Apply SDXL Style TWRI

Apply SDXL Style TWRI is the everything-bagel of the mozman pack: 89 styles in one dropdown, organized by category. The prefixes tell the story - ads- (advertising, automotive, corporate, food photography, real estate), artstyle- (abstract, baroque, anime...), futuristic-, game-, papercraft-, photo-, plus a big misc- bucket. Whatever direction you're pointing a subject, there's probably a template for it.

This is the node you reach for when you don't know what you want yet. Product shot → ads-luxury or ads-food photography. Concept art → game- or futuristic-. Illustration → artstyle- or papercraft-. It's the same "prefix/suffix style presets are a cheat code" trick the community swears by, just with the widest menu in the pack. Where the SAI set is focused and curated, this one is broad and shallow - which is precisely its value for exploration.

Fair warning about the name: "TWRI" is unexplained anywhere I can find, and the set circulates as sdxl_styles_twri.json with no solid origin story. It's a community collection, well-sorted into categories, and that's honestly all you need to know to use it.

How it works

Same machinery as every node in the pack. Pick a style, the node drops your prompt into the template's {prompt} placeholder and appends the template's negative to yours. Strings in, strings out, before CLIP Text Encode. No model, no API, no downloads.

Inputs and outputs that matter

• positive_prompt / negative_prompt - your prompt strings (multiline).
• style - the dropdown, 90 entries including bypass at the top. At this size the menu is unwieldy, so the practical workflow is: leave it on bypass, pick a style when you actually want one, and use bypass as your unstyled baseline for A/B tests.
• apply_negative_style (default yes) - off if you're supplying your own negative.
• log_prompt (default off) - with 89 templates you can't memorize, turning this on once and reading the console is the fastest way to learn what's actually in a style.

Outputs are positive_prompt and negative_prompt (STRING) → your CLIP Text Encode node.

Install

Part of ComfyUI_mozman_nodes - one install, seven nodes. ComfyUI Manager (search ComfyUI_mozman_nodes), or:

cd ComfyUI/custom_nodes
git clone https://github.com/mozman/ComfyUI_mozman_nodes

Restart after. No requirements.txt, no model files, pure stdlib, from Manfred Moitzi (the ezdxf author). Boring and safe.

Where people get burned

The 89-entry menu is the obvious pain point - it's a long scroll, and there's no search or preview in the dropdown. If you find yourself digging through it constantly, the workaround is to pick your handful of winners, note them, and leave the node set on one of them with bypass as your off switch.

Then the usual family caveat, which matters more at this scale: these are SDXL-CLIP templates. On guidance-distilled or LLM-encoded models the negatives are inert at CFG 1 and the phrasing doesn't transfer. Keep this one on SDXL lineage and let the breadth do its job.