✨KonText Extractor

This is the smallest node in the ✨Prompt Assistant pack, and it knows it. ✨KonText Extractor doesn't call any API and doesn't generate anything - it's a dropdown of preset system prompts. If you work with FLUX Kontext, you'll immediately see why it exists: the hard part of that workflow isn't the model, it's wording a good edit instruction. Kontext is BFL's 12B instruction-based editor - feed it an image plus a sentence, get the edit - and the quality of that sentence is the whole game.

What it actually does

The node has exactly one input, kontext, a combo box listing the presets. Pick one and you get the corresponding system_prompt string out. That's the entire job: it's a config front-end that reads the pack's preset library and hands you a ready-made system prompt.

Those presets are the real value. The pack ships around thirty of them, each a carefully worded instruction block for a specific edit. Deep Contextual Fusion (情境深度融合) for grafting a head onto a mismatched body, Seamless Integration for invisible neck blends, Scene Teleportation, Camera Movement, Relighting, Text Removal, Movie Poster, Cartoon Style, Miniature World, Material Replacement - the list reads like a catalogue of every edit people actually ask for. They're English in content despite the Chinese names, and they were contributed by the pack's community, so the quality is the good kind of community quality: battle-tested by people doing the edits.

How you use it

The output is a system prompt, so it slots into anything that consumes one. The natural move is to feed it to an LLM node - PromptExpand's custom_rule_content works perfectly - so the LLM acts as your prompt engineer and produces a Kontext-style edit instruction for your image. Wire that instruction into your Kontext image-editing workflow and you're done.

A light aside: don't feel married to Kontext specifically. The same "describe the edit in detail" paradigm is what Qwen-Image-Edit wants, so these presets carry over to the Apache-2.0 alternative - and the pack's other nodes already ship Kontext-flavored rules for expand and vision workflows, so this preset node is the standalone version of that idea.

Inputs, outputs, install

Just kontext in, system_prompt out. Install is the pack-wide routine:

cd ComfyUI/custom_nodes
git clone https://github.com/yawiii/ComfyUI-Prompt-Assistant.git

or ComfyUI Manager → search Prompt Assistant, then restart.

The one thing that can confuse you: if the dropdown shows only 默认提取预设 (or is empty), the preset config didn't load. The presets ship as a template in the pack and get read from its config - if you've been tinkering with the pack's config directory, that's where to look. Also worth knowing: like every node in this pack, it targets the newer V3 ComfyUI API, so a stale ComfyUI install won't see it. It's a thin utility, but when you're three edits deep into a Kontext session and can't remember how to word a seamless skin-tone merge, it's the exact right tool.