUI console on every run. Harmless, mildly useful for debugging, mildly annoying if you generate a lot.

The inputs that matter

• text - your base prompt, multiline. The part you actually type.
• clip - the CLIP model, from your checkpoint loader.
• tags - the string to append. This is the interesting one: it's marked forceInput, which turns a text field you'd normally type into into a real input wire on the canvas. That's the design choice that makes auto-tagging possible.

The single output, CONDITIONING, wires into the KSampler's positive input exactly like the output of any text encoder.

Install

Same pack as the loader - you get both in one shot:

cd ComfyUI/custom_nodes/
git clone https://github.com/iiTzMYUNG/ComfyUI-LoraTags.git

Restart ComfyUI and hard-refresh your browser (Ctrl+Shift+R) so the custom UI loads. No extra dependencies, no models to download; the whole pack runs on ComfyUI's own internals.

What to watch for

The node merges your prompt after you've written it, so if you like putting trigger words at the front of your prompt, this fights you a little - the tags always land at the end. Most prompts don't care; the tag position isn't usually what decides the result. And since it only appends whatever string arrives on tags, an empty TAGS from the loader (every LoRA toggled off) just means you get your raw prompt back - no error, no harm.

One honest caveat: because the auto-appended tags go through the same token budget as your base prompt, a long prompt plus a pile of trigger words can push you toward the 77-token truncation on SD 1.5/SDXL-class CLIP encoders. If you're stacking half a dozen LoRAs with verbose triggers, keep an eye on that. It's a small node with a small job - and for the trigger-word-forgetting crowd, it's genuinely the part that sells the pack.