easy negative

This one's about as simple as a node gets: a multiline text box that outputs a string, meant to hold your negative prompt. That's it. No encoding, no model - just text in, string out. The point isn't power, it's tidiness: instead of retyping low quality, blurry, bad hands, extra fingers, watermark into every workflow, you keep it in one labeled node and wire that string wherever a negative prompt is needed. Its sibling easy positive does the same for the positive side.

Why would a pack full of clever loaders bother shipping a glorified text field? Because Easy-Use is a quality-of-life pack, and half of quality-of-life is naming things. A generic "primitive string" node in the middle of a graph tells you nothing; a node literally labeled easy negative tells you what that text is for at a glance. When you're staring at a workflow you built three weeks ago, that labeling earns its keep.

How it works

There's no mechanism to speak of. You type into the box, the node outputs exactly that text as a STRING. It doesn't run CLIP or touch a model - it's a value holder. The encoding happens later, wherever you feed the string: a CLIP Text Encode node, an Easy-Use loader's negative input, or a easy pipeEditPrompt. Keeping the text separate from the encode is what lets you route the same negative into several places, or swap it in one spot and have it propagate.

The input and output

• negative - the multiline text box. Put your negative prompt here.
• Output: negative (a STRING). Wire it into any node that accepts prompt text.

That's the entire node. No settings to get wrong.

How to install it

Via ComfyUI Manager: search "ComfyUI Easy Use", install, restart. Or clone it:

cd ComfyUI/custom_nodes
git clone https://github.com/yolain/ComfyUI-Easy-Use

Then install.bat / pip install -r requirements.txt, restart. Preinstalled on comfy.icu.

Common issues

The trap here isn't the node, it's the era. A big negative prompt is an SD 1.5 / SDXL habit, and on those models it works fine. But on the guidance-distilled models most people run in 2026 - Z-Image Turbo, Flux 2 Klein distilled, and the like - the negative prompt is inert. Those models run at CFG 1, and at CFG 1 there's no unconditional pass for a negative to steer against, so the text just does nothing. It's not weak, it's off. If you drop easy negative into a Flux or Z-Image workflow, load it up with blurry, deformed, bad anatomy, and see zero effect, that's expected - the node is fine, the model doesn't read negatives. State those constraints positively in your prompt instead, or look at a NAG-based approach.

On SD 1.5/SDXL, none of that applies and this is just a clean way to park your go-to negative. Also worth knowing: it outputs a raw string, so if you're wiring it somewhere that expects already-encoded CONDITIONING, you still need a CLIP Text Encode (or an Easy-Use loader) in between.