LoadArtistTag

If you run Illustrious, NoobAI, WAI or any of the booru-trained anime checkpoints, artist tags are the single biggest lever you have on style. Drop wlop or ciloranko into your positive prompt and the whole aesthetic shifts, no LoRA needed. The catch is that nobody remembers 1,300 Danbooru artist handles, and half of them are typo-bait like amyu (amm asl pa). LoadArtistTag is the fix: a searchable dropdown of them baked right into ComfyUI, plus a big pile of pre-tuned artist blends. That's the whole node. It's small, and it earns its keep.

Why you'd reach for it

Artist tags are the anime family's real superpower and the biggest single thing that separates Illustrious and NoobAI from Pony, which stripped artist tags out entirely. A name that had a decent presence on Danbooru resolves straight from the tag, so you get that artist's linework and palette for free. The community even shipped a whole web tool for browsing these styles (the Illustrious & NoobAI Style Explorer, which went from 5,000 to 16,000 styles in five days), which tells you how much people care about discovery here. LoadArtistTag is that same idea living inside your graph instead of a browser tab. You don't have to memorize anything or keep a text file of favorites around.

The other half of the value is the mixed list: 435 hand-built artist blends like (torino aqua:1.2),(as109:0.7),(asanagi:1.2).... Blending three to six artists at fractional weights is how you build a "house style" that leans on real training data without being a straight copy of one person. Rolling those by hand is fiddly. Having a curated set to pull from is genuinely nice.

How it works

There's no model, no download, no magic. The node ships two big lists of tag strings and hands one back as text. mode decides which list it reads and whether it picks for you:

• single / mixed - you choose an exact entry from the dropdown.
• random_single / random_mixed / random - the node rolls one for you, driven by seed. random draws from both pools.
• none - returns an empty string, so you can leave the node wired in and effectively switch it off.

The inputs that matter are mode, single_artist (the ~1,349-entry single-artist list), and mixed_artists (the ~435 pre-built blends). seed only does anything in the random modes - it's what makes the roll reproducible. The one output, artist_tags, is a plain STRING. You concatenate it into your prompt: wire it into a text-combine node ahead of your CLIP Text Encode, or paste it beside your other tags. That's it - the artist string joins the rest of your positive prompt like any other tag.

Where the random modes shine: feed random_single a changing seed and batch a dozen generations. Same subject, a different artist style each time. It's a fast way to go style-shopping without babysitting a dropdown between every run.

Installing it

Easiest path is ComfyUI Manager - hit Install Custom Nodes, search ComfyUI-Artist-Selector, install, restart. Manual works too:

cd ComfyUI/custom_nodes
git clone https://github.com/ReBeating/ComfyUI-Artist-Selector

Then restart ComfyUI. No extra Python packages, no model files to fetch - the tag lists are the whole payload, so this one installs clean and cold-starts instantly.

Where people get burned

Wrong model, zero effect. This is a booru-tag tool. On Flux, Z-Image, Anima's LLM-encoded turbo variants, or vanilla base SDXL, an artist name is just meaningless tokens - the vocabulary was never trained in. Use it on Illustrious, NoobAI, WAI, or a booru merge, full stop. If your styles all look identical, that's almost always the reason.

A name in the list isn't a promise the model knows it. The dropdown has 1,300+ entries; your specific checkpoint only actually learned an artist if that artist was well represented on Danbooru - the rough rule of thumb people repeat is around 200+ works. Pick an obscure one and you'll get a generic result, not an error. Try a few and keep the ones that clearly bite.

Don't fight the tags. The blends already carry weights, and stacking them much past ~1.5 gets you oversaturation and artifacts. Quality tags like masterpiece, best quality also have a documented habit of saturating colors, so if a painterly artist blend comes out blown-out or garish, try dropping the quality prefix before you touch the artist weights. And leave the spaces alone - the tags here are correctly written as kedama milk, not kedama_milk; re-adding underscores just makes results worse.