Nodes/XBHH Live2D Pet & LoRA/XBHH Prompt Formatter
ComfyUI Node

XBHH Prompt Formatter

The prompt janitor that strips bad weights, fixes underscores, and dedupes your tags

By xbhh123456·Created 10 months ago·Updated 23 days ago· 13
XBHH Prompt Formatter
    • formatted_text
    ◄text—►
    ◄to_lowercasetrue►

    The XBHH Prompt Formatter is a cleanup pass for your prompt text. Paste a messy prompt in - the kind that accumulates bad weight syntax, underscores where spaces should be, duplicate tags, and the odd 5::word:: leftover from some generator - and it returns a deduplicated, normalized string you can feed straight into a text encoder. It's the node equivalent of linting your code before you ship it, and it matters more than it sounds because prompt hygiene is a real quality lever when you're pasting tags from multiple sources.

    It ships in the same XBHH kit as the LoRA loaders, and it slots naturally next to them: the Plus loader's triggers output plus a scraped tag list plus a hand-written phrase all get mashed together in one text field, and this node is the thing that makes that mashup presentable.

    What it actually does

    The format_prompt function runs a defined pipeline over the text:

    1. Fixes broken weights - strips number::word:: patterns (like 5::postcard::), which some tag generators and datasets emit, leaving just postcard.
    2. Underscores → spaces - long_hair becomes long hair, which is what most encoders want from a civitai-style tag list.
    3. Splits on commas, but not inside brackets - it tracks parenthesis/brace/bracket depth so compound weights like (cat:1.2), (dog) don't get torn apart at the comma inside the group.
    4. Converts NAI-style weights - the NovelAI brace shorthand {tag}/[tag] gets rewritten to explicit (tag:weight) notation, { being ~1.05^n and [ being its inverse, since ComfyUI's encoders mostly don't understand the brace syntax.
    5. Dedupes - repeated tags collapse to one, keeping the first occurrence and dropping case/whitespace-insensitive duplicates.
    6. Lowercases - the to_lowercase toggle (default true) makes the whole output lowercase, matching the tag-style conventions most anime/SDXL taggers use.

    Output is one formatted_text string: tags joined back with commas.

    The inputs

    • text - the raw prompt, multiline. The only thing you'll really touch.
    • to_lowercase - a two-option dropdown (true/false). Default true; set false if you're writing natural-language prompts where capitalization matters.

    Installing it

    It's a node in the XBHH pack:

    cd ComfyUI/custom_nodes
    git clone https://github.com/xbhh123456/Comfyui-xbhh-main.git
    cd Comfyui-xbhh-main
    pip install -r requirements.txt
    

    Or ComfyUI Manager → search xbhh-lora → restart. Nothing heavy.

    Honest caveats

    This is opinionated cleaning, and its opinions are tag-style opinions. If you write prompts as full sentences, the lowercase toggle and comma-splitting will fight you - leave to_lowercase off and keep your own punctuation. The NAI weight conversion targets the NovelAI shorthand specifically; plain ComfyUI weights like (word:1.2) pass through untouched, and the dedup treats (word:1.2) and (word:1.3) as different tags because their text differs. For the common case - you pasted three sources' tags together and want one clean list - it does exactly what it says. Think of it as the formatter at the front of the pipe: dynamic text or a selector produces the raw string, this tidies it, and the CLIP encoder gets the clean version.

    Categoryxbhh/text

    Inputs (2)

    NameTypeDefaultDescription
    textSTRING—
    to_lowercaseCOMBOtrue2 options: true, false

    Outputs (1)

    NameTypeDescription
    formatted_textSTRING—