Style Chooser

Lots of "style" nodes in ComfyUI are actually a pile of hidden tricks - baked-in embeddings, negative-prompt swaps, weights you can't see. Style Chooser is the opposite of that. It's a dropdown of eight named styles, a custom text box, and a string you plug straight into a CLIP Text Encode. You reach for it when you want an artist's look without re-typing the same clause across ten workflows.

What it is

Style Chooser gives you eight presets: Modern Comic Book Art, Anime, Gil Elvgren, Boris Vallejo, Casey Baugh, Frank Frazetta, George Hurrell, and Humberto Ramos. The painter list - pin-up and fantasy names, a Hollywood portrait photographer, a couple of comic artists - tells you something about the author's own taste, and also about its limits: if you want cyberpunk or film noir, that's what the custom text box is for.

How it works

Each dropdown label maps to a hardcoded prompt fragment with a trailing comma. Modern Comic Book Art → modern comic book art style,, Frank Frazetta → art style of Frank Frazetta,, Anime → anime art style,. Then it builds two outputs:

• style_text - the mapped fragment as-is, comma and all.
• combined_text - the fragment plus your custom_text if you typed any, or the fragment with the trailing comma stripped if you didn't.

No weights, no embeddings, no hidden state. Just string assembly.

The inputs and outputs that matter

• style - the dropdown, default Modern Comic Book Art. Pick one and it becomes a prompt fragment.
• custom_text - a multiline text box, empty by default. Your actual subject, or extra style direction.

Outputs:

• style_text (STRING) - the clean preset on its own.
• combined_text (STRING) - preset plus your text, e.g. art style of Frank Frazetta, a barbarian queen in a torchlit hall. This is the output you'll usually wire - one cable into the positive input of a CLIP Text Encode and you're done.

The model-family caveat

Artist-name clauses are an SDXL-era staple, and how well they land depends entirely on the text encoder. On SDXL, Pony, and Illustrious-derived checkpoints, art style of Frank Frazetta, behaves exactly the way the tag ecosystem expects. On the 2026 LLM-encoded models - Flux, Qwen, Chroma - the same fragment is read as natural language and it's thin. "In the heroic fantasy style of Frank Frazetta, dramatic lighting, rich oil-paint texture" pulls far harder than the bare name. So this node is most at home on SDXL-lineage models; on newer ones, lean on custom_text and treat the dropdown as a starting point rather than a complete prompt.

Installing it

Style Chooser ships in the Candlehead-Comfy-Nodez pack, which has zero dependencies - no requirements.txt, no model files. Install once, get all four nodes:

ComfyUI Manager → Custom Nodes → search "Candlehead" → install → restart.

Or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/Candlehead/Candlehead-Comfy-Nodez

Restart, then search the menu for "Style Chooser" (the class name is StyleChooser).

Gotchas

• The trailing comma is a real thing. style_text always ends with one. If you wire it into a prompt you're building yourself, you'll get style,, more text. Use combined_text and the comma is handled for you.
• "Anime" is the only broad category - everyone else is a named artist, so the preset list has real gaps.
• No weights, no magic. This node returns raw text; there's no (style:1.2) under the hood. Weigh it yourself if you need to.
• Defaults matter. Clear the dropdown and it snaps back to Modern Comic Book Art - a silent surprise if you expected an empty prompt.