Remove Weights

If your prompt has ever come back from an LLM prompt helper or a TIPO-style tagger with (masterpiece:1.2) and ((best quality)) welded onto every other tag, this is the node you're looking for. Remove Weights takes a weighted prompt and hands back the plain tags: (cat:1.2) becomes cat. That's the whole job, and it does it cleanly.

Why you'd reach for it

Weighted syntax is an A1111 convention that ComfyUI also understands - (word:1.2) and ((word)) mean "emphasize this" to CLIP-era encoders (SD 1.5, SDXL, and the Danbooru-trained anime family like Illustrious and NoobAI). But a lot of pipeline stages want the unweighted version:

• You're saving captions for a training dataset and don't want (foo:1.2) noise in your .txt files.
• A prompt generator injected weights and you want the raw tag list back so you can edit or reorder it.
• You're feeding a prompt into another pack node that parses tags itself and chokes on the parenthesized syntax.

It's the inverse of the "add weights" workflow - every once in a while you just need the plain words.

How it works

The node splits your text on BREAK first, keeping the separators and their whitespace intact, then splits each segment into tags. The tag splitter is parenthesis-aware, so a comma inside (highres, absurdres) doesn't break the group. Each tag then gets regex-stripped:

• (cat:1.2) → cat
• (cat:1.2:1.3) → cat (the two-number form, for SDXL model/clip weights)
• (cat), ((cat)), [cat], [[cat]] → cat

The parens-and-brackets unwrap is worth knowing: this node doesn't just remove explicit weights, it flattens emphasis syntax entirely. That's usually exactly what you want from a node called Remove Weights, but if you were deliberately using ((word)) emphasis, this removes it.

Inputs and outputs

Just two things, and one of them is on the left edge of the node:

• text - the weighted prompt you want cleaned.
• processed_text - the output, wired into whatever wants plain tags (a CLIP Text Encode, a caption saver, or the next prompt-cleaning node in a chain).

No CLIP input, no model, nothing to configure. It's a pure string transformation.

Installation

This node ships in the ComfyUI-Alchemine-Pack repo, together with its siblings like SDXLAutoBreak, SDXLTokenAnalyzer, and SaveImageWithText. The easiest route is ComfyUI Manager: search for "Alchemine" and install the pack, then restart. Or clone it by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/alchemine/comfyui-alchemine-pack
pip install -r requirements.txt

Then restart ComfyUI. The only hard dependency in the whole pack is python-dotenv - no model downloads, no heavy packages. The .env file the README mentions is optional and only matters for the pack's API-key-hungry nodes; none of the prompt nodes need it.

Common issues

There aren't many failure modes for a string cleaner, but one gotcha: the node runs tag-by-tag on comma-separated tags, so if your prompt uses sentence-like prose rather than tag lists, the results can look a little flattened - (a girl:1.2) standing in (a field:1.1) becomes a girl standing in a field, weights gone, grammar intact. That's the point. If you have BREAK separators, they're preserved, so a weighted prompt with BREAK still comes back with its segment structure. Feed the output into a text-preview node on your first run and you'll see exactly what changed.