Banana_Styles Styler

Let's be honest up front: this node is a dropdown of ten bananas. Standard Yellow Banana, Plantain, Red Banana, Frozen Banana, Banana Smoothie. It is not a serious styling tool and nobody pretends it is. What it is - and this is the useful framing - is the pack's worked example: a tiny custom menu that shows exactly how MileHighStyler turns a folder of JSON into a working node. If you've ever wondered "how would I add my own styles," Banana_Styles is the answer sitting right there in the node list.

So don't come here to make good images. Come here to understand the machine, then go build your own menu with it.

How it works (which is the whole point)

MileHighStyler is a fork of SDXL Prompt Styler. Every menu is just a folder of JSON entries; each entry is a template like "organic banana, {prompt}, ripe" with a {prompt} slot where your text goes. At startup the node loads those folders and generates one node per menu - Banana_Styles included. Pick an entry, and the node drops your text_positive into the {prompt} slot and wraps the style words at a strong weight (~2.0). Output is text, straight into your CLIP Text Encode.

Want your own menu? Copy the banana folder, replace the JSON with your styles, restart ComfyUI, and you've got a new styler node named after the folder. That's the trick this node exists to demonstrate. The bananas are the "hello world."

The inputs and outputs that matter

• banana_styles - the ten-entry dropdown.
• text_positive - your subject, dropped into the template.
• text_negative - optional negative text.

log_prompt prints the assembled prompt to the console - which, on a demo node, is genuinely the most useful thing here: run it once, read the log, and you'll see precisely how a template becomes a prompt. Outputs are text_positive and text_negative strings.

Installing it

ComfyUI Manager: search ComfyUI_MileHighStyler, install, restart. Or:

cd ComfyUI/custom_nodes && git clone https://github.com/TripleHeadedMonkey/ComfyUI_MileHighStyler

then restart. No models, no dependencies - Python plus a data/ folder of JSON. Nodes appear under ali1234/stylers.

Where people get burned

• Taking it seriously. It's a demo. "Chocolate-Dipped Banana" at weight 2.0 will absolutely try to put chocolate and a banana in your image. Working as intended; not a bug.
• Not using it to learn. The one real mistake is scrolling past this node without opening the JSON. It's the clearest thirty-second lesson in the whole pack on how custom stylers work.
• Model era. Like every node here, the weighting is SDXL/SD1.5 syntax - silently ignored on 2026 LLM-encoded models (Flux 2, Z-Image). Fine, because you're not shipping banana images anyway.