Nodes/ComfyUI-Anima-Style-Explorer-Prompt/Anima 2B Prompt Template 1
ComfyUI Node

Anima 2B Prompt Template 1

Ten artist slots and a clipboard that fills them

By mizukir0418-gif·Created 3 months ago·Updated 13 days ago· 6
Anima 2B Prompt Template 1
    • text
    artist_1
    artist_2@nakamura takeshi
    weight_20.60
    artist_3@zhibuji loom
    weight_30.60
    artist_4@momoko \(momopoco\)
    weight_40.60
    artist_5@koi han
    weight_50.65
    artist_6
    weight_61.00
    artist_7
    weight_71.00
    artist_8
    weight_81.00
    artist_9
    weight_91.00
    artist_10
    weight_101.00
    upstream_text

    If you've spent any time with Anima, someone has told you the artist tags are overpowered. They're right - "artist tags is pretty overpowered on this model" is basically the community consensus, and style exploration is the setup work that separates "Anima is bad" from "Anima is the best anime model I've used." This node is where you park up to ten of those artist tags at once, with a weight per slot, and get a clean prompt string out the other end. It does nothing clever and nothing magical. That's the appeal.

    What it actually does

    Anima2BPromptTemplate1 is a pure string assembler. No model files, no API calls, no network, no inference. Feed it artist tags, it concatenates them into one line of prompt text. The whole pack is built around this idea - everything else is just organization around prompt assembly for Anima.

    Mechanically it's simple. artist_1 is treated as a raw bypass channel: it's dropped into the prompt as-is, no weight wrapper. Slots 2 through 10 each get wrapped in Danbooru-style weighted parentheses - (@zhibuji loom:0.6) - with the weight formatted to two decimals. Empty slots (or a literal "0") are skipped, the survivors are joined with comma-space, a period is appended, and the result is your text output. If you wire upstream_text in, it gets appended after the artist block.

    The inputs that matter

    Honestly, there are only two things you'll touch: the ten artist_N text fields and the nine weight_N sliders (slots 2–10; slot 1 has no weight).

    • artist_1 - your primary artist, left unweighted. Keep it raw on purpose: it's the one that's guaranteed to appear with full strength.
    • artist_2 … artist_10 - supporting artists, each with a weight from 0 to 2 (step 0.05). Defaults sit around 0.6 for a light influence, 1.0 for a stronger one. For Anima specifically, a light touch reads better than a wall of maxed weights.
    • upstream_text (optional) - a STRING input, typically from an Image Interrogator or a character prompt, appended after the artist block.

    The single text output goes into the positive prompt of a CLIP Text Encode node - or, the intended route, into template1_text on the pack's Anima 2B Prompt Template 2, which adds the quality-tag skeleton around it. The defaults are real Danbooru artist tags (@nakamura takeshi, @koi han, …), so you can hit Run with nothing but the default and see what the node does.

    The clipboard trick

    This is where the pack earns its "Style Explorer" name. Copy an artist ID from an image in the pack's sidebar, and the sidebar's sync button writes it into the first empty artist_N slot on a selected Template 1 node - and if every slot is full, it overwrites artist_1 rather than silently failing. It's a small convenience, but it turns "browse a style you like" into "click, copy, done" with no retyping of @-tags.

    Install

    The core nodes have zero dependencies - the repo's requirements.txt is literally "no heavy third-party deps, runs natively." Clone and restart:

    cd ComfyUI/custom_nodes
    git clone https://github.com/mizukir0418-gif/ComfyUI-Anima-Style-Explorer-Prompt.git
    

    Restart ComfyUI and the node appears under PromptHelper. Or search "Anima Style Explorer" in ComfyUI Manager and let it handle the clone. The only optional heavyweight is the sidebar's visual style database (~1.6 GB, pulled from HuggingFace when you first open the browser) - the node itself needs none of it.

    Where people get burned

    The README is ahead of the code. It documents an Artist Pre-Config Outline node (AnimaArtistPreConfig) and an artist_config input on Template 1 for wrapping artists in category brackets like [Lineart:(@mako:0.8)]. None of that is in the shipped source - the outline wrapping was removed, and only the three core nodes are registered. If you follow the README and go looking for the PreConfig node, you won't find it. It's not you.

    One quirk worth knowing: artist_1 is deliberately unwrapped and unweighted. That's not a bug - the author keeps it as a bypass channel so one artist always lands at full strength. If you want a pure "ten weighted artists" workflow, treat slot 1 as just another tag and set its weight by wrapping it manually.

    CategoryPromptHelper

    Inputs (20)

    NameTypeDefaultDescription
    artist_1STRING
    artist_2STRING@nakamura takeshi
    weight_2FLOAT0.600–2
    artist_3STRING@zhibuji loom
    weight_3FLOAT0.600–2
    artist_4STRING@momoko \(momopoco\)
    weight_4FLOAT0.600–2
    artist_5STRING@koi han
    weight_5FLOAT0.650–2
    artist_6STRING
    weight_6FLOAT1.000–2
    artist_7STRING
    weight_7FLOAT1.000–2
    artist_8STRING
    weight_8FLOAT1.000–2
    artist_9STRING
    weight_9FLOAT1.000–2
    artist_10STRING
    weight_10FLOAT1.000–2
    upstream_textoptSTRING

    Outputs (1)

    NameTypeDescription
    textSTRING