CLIP Text Encode (Simple) ✍️

Let's be upfront: this node is a thin, branded wrapper around ComfyUI's own core CLIPTextEncode. The tokenization, the embedding resolution, the embedding:name syntax - all identical to the standard node, because the code literally calls the core node's encode() method. So why does it exist? Two reasons, and both are about graph hygiene rather than capability.

First, it adds an is_positive BOOLEAN label. Here's the honest catch from the source: that flag does not change the encoding. CONDITIONING has no inherent positive/negative flag until a sampler treats it as one. It's purely a label so you can glance at a graph and know which wire is the positive prompt and which is the negative - genuinely useful when you're re-opening a workflow you built three weeks ago. Second, it returns a summary STRING that shows a short preview of what was encoded, labeled [positive] or [negative]. Wire that into a text display and you get a running log of exactly what prompt produced each render.

The two inputs you'll actually touch

• text - your prompt, multiline, with the standard CLIP weighting syntax ((word:1.3), [word], BREAK) working exactly as it does in the core node.
• clip - your CLIP model from the checkpoint loader.

is_positive is the label, as covered. Output is conditioning (straight into a KSampler) plus the summary.

When to use it (and when not to)

If you're rebuilding a workflow from scratch and want the graph to read well, this is a reasonable drop-in for core CLIPTextEncode - same behavior, bonus label, bonus summary. If you're editing an existing shared workflow, there's zero reason to swap nodes; the output is identical and you gain nothing by replacing a working core node. And if you need to measure or manipulate the conditioning rather than just encode it, OmniNodes has better tools for that: CLIP Text Compare for similarity scoring and CLIP Text Weight for scaling the conditioning.

One honest caveat from the prompt-engineering side of things: weight syntax only does anything on CLIP-era encoders (SD1.5, SDXL, Illustrious, Pony). On LLM-encoded models like Flux or Z-Image, (word:1.4) is passed through as literal punctuation - that's an encoder limitation, not this node's, since it inherits core behavior exactly.

Install

ComfyUI Manager → search OmniNodes, or:

cd ComfyUI/custom_nodes
git clone https://github.com/TensorVizion/OmniNodes

Restart ComfyUI; it lives under TensorVizion/Prompt. No extra dependencies - this is a pure wrapper around a core node, so install is zero-friction.

Is it essential? No - that's the honest review. Is it a nicer-looking, self-documenting replacement that costs you nothing? Also yes. If you like graphs you can read at a glance, grab it.