Prompt Slide ♾️Mixlab

If you've ever hand-edited a prompt to read (golden hour lighting:1.3) and nudged that number up and down by trial and error, Prompt Slide is that exact workflow turned into an actual UI control. You type one keyword or phrase into prompt_keyword, drag weight between -3 and 3 (default 1), and the node builds the weighted syntax string for you and outputs it as a STRING ready to feed into a CLIPTextEncode node or concatenate with the rest of your prompt.

It's also one of the small handful of node types the README specifically whitelists for Mixlab's Workflow-to-APP export - the same nine-node list that includes IntNumber and FloatSlider. If you're publishing a workflow as a web app and want visitors to be able to nudge how strongly one concept shows up without touching the underlying prompt text, Prompt Slide is the standard way to expose that as a real slider control on the published page.

Now, the part that actually matters and that the node's simplicity can hide from you: weighted-attention syntax like this only means something to certain models. It's a CLIP-encoder-era convention - it works on SD 1.5, SDXL, and SDXL-lineage checkpoints (Illustrious, Pony, NoobAI, and the rest of that family). On the newer wave of models that use a general-purpose LLM as their text encoder instead of CLIP - Z-Image, Flux 2 Klein, Anima, and similar - that same (word:1.3) syntax gets silently discarded by the encoder wrapper. Not an error, not a warning - the weight just does nothing, and the text passes through as if the parentheses and number were never there. If you build a workflow around Prompt Slide and switch checkpoints to one of those newer models, don't be confused when the slider stops visibly changing anything; it's not the node breaking, it's the target checkpoint no longer reading that syntax at all. Check what your checkpoint's text encoder actually is before assuming a weight problem is a Prompt Slide problem.

Within CLIP-based models, the range itself is worth understanding too: weight above 1 emphasizes the keyword more strongly, below 1 de-emphasizes it without removing it, and negative values push in the opposite direction - functionally closer to a negative-prompt-style suppression of that concept than a simple "turn it down." A weight of exactly 1 is a no-op, equivalent to just typing the keyword plain.

Install is the standard Mixlab path. Through ComfyUI Manager: search "comfyui-mixlab-nodes," install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/shadowcz007/comfyui-mixlab-nodes.git
cd comfyui-mixlab-nodes
install.bat

or pip3 install -r requirements.txt in a venv, or the embedded-python pip call on a portable build. No model download needed - this node just builds a text string, it doesn't run any inference of its own, so once the pack loads, Prompt Slide itself can't really break in a way that needs troubleshooting. If your weighted keyword doesn't seem to be doing anything, work through it in this order: confirm the string is actually reaching your CLIPTextEncode node (a broken wire is the most boring and most common cause), then confirm your checkpoint's encoder actually respects attention weights at all, before concluding the node itself is misbehaving.