Generate Draw Artist Prompt

Styling a character image is where you end up naming art movements you barely remember: Art Nouveau, Ukiyo-e, Baroque. Draw Artist Prompt turns that impulse into a form - five dropdowns (draw_artist_1 through draw_artist_5), each with 127 styles from the pack's dataset, each with its own weight slider. Pick one and it's the whole style; pick several with different weights and you get a blended art direction. It's from ComfyUI-PromptWrapper, and it's the artist-half of a two-node style system, with Draw Style Prompt covering the material/texture side.

How it works

The node reads from a single bundled draw_artist_datas.jsonl dataset (Chinese or English, per the language input). Each slot behaves like every builder in this pack: -- omits it, random 🎲 picks via seed % dataset_length (and note the slots use seed, seed+1, seed+2… so a single seed gives you five different draws), and weights wrap the phrase as (phrase:1.3). The defaults are sensible: slot 1 at weight 1, the rest at 0 - so out of the box you get exactly one artist, and you dial in the others only when you want a mix.

Inputs that matter

• draw_artist_1 and draw_artist_1_weight - the primary style. Weight 1 by default.
• draw_artist_2..5 plus their weights - the blending slots. A low weight like 0.4 on a second artist nudges the style rather than shouting over it.
• seed - reproducibility for random picks.
• preset_prompt and optional input_prompt - prepended text (e.g., your fixed subject line).
• enable - master switch; off, and only the preset/input text is output.

One text output, comma-joined, ready for a CLIP Text Encode.

Install

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

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

Datasets ship with the pack; no downloads.

Where people get burned

Two caveats worth knowing. The dataset entries are movement/style names, not artist names - think "Art Nouveau" rather than "Alphonse Mucha" - so if you specifically want a living artist's name, this won't do it, and honestly that's often a license minefield anyway. And the weighting syntax is CLIP-encoder convention: on SDXL-lineage models the blend works beautifully, but on LLM-encoded 2026 models the weights are discarded and you get a flat list of styles. Also remember the default language is Chinese - switch to English before you end up with a prompt full of 新艺术主义 unless that's your goal.