Pick Style

Pick Style is the read side of the Immac Style Mixer pack, and the node you'll probably touch first. It's a single dropdown of every style you've saved - pick one, and out come its name, its id, its prompt text, and its negative. No models, no API calls, no downloads. If you've ever retyped the same golden hour, film grain, shot on 35mm snippet into ten different workflows, this is the cure.

It's also the node that appears automatically when you drag a style card from the Immac Style Mixer sidebar tab onto the canvas. The drag-drop wiring is the cleanest onboarding trick in the pack: you organize prompts in the sidebar, then literally throw one at your graph. What you get on the other end is this node, pre-wired to that style.

How it works

Under the hood it's embarrassingly simple, which is a compliment. The dropdown is populated from style_mixer_data.json, the file where the whole pack stores your library. Select a name and the node looks it up, returning the stored fields. There's a fingerprint_inputs hook keyed to the data file's modification time, so when you edit a style in the sidebar the dropdown refreshes on the next run instead of showing stale names.

The whole point of the pack is that a "style" is just a reusable prompt chunk with a name and an optional thumbnail - not a LoRA, not a model, not a Redux embedding. That keeps everything fast and portable, and it means there's no heavy dependency anywhere in the install.

The inputs and outputs that matter

Only one input:

• style - a dropdown of your saved style names. If you've saved none yet it shows a (no styles saved) placeholder. That's your signal to open the sidebar tab and add some.

Four outputs, and you'll use most of them:

• style_value - the prompt text. This is what you wire into your positive conditioning, usually a CLIPTextEncode or your sampler's positive input.
• style_negative - the per-style negative prompt, if you set one. Wire it to the negative side. Worth noting that negatives still earn their keep on SDXL-lineage models (Pony, Illustrious, NoobAI) even in 2026 - that's who this pack is aimed at.
• style_id - the internal UUID. This feeds Weight Style when you want to weight the snippet instead of using it as-is.
• style_name - handy for logging or for driving a Save Style loop later.

Installing the pack

The node ships inside ComfyUI-ImmacStyleMixer, so it's one install for all six nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/Immac/ComfyUI-ImmacStyleMixer

Then restart ComfyUI. There are no pip requirements (the pack uses comfy_api, Pillow and numpy, all already inside ComfyUI) and no model files to hunt down. The UI is pre-built and committed, so you never need Node.js as an end user. ComfyUI Manager can install it too if you paste the repo URL, though this is a young pack so don't be surprised if it's not in Manager's registry yet. One real requirement: the nodes are built on ComfyUI's newer extension API (comfy_entrypoint + comfy_api), so you want a current ComfyUI, not a year-old build.

Gotchas

The library lives in style_mixer_data.json inside the pack's folder. Re-cloning the repo or deleting custom_nodes/ComfyUI-ImmacStyleMixer wipes your styles, which is exactly why the pack ships JSON/ZIP backup and restore in the sidebar. Make it a habit. And if a style you just saved isn't in the dropdown, rerun the graph once - the fingerprint refresh kicks in on execution, not before.