prompt - add weight

Want to emphasize a single word in a prompt without hand-writing (keyword:1.3) and praying the parentheses balance? Prompt - add weight is the honest little node for it: a prompt_keyword text field and a weight slider from -3 to 3. Move the slider, and it outputs either the bare keyword (at exactly 1.0) or the wrapped (keyword:weight) form. That's the whole node, and that's fine.

It's meant to sit upstream of whatever assembles your final prompt - feed its prompt output into a join or directly into your CLIP encode, alongside your other fragments. The slider being a float (not a text field) is the quiet improvement over the rest of this pack's string-obsessed text nodes: you can actually drag it while testing.

How it works

One rule, applied consistently:

• weight == 1.0 → output is just the keyword, no wrapper (adding (keyword:1.0) is noise, so it's skipped)
• anything else → (keyword:weight), with the weight rounded to 3 decimals

Weights below 1 - down to the -3 floor - are supported, which gives you the suppression direction for free: (keyword:0.7) is a perfectly cromulent way to de-emphasize a token on SDXL-lineage models.

Inputs and outputs

• prompt_keyword - the word or phrase to weight
• weight - FLOAT slider, default 1, range -3 to 3, step 0.01
• Output: prompt - STRING, the keyword or its weighted wrapper

Installing it

# ComfyUI Manager: search "ymc suite" and install ymc-node-suite-comfyui
# or:
cd ComfyUI/custom_nodes
git clone https://github.com/YMC-GitHub/ymc-node-suite-comfyui

Restart after installing. Dependencies are four small pure-Python PyPI helpers that self-install; no models, no GPU footprint.

Gotchas

• The (word:weight) syntax is an SD1.5/SDXL-lineage feature. On the LLM-encoded encoders (Flux 2 Klein, Anima, Z-Image, most 2026 models), the wrapper is discarded - you're producing text that gets read as a literal instruction. Check what encoder your checkpoint uses before you spend a session tuning weights on a model that ignores them.
• At exactly 1.0 the wrapper vanishes. That's intentional, but it means the output silently changes form as you drag through 1.0 - if a downstream node is parsing the string, that can surprise you.
• The sibling prompt - add is for adding whole tags to a list; this one is for weighting one keyword. They compose well together, but they're not interchangeable.
• WAS-conflict note as with the rest of the pack: if WAS Node Suite is installed too, Manager may flag shared node names between the packs.