ComfyUI Mexx Styler

MexxSDXLPromptStyler is a style selector: you type a plain prompt, pick a look from a dropdown, and it hands back the same prompt wrapped in that style's keyword template. It never touches the sampler and never touches a model - it's a pure text node that rewrites strings. It's part of SoftMeng's ComfyUI_Mexx_Styler pack, a small Chinese-authored rework of the classic twri/sdxl_prompt_styler idea, and the value is entirely in the 19 curated templates it ships.

Know your audience before you install. The default prompt is "a beautiful k-pop girl with soaked dress", and roughly half the dropdown is portrait photography variants, several explicitly tagged 亚洲人 (Asian). This is built for the photoreal K-pop/AI-photo crowd, and the templates lean that way hard. If that's your genre, this is one of the nicer drop-in style packs around. If it isn't, you'll be fighting the baked-in "k-pop" bias from the first image - probably worth skipping.

How it works

At load, the node reads mexx_styles.json from its own directory - a plain list of {name, prompt, negative_prompt} entries. When you run it, it finds your chosen style, replaces the {prompt} placeholder with your positive text, and prepends the style's keywords (think Leica M50 F/2.8, deep Shadows, sharp focus for 摄影棚/Studio). Your negative gets appended after the template's built-in negative - most photo styles prepend anime, drawing, painting, crayon, sketch to keep you out of illustration territory. The 无 (None) entry is the identity template, so it passes your text through untouched.

That JSON is the whole product, and it's editable. Open mexx_styles.json in the pack folder, add your own {name, prompt, negative_prompt} object, restart, and your style appears in the dropdown. Duplicate names get auto-renumbered, so you can't break the list by being sloppy.

Inputs and outputs

Three inputs matter:

• text_positive / text_negative - your base prompt and negative. Multiline, so paste freely.
• style - the dropdown of 19 looks: studio, lifestyle, model, portrait, Fujifilm (Provia 400X), dark/horror, biomechanical, line art, abstract illustration, and a couple of generic "complex detail / all-around enhancement" boosters.
• log_prompt - defaults to Yes, and this one's a trap: it prints the styled prompts to the ComfyUI console on every run. Handy when you're learning, annoying once it's noise. Flip it to No.

Outputs are just two: text_positive and text_negative, styled. Wire them into a CLIPTextEncode node - positive and negative respectively - and you're done.

Install

No dependencies, no model downloads, no API keys. ComfyUI Manager will find it by searching "Mexx Styler", or:

cd ComfyUI/custom_nodes
git clone https://github.com/SoftMeng/ComfyUI_Mexx_Styler

Then restart ComfyUI. The README's example settings are SDXL at 768×1024, dpmpp 2M Karras at 10 steps, or an LCM checkpoint for fewer. There's a demo workflow that pairs this pack with the author's ComfyUI_Mexx_Poster and an LCM model, if you want the full SoftMeng stack.

Where people get burned

The README is Chinese-only, which is the first hurdle for most English users - though the style names mix in enough Latin photography terms (Fujifilm, Lifestyle, Model) to be guessable. The author themselves flags the two line-art styles (线稿) as "时灵时不灵" - hit or miss. Don't tune your whole workflow around them. Otherwise it's a tiny, honest node: nothing to break, nothing to download, just a curated set of prompts. For a one-click photography styling step on SDXL, it's a solid reach.