APNext Artist

Naming an artist is one of the highest-leverage things you can do to a prompt's look, and this node turns that into a set of dropdowns instead of you memorising a hundred spellings. Pick from illustrators, concept artists, classical painters, or a "dark" list of the moody surrealist crowd, and it welds those names onto your prompt string. That's the whole job. It's part of dagthomas's big "SDXL Auto Prompter" suite - the same pack that carries the Cinematic, Art, and Character builders - and it's the one you reach for when you want a stylistic fingerprint you can't get from adjectives.

Be honest with yourself about where this works, though. Artist-name steering is a CLIP-encoder trick, and it's strongest on the SDXL lineage - SDXL itself, Illustrious, Pony, NoobAI. On the 2026 LLM-encoded models (Flux 2, Z-Image, Qwen) the name still goes in as text, but the encoder reads your whole prompt as an instruction and famous-painter steering gets hit-or-miss. And a lot of newer base models were deliberately trained away from living-artist names, so a pick that nails the vibe on one checkpoint can do nothing on another.

How it works

Under the hood it's dead simple: your prompt text plus whatever you select from the artist menus, joined by the separator (a comma by default) into one string. No model, no API, no key - pure Python, so it runs anywhere and costs nothing. Each menu also carries a None / Random / Multiple Random option at the top, so you can let the seed roll a random painter for you instead of choosing.

The inputs and outputs that matter

• prompt - your base prompt; the artist names get appended to this.
• The four artist menus - illustrator (59 names - Rackham, Beardsley, the storybook and comics crowd), concept (36 - Mucha, Bierstadt, the concept-art and sci-fi illustrators), painters (41 - Monet, Caravaggio, the fine-art canon), and dark (18 - Kubin, McKean, Delville, the gothic/surreal end).
• seed - drives the Random and Multiple Random picks so you can batch variety.
• attributes (boolean, off by default) - when on, it pulls in the extra descriptor words the pack attaches to each entry, per the "advanced format" in the README.

Two string outputs: prompt (your text with the picks folded in) and random (the randomised variant). Wire either into a CLIP Text Encode.

How to install it

ComfyUI Manager: search comfyui_dagthomas, install, restart. Manual:

cd ComfyUI/custom_nodes && git clone https://github.com/dagthomas/comfyui_dagthomas
cd comfyui_dagthomas && pip install -r requirements.txt

then restart. Heads up: the pack's requirements.txt is heavy - it pulls openai, google-generativeai, anthropic, transformers, decord, scipy and more, because the same repo also ships cloud-LLM and video nodes. This node needs none of that, but the install grabs it all anyway, and decord is the one that most often refuses to build on odd platforms. If pip chokes there, the prompt-builder nodes usually still load once the rest is installed.

Where people get burned

• Wrong-model steering. On an LLM-encoded 2026 model the artist name is just words in a sentence; don't expect the SDXL-style transformation. Use it on the SDXL/Illustrious/Pony family and it sings.
• Stacking too many names. Two or three artists blend; six average into mush. Pick one strong name, or use Multiple Random and treat the misses as free exploration.
• Names that do nothing. If a pick has no visible effect, the base model probably never learned that artist. Fix the seed, toggle the name on and off, and A/B - that's the only reliable way to tell a real steer from a placebo.