心宝❤提示词助手

If you generate images for a living - product shots, e-commerce detail pages, anything where you say the same things over and over - you type the same quality/lighting/style tokens into every single prompt. 心宝❤提示词助手 (XinbaoPromptAssistantNode) is the author's fix for that: a visual prompt library that lives right on the node and inserts text with a click. No more scrolling back through your history for that one lighting phrase you nailed three weeks ago.

It comes from the comfyui-banana-li-linux pack by Li Xinbao (李心宝), an e-commerce designer who shares workflows on Bilibili. The pack's headline act is the 心宝❤Banana Gemini image node; this node is the prompt-management sidekick that makes the paid generation less painful. Every generation costs money, so having your prompt vocabulary organized isn't a luxury - it's how you stop burning credits on typos.

What the node actually is

Honestly, the node itself is almost embarrassingly simple. It takes two inputs - text (the multiline box your fragments assemble into) and an optional prefix_text (prepended to the result) - and outputs a single text STRING. The machinery is a web extension that turns the node into a full prompt manager backed by ComfyUI's local API. Your library of fragments lives in a snippets.toml file inside the pack folder, so it's portable: copy that one file and your whole vocabulary moves to another machine.

The output wires straight into the prompt input of whatever generator you use - 心宝❤Banana, or any other node that takes a STRING prompt.

The interactions are the whole point

• Click a snippet to append it to the prompt (comma handling is automatic).
• Click it again to remove it - so it doubles as a toggle, not just an inserter.
• Shift+click to force-append even if the snippet is already present.
• A red corner badge shows how many times each snippet appears in the current prompt, so you can see at a glance that you've doubled up.
• Hover a long snippet for a full preview - useful when your library entries are multi-line descriptions, not single tags.

Building the library is as simple as clicking the green add button. You get a content field, a category label (you can type a brand-new tag right there), and a choice of ten preset colors so you can color-code your vocabulary - say, red for quality tags, blue for lighting, green for subject matter. The top tab bar filters by category, which is what makes a library of a few hundred fragments usable instead of a wall of text.

Installing it

Same install as the whole pack:

cd ComfyUI/custom_nodes
git clone https://github.com/98624017/comfyui-banana-li-linux

Then restart ComfyUI. This is the Linux build - Windows users want comfyui-banana-li, macOS comfyui-banana-li-mac. In ComfyUI Manager you can search "banana-li-linux" or the pack title. The pack's requirements.txt pulls in the usual suspects (requests, opencv-contrib-python, Pillow, and friends), and Manager will grab them on install. One heads-up: this pack ships precompiled modules rather than readable .py, and a bootstrap may fetch a binary on first load, so the first launch can take a minute.

Where people get burned

Your whole library is snippets.toml. If you delete the pack folder or reinstall, that file goes with it - back it up before you fiddle. The node's UI is Chinese-first (search "心宝" or "提示词" to find it), and while the pack's other nodes are API wrappers that hold keys, this one does nothing but pass text around - it's the one node in the pack you can wire up with zero network and zero risk. Still, it lives inside a pack that ships compiled binaries and phone-home nodes, so the usual "read what a fresh pack does before you trust it" applies to the pack as a whole.

If you keep a stable prompt vocabulary for your work, this is the node that finally gets you out of the text box.