🎨 Prompt Stylist

Here's the twist in this pack: 🎨 Prompt Stylist is the one node that never touches a model. No LLM, no Ollama, no network call. It's a deterministic string-prepender - take your prompt, stick a style: header on top, hand it back. That makes it the odd one out in OllamaTools, and also the most predictable one in it. Same input, same output, every single time. In a pack of LLM-powered generators, that determinism is the feature.

Why would you want that? Because styling through an LLM is a dice roll. Every run you get slightly different wording, which makes batch comparison maddening. With this node you can flip the style dropdown, queue five gens, and know the only thing that changed is the style line. That's a clean A/B experiment. It's also instant - no model load, no warm-up, works even while Ollama is down.

How it works

It's genuinely just dictionary lookups. Pick a style_tone and the node prepends a fixed phrase to your prompt - for example style: oil painting, textured brushstrokes, painterly composition followed by a newline and your text. That's the whole mechanism, and it's the standard "style prefix" trick that SDXL and Flux-era models respond to: put the style early, where the model pays the most attention. There's no magic, and honestly there doesn't need to be.

The inputs that matter

• prompt - your existing prompt, multi-line. This is what gets the style slapped onto.
• style_tone - the dropdown: None, Oil Painting, Photography, Comic Drawing, Anime Drawing, Commercial. None passes your prompt through untouched.
• nsfw - a boolean that swaps each style for an uncensored variant (glamour photography, erotic comic, "lewd anime," and so on). Note that it only does anything when a style is actually selected - with style_tone: None it's inert.

Output: a single styled_prompt string. Wire it straight into the positive input of a CLIPTextEncode or CLIPTextEncodeSDXL.

Installing it

Same pack as the others - ComfyUI Manager, search OllamaTools, or:

cd ComfyUI/custom_nodes
git clone https://github.com/max-dingsda/OllamaTools.git

Restart ComfyUI and you're done. This node specifically needs nothing else: no model downloads, no Ollama install, no API key. It's pure Python string work.

Where people get burned

The honest take: this is a lightweight utility, not a stylization solution. A style: prefix is a nudge; a style LoRA or an IPAdapter is a guarantee. If you want convincing oil painting, use a proper style model - this node gets you a reasonable starting flavor for iteration, and it's great for quickly testing whether a style direction has any legs before you commit to heavier tooling. One more gotcha: whether the style line actually lands depends on your checkpoint. SDXL and Flux-style models read natural-language style instructions well; pure tag models (Danbooru-trained) may largely ignore the English prose. Know what you're prompting, and you won't be disappointed.