Illustrious 📋

First thing to clear up, because the name confuses people: this node is not the Illustrious model. It doesn't load a checkpoint and it doesn't generate anything. It's a prompt list - a curated dropdown of danbooru artist tags and character tags picked to resolve on Illustrious/NoobAI-family checkpoints. You use it with an Illustrious model, to feed it the exact tag strings that model was trained to recognize.

And that's a bigger deal than it sounds. The single most under-documented thing about the whole Illustrious family is that artist tags and baked-in characters are its superpower - the model knows thousands of danbooru artists and characters by name - but nobody writes down what those names are. You're expected to already know that the tag for that one artist you like is maruki_(punchiki), or that Asuka is 1girl, souryuu asuka langley, neon genesis evangelion. WildPromptor's author calls the pack a "keyword GPS," and for this node that's exactly right: it turns "I can't remember the tag" into a scrollable menu.

What it actually gives you

Two big lists, wired into dropdowns. The ArtStyle picker holds ~710 artist/style tags. The Char picker holds ~3,900 character tags, already formatted the danbooru way - 1girl prefix, series in the tag, parentheses escaped as \(...\) so ComfyUI reads them as literal text instead of attention weights. Both are the sort of thing you'd otherwise be copy-pasting out of a booru wiki all day.

Each dropdown has three special options above the actual tags: ❌disabled (contribute nothing), 🎲Random (pick one for you), and 🔢ordered (walk through the list in sequence across a batch). So you can lock a character and roll the dice on the artist, or lock a style and cycle characters - mix control and surprise however you like.

The inputs that matter

• ArtStyle and Char - the two pickers above. Set either to a specific tag, or to Random/ordered, or leave disabled.
• batch_size (1–1000, default 1) - how many prompt strings to emit. This is what makes Random and ordered worth using: crank it to 8 and you get 8 different combinations in one go.
• seed - controls the randomization. Same seed, same picks; change it (or set it to randomize each run) to reroll.
• allow_duplicates (default off) - whether the same combination may repeat inside a batch. Leave it off and each row is unique; note that with a big batch and few enabled options you can run out of unique combos.

The output is a single socket, prompt, and it's a list of strings (one per batch item). Wire it into your positive CLIP Text Encode, or - more usefully - into one of WildPromptor's Prompt Concat / Prompt Builder nodes first so you can glue your quality prefix and scene tags around it before it hits the encoder. Because it's a list, downstream nodes fan out over it and you get a batch of images, one per prompt.

Installing it

Nothing heavy here - no models to download, no API key, no chunky Python deps. Either search ComfyUI-WildPromptor in ComfyUI Manager and install, or drop to a terminal:

cd ComfyUI/custom_nodes
git clone https://github.com/1038lab/ComfyUI-WildPromptor

Restart ComfyUI. The nodes show up under the 🧪AILab/🧿WildPromptor category (this one lives in the 📋Prompts List subgroup). The tag lists live in the pack's data folder as plain text files, so if you want your own artists or characters you just edit the files and the dropdown updates. 1038lab is the same author behind the widely-used ComfyUI-RMBG background-removal pack, so this isn't some abandoned one-off - it's a small, tidy utility from someone who ships.

Where people get tripped up

The tags do nothing on the wrong model. This is the big one. souryuu asuka langley and artist handles are danbooru vocabulary - they only resolve on Illustrious, NoobAI, WAI and other danbooru-trained anime checkpoints. Feed them to Flux, or to a realism SDXL merge that fried its anime knowledge, and you get a generic girl and a shrug. Match the list to the model.

A character comes out generic. Baked-in characters need enough danbooru presence to have been learned - roughly 200+ works is the rule of thumb. Obscure picks land soft, and that's when you reach for a LoRA instead. An aggressive artist tag can also stomp a character's canonical look, so if the face drifts, try dropping the ArtStyle to disabled and add it back once the character's stable.

It's a picker, not a whole prompt. This node doesn't add quality tags or a rating for you. You still bring masterpiece, best quality (and worst quality, low quality in the negative - long negative blocks are cargo on this family) and put the important concepts first, since earlier tokens get stronger attention. The node hands you the hard-to-remember middle; the framing is still on you.

Node doesn't appear after install? You skipped the restart, or dropped the folder somewhere other than custom_nodes. Standard ComfyUI custom-node rules apply.