ComfyUI Styler (Advanced)

The name undersells it. "ComfyUI Styler" sounds like a utility; in reality ComfyUIStylerAdvanced is the whole buffet - 34 dropdowns and over 107,000 style options bolted onto a single node. If you've ever wanted to pick an art style, a camera, a lighting mood, and an adjective all from menus instead of remembering prompt syntax, this is the node you're actually looking for.

It ships in aegis72/comfyui-styles-all, a preservation clone of the SDXL prompt-styler styles azazeal04 removed from GitHub in January 2024. The author copied the catalog so it wouldn't vanish - which makes this pack the reason those styles are still downloadable today. ComfyUIStylerAdvanced is the pack's flagship: every theme menu exposed at once.

What's on the node

The 34 dropdowns split into a few sensible groups:

• Theme families - Aesthetic, Anime, Fantasy, Gothic, Halloween, Line_Art, Movie_Poster, Punk, Travel_Poster. These are hand-written scene/art-movement presets (the pack's own single-theme nodes each expose just one of these).
• Technical categories - camera (127 bodies), lighting, composition, filter, LUTs, focus, depth, timeofday, mood, subject, theme.
• Content modifiers - artist (7,743 artists), materials (1,237), nouns (90,965), verbing (3,501), adjective, environment, face, preposition.
• Some NSFW-leaning ones - breast_state, clothing_state, and focal point include body-focused options. They're just style templates, but don't be surprised.

Every menu's first option is None (pass-through), so you can leave most of them alone and still get value from the two you actually want.

How it works

Each option is a JSON template: a prompt string containing a {prompt} placeholder plus a negative_prompt. The node substitutes your text into the placeholder and appends the style's negative to your negative. Dropdowns are applied in a fixed order, each one wrapping the result of the previous - so the menu order matters, and picking a lot of styles at once stacks parentheses-heavy keyword blocks.

The "Advanced" half is SDXL's dual-encoder split. Templates containing a . separator (mostly the milehigh menu) split into a G (global) part and an L (local) part, so the node can feed SDXL's OpenCLIP-G and CLIP-L encoders separately.

The inputs and outputs that matter

• text_positive_g / text_positive_l - your prompt, split for SDXL's two encoders.
• text_negative - your negative; the style's negative gets appended.
• negative_prompt_to - Both (default), G only, or L only. Controls which encoder gets the negative.
• log_prompt - defaults on; prints every selection and the before/after prompts to the console. Genuinely useful for seeing what 34 menus did to your sentence.

Outputs: text_positive_g, text_positive_l, text_positive (combined), and the matching text_negative_g, text_negative_l, text_negative. For SDXL, wire text_positive_g/text_positive_l into CLIPTextEncodeSDXL and the two negatives into its negative inputs. For SD 1.5, just use the combined text_positive into a normal CLIPTextEncode - the G/L split is optional there.

Installing it

ComfyUI Manager (search ComfyUI-styles-all), or:

cd ComfyUI/custom_nodes
git clone https://github.com/aegis72/comfyui-styles-all

Then restart ComfyUI. No requirements.txt, no model downloads - pure stdlib Python plus JSON.

Where people get burned

• Menu changes need a restart. The dropdowns are built from data/*/ JSON files when ComfyUI starts. Edit a file or add your own JSON and you must restart before it appears. (That's also how you add custom styles.)
• It's weaker on Flux. These templates were written for SDXL's CLIP encoders. On Flux and other LLM-encoded models the same keywords do far less - a common complaint with every SDXL styler.
• Don't set everything at once. A beginner instinct is to fill all 34 menus; you'll get an unreadable prompt. Start with one theme family and maybe a lighting, and read the console output.