Generate Draw Style Prompt

Style words are the flakiest part of prompting - "painterly" means something different to every checkpoint. Draw Style Prompt fixes that by giving you the actual style vocabulary as a form: five dropdowns (draw_style_1 through draw_style_5), 135 styles each (watercolor, pixel art, anime, sketch, and plenty more), each with its own weight slider. It's the material-and-finish half of ComfyUI-PromptWrapper's style pair - Draw Artist Prompt covers movements, this one covers the look and texture of the render.

How it works

Identical machinery to its sibling: it reads from a bundled draw_style_datas.jsonl dataset, in Chinese or English depending on the language input. -- omits a slot, random 🎲 draws via seed + slot_index against the dataset length, and each weight (0–2) wraps its phrase as (style:1.3). Out of the box, slot 1 sits at weight 1 and the rest at 0, so the default behavior is "one clean style", and you add others when you want a blend.

Inputs that matter

• draw_style_1 and draw_style_1_weight - your main style. This is the one you set.
• draw_style_2..5 + weights - blending. A second style at 0.3–0.5 gives you a hybrid look; push two styles to 1.0+ and they fight.
• seed - reproducible random picks.
• preset_prompt, input_prompt (optional) - prepended fixed text.
• enable - bypass switch for the whole style block.

One text output - a comma-joined, weight-wrapped style phrase ready for a CLIP Text Encode, usually wired in front of or behind your subject prompt.

Install

ComfyUI Manager → Custom Nodes Manager → search ComfyUI-PromptWrapper → Install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/clouddreamfly/ComfyUI-PromptWrapper

No downloads - the dataset ships with the pack.

Where people get burned

The weighting caveat is the recurring theme of this pack: (style:1.3) is honored by SDXL-lineage checkpoints and silently ignored by LLM-encoded models, so the blend feature is a CLIP-encoder party trick. On those models you're still getting a useful style tag list, just not the emphasis math. Second, styles are dataset-defined, so you're choosing from what the author curated - if the exact style you want isn't in the 135, you can't type your own into this node (that's what Input Prompt + Combine Prompt are for). And the default language is Chinese; flip it if your model reads English better.