Anima Artist Style T8

Artist tags are Anima's superpower. One @artistname token can change the entire look of a generation - the model was trained on Danbooru's tagging with artist fidelity that SDXL anime models can't touch, and the community's standard advice is to find a dozen artists you like and build your workflow around them. Anima Artist Style T8 is how this pack hands you that power: you give it a list of artist names, it builds the weighted style string, and it pulls preview images of each artist's representative work so you can see what you're about to generate before you burn a single sample. There's a whole little niche of "browse Danbooru inside ComfyUI" nodes out there; this is the one wired specifically for Anima's @ artist convention.

How it works. The artist_tags textarea takes comma- or newline-separated names, optionally @-prefixed and with a :weight suffix - wlop:1.1, hiten, @somebody:0.9 is the language it speaks. default_weight (0.1–2.0, step 0.05) applies to any tag you didn't weight by hand. use_artist_prefix toggles the @ marker, and that's the load-bearing detail: in Anima's world, @name says "this is an artist", while a bare name is an IP, character, or style-meta token. The node's build() then emits a proper ComfyUI weight string like (@wlop:1.1), @hiten from STYLE_PROMPT.

The second output is the fun one. PREVIEW_IMAGES is a real IMAGE tensor, and it fetches each selected artist's representative first image from Danbooru through the pack's own same-origin proxy - up to 16 images, pulled concurrently, usually a couple of seconds. Wire it into a PreviewImage node and you get a contact sheet of what you're selecting. The previews are letterboxed to 512×768 with black bars, so don't mistake the bars for broken images. If a name returns zero Danbooru hits, you get a small black placeholder tile - that's the node telling you the tag doesn't resolve, which usually means the artist is too obscure or you typo'd the slug.

The hidden input. There's an optional last_picked field you'll rarely touch by hand. When you use the pack's artist gallery (the 📚 button above the node) and hit "add selected", the front-end writes your latest picks into last_picked. The logic: STYLE_PROMPT always includes everything in artist_tags, but PREVIEW_IMAGES only shows the current session's picks when last_picked is set. So your preview sheet reflects what you just chose, not the whole accumulated history. If you clear last_picked, previews fall back to all of artist_tags.

Wiring it up. STYLE_PROMPT goes into Anima Prompt T8's style box (or straight into a positive CLIPTextEncode), and PREVIEW_IMAGES into PreviewImage. The artist library itself - Danbooru artists, plus copyright/IP, character, style-meta, and the curated mooshieblob gallery - opens from the button above the node. First time you switch tabs it backfills tens of thousands of tags in the background (2–5 seconds), then everything is cached locally in the pack's SQLite file.

Installing. Same pack, same drill as the rest:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/comfyui-anima-t8
cd comfyui-anima-t8
pip install -r requirements.txt

or search comfyui-anima-t8 in ComfyUI Manager. The only extra dependency is requests; the model and checkpoint are separate. One honest caveat: this node needs the internet to fetch previews, and it proxies Danbooru's CDN through your local ComfyUI server. If you're on a network that blocks Danbooru, the tags still work but the preview sheet won't fill. And remember - artist knowledge in Anima is deep but uneven, so an artist with a low Danbooru post count may preview poorly. That's the model's quirk, not the node's.