ComfyUI Mexx Styler Advanced

Same pack, same 19 templates, different plumbing. MexxSDXLPromptStylerAdvanced is the sibling of MexxSDXLPromptStyler for the part of the SDXL crowd that doesn't feed one flat string to a single text encoder. SDXL runs two CLIP encoders, and if your workflow uses the CLIPTextEncodeSDXL node with separate text_g and text_l inputs, this is the styler that speaks your language.

It's still a pure text node - no models, no API, nothing to download. It takes your prompts, wraps them in the same Chinese-named style templates from mexx_styles.json, and hands back separate G and L streams instead of one blob.

How it works

The template engine is identical to the base node - find the style, replace {prompt}, append the baked-in negative. The difference is split_template_advanced, which looks for the marker {prompt} . inside a template and, when it finds it, splits the style keywords between the G (main) and L (auxiliary) halves.

Here's the honest catch: none of the shipped 19 styles actually contain that marker. So in practice the whole style template lands in the G prompt and your text_positive_l rides along untouched. That's not a bug - it means the per-encoder splitting is scaffolded and waiting, but the current templates don't use it. What you get today is a styled G prompt plus your raw L prompt, and a combined text_positive built as G . L - exactly the separator convention the dual-encoder workflows expect. Add {prompt} . to a template in mexx_styles.json yourself and the split starts doing real work.

Inputs that matter

• text_positive_g and text_positive_l - your two encoder-targeted prompts. The style keywords go into G; L stays as you wrote it.
• text_negative - your negative, merged with the template's.
• style - same 19-look dropdown as the base node.
• negative_prompt_to - where the merged negative goes: Both (default), G only, or L only. Set it and forget it unless you're deliberately starving one encoder.
• log_prompt - defaults to No here, the opposite of the base node, so you won't get console spam unless you ask.

Outputs and wiring

Six outputs, a clean grid: text_positive_g, text_positive_l, text_positive (the combined string), then text_negative_g, text_negative_l, text_negative. Two ways to use them:

• Feed the separate g/l outputs into a CLIPTextEncodeSDXL node's text_g/text_l (and the negative streams into its negatives) for full per-encoder control.
• Or just take the combined text_positive and text_negative into a plain CLIPTextEncode and treat it like the base node.

The combined output exists precisely so you never feel forced into the dual-encoder path.

Install and quirks

Same as the base node - Manager search "Mexx Styler", or git clone https://github.com/SoftMeng/ComfyUI_Mexx_Styler into custom_nodes, then restart. No requirements.txt, no models.

One quirk to know before you hunt for it in the UI: the two nodes aren't in the same menu. The base styler sits under ComfyUI_Mexx while this one is filed under utils, so it won't appear next to its sibling. Minor, but it has tripped up people clicking through the node menu expecting them grouped. Otherwise it's the same tiny, dependency-free pack - worth a look if you run SDXL with dual-encoder prompt splitting and want a curated style dropdown that doesn't fight your setup.