Embedding Prompt ♾️Mixlab

Pick an embedding from a dropdown, set a weight, and it spits out the little (embedding:name:weight) string you'd otherwise hand-type into your prompt. That's the whole job - it's a convenience node that saves you from remembering the exact syntax and from misspelling the filename. Small, but if you use embeddings a lot, it removes a real papercut.

What an embedding even is

Worth a quick refresher, because the term gets muddled. A textual-inversion embedding is a tiny learned file (10–80 KB) that teaches the text encoder one new "word" - a trigger token that pulls the model toward a specific style, character, or (the classic use) a bundle of quality/negative traits. It never touches the model's weights; it just adds vocabulary. That's why they're so small and why you invoke them by name in the prompt.

The honest context: embeddings are a fading technique. They're locked to the exact text encoder they were trained on, so an SD 1.5 embedding does nothing on a Flux or a 2026 LLM-encoder model - and nothing warns you, it just silently reads the name as plain text. New embeddings have slowed to a trickle, mostly SDXL-lineage anime stuff now. So this node is genuinely useful if you're working in SD 1.5 / SDXL with embeddings you already rely on, and irrelevant on newer bases.

The inputs

• embedding - a dropdown populated from your ComfyUI/models/embeddings folder. If it's empty, you have no embeddings installed - that's the usual reason the list looks blank, not a bug.
• weight - how strongly to apply it, from -2 to 2 (default 1). Above 1 leans harder into the embedding; below 1 softens it; negative flips it. Same idea as weighting any prompt term.

Output is a single STRING - the formatted embedding tag. You concatenate it into your prompt (a text-combine node, or straight into a CLIP encode alongside your other words).

Installing it

Get the pack via ComfyUI Manager (search mixlab, install comfyui-mixlab-nodes, restart), or clone it:

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

then install requirements and restart. The node needs no download of its own - but it's only useful if you've put .pt/.safetensors embedding files in ComfyUI/models/embeddings. Grab those from Civitai or wherever you source them.

Common snags

• The dropdown is empty. No embeddings in models/embeddings. Add files, restart ComfyUI (or refresh the node), and they'll show up.
• The embedding does nothing. Almost always a base-model mismatch - the embedding was trained for a different text encoder than your current checkpoint. An SD 1.5 embedding on an SDXL or Flux model is inert, and you get no error, just no effect. Match the embedding to your checkpoint's family.
• You could just type it. True - (embedding:myEmbed:1.0) in your prompt does the same thing. This node is for people who want it as a wired, selectable input (handy in Mixlab's workflow-to-app setups) rather than hardcoded text.