Apply Style Prompt (Texturaizer)

A small text-combination node that does exactly what its name says: it takes your base prompt and a separate style prompt and merges them, with a boolean to switch the merge off entirely so you can A/B a render with and without the style applied. In Texturaizer's Blender workflow this is how a per-object or per-scene "base" description stays separate from a reusable style you want to apply consistently across renders - you don't have to hand-edit the prompt string every time you want to toggle a style on or off.

How it works

It takes three prompt strings and one switch. If apply_style is true, it concatenates (or otherwise merges - the exact string logic isn't detailed in the schema, just the behavior) style_prompt into base_prompt_g and base_prompt_l and returns the combined result as two output strings. If apply_style is false, you'd expect the base prompts to pass through unmodified - which is the whole point of having the toggle live on the node instead of just deleting the style text by hand.

Why two base-prompt fields instead of one? This node is built to feed directly into Texturaizer_ClipEncodeSwitchVersion's SDXL-mode inputs - that node takes text_g and text_l as SDXL's dual prompt fields, and this node's two prompt outputs are shaped to match.

The inputs and outputs that matter

• apply_style (default true) - the on/off switch for the whole merge.
• base_prompt_g / base_prompt_l - your starting prompt text, split the same way SDXL's dual encoder wants it.
• style_prompt - the text you want layered in.

Outputs are PROMPT G and PROMPT L, the merged (or passed-through) result - wire both straight into a Texturaizer_ClipEncodeSwitchVersion node with use_sdxl on, or into any other dual-prompt text encode you're using.

Installing it

ComfyUI Manager → search Texturaizer → Install → restart, or clone directly:

cd ComfyUI/custom_nodes
git clone https://github.com/LatentSpaceDirective/ComfyUI-Texturaizer

Restart ComfyUI. It's a pure text-manipulation node - no models, no extra dependencies, nothing to download.

Where people get burned

If you're not on an SDXL-style dual-encoder setup, feeding both base_prompt_g and base_prompt_l the same text and only using one of the two outputs downstream is a perfectly reasonable way to use this node on a single-prompt model - just be aware the node itself doesn't know or care whether your checkpoint actually wants two separate fields; that judgment call is yours. And because apply_style defaults to true, it's easy to forget it's there: if a render looks "styled" when you expected a clean base prompt, this switch - sitting quietly upstream of your text encode - is the first place to check before you go hunting through the rest of the graph.