Prompt All in One · Positive

On the canvas it's a text box. Off the canvas it's the whole point

Don't judge "Prompt All in One · Positive" by what you see when you add it to a workflow, because what you see is deliberately boring: one multiline input called prompt, one STRING output, and a plain passthrough between them. Wire it up and it returns exactly the text you typed. That's not a bug. This pack is a port of Physton's sd-webui-prompt-all-in-one, one of the most-installed A1111 extensions ever, and its model of the world is that prompt management shouldn't live in a cramped text box - it should live in a floating editor that follows your node around.

Click the Open Prompt Editor button on the node and a draggable, resizable panel appears over the canvas. That panel is the tool. The tag library is split into tagged groups (English and Chinese both ship with the pack) with search and one-click insert. Your prompt is broken into individual tags you can drag to reorder, toggle on and off one at a time, batch enable/disable/delete, blacklist, and wrap or weight-bump. There's a history of your last 100 positive edits, a separate favorites list, and all of it is isolated per ComfyUI user.

Here's the part that makes it practical rather than just fancy: the editor state (which tags are disabled, the ordering, translations) is serialized into the workflow's node properties, but the actual output is still a plain string. So the workflow stays portable - save it, share it, drag it onto another machine and the text survives, even if the other machine never opens the editor.

The one-button trap

You'd think clicking a node on the canvas would aim the editor at it. It doesn't. The floating window stays bound to whichever prompt node's button you pressed last, and you rebind it by clicking Open Prompt Editor on a different node. That's a deliberate design (each positive and negative node is its own editor) but it trips people up constantly. Also: the editor window's position and open-state live in your browser, not in the workflow, so a page reload won't restore which node it was editing.

Wiring it up

Take the STRING output and feed it into the text input of whatever encodes your prompt - a CLIPTextEncode node (the one that feeds positive conditioning) for the SD 1.5/SDXL lineage. Because it's just a string, it drops in anywhere a text box does. Normal generation triggers zero network activity; the pack's translation and AI-prompt features only fire when you explicitly select text and invoke them, which sends that selection to whatever third-party service you configured.

Installing it

ComfyUI Manager, searching for comfyui-prompt-all-in-one, is the easy path - the registry release ships with the React frontend pre-built. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/DaNiuNai/comfyui_prompt_all_in_one.git
cd comfyui_prompt_all_in_one
pip install -r requirements.txt

Then restart ComfyUI. Two real caveats. First, dist/ (the built UI) is gitignored, so a raw clone has no frontend - if "Open Prompt Editor" does nothing, you installed from source and need cd ui && npm ci && npm run build. Second, this pack registers its nodes through the new V3 extension API (comfy_api.latest), with no old-style fallback, so it needs a recent ComfyUI. If the nodes don't appear at all, update ComfyUI first. Runtime deps are requests, PyYAML, openai, and translators - the latter two only matter if you use the translation/AI features.

Fair warning

This pack is brand new - the repo was created the same week I'm writing this, zero stars, no battle testing yet. Expect churn. And one ComfyUI-specific gotcha the A1111 original hid from you: the editor will happily flag <lora:name:0.8> references against your installed LoRAs, but ComfyUI does not load a LoRA just because it's in your text. You still need a LoRA Loader node in the graph. The editor validates your references; it doesn't make them real.