Rabbit Prompt

Rabbit Prompt is a web app for composing prompts visually - think a little instrument where you snap together blocks, connectors, weights and locks instead of typing one long sentence into a text box. This node is how that work reaches ComfyUI. It doesn't generate anything itself; it pulls a prompt that's already been composed for you, server-side on rabbitprompt.com, and hands it to your CLIP Text Encode as plain text. If you've built up a library of projects in Rabbit's tool, this is the missing bridge between "I designed this prompt in a browser" and "I want to run it in my workflow." If you've never touched rabbitprompt.com, this pack is just a storefront you'll need an account for - nothing about it makes sense without the SaaS.

How it works

The node is an API wrapper, but a lightweight one. It calls two endpoints on Rabbit's cloud functions: /apiGetProjects to list your projects, and /apiGenerate to compose a prompt from the project you picked. Your API key is sent as an x-api-key header, and - this is the part the author got right - it's stored in a config.json inside the custom node's own folder, never in your workflow file. The frontend goes further and refuses to serialize the key field into the graph at all, and masks it like a password on screen. Your project list is cached locally too, so the dropdown is a real combo widget instead of a free-text field.

One thing the author did for you: the moment you drop the node on the canvas, the frontend creates a CLIPTextEncode node beside it and connects the output automatically. You don't wire it up, you don't even think about it. A "Generate new prompt" button fetches fresh text into the prompt box, and the same text is mirrored into the connected CLIP node's text widget.

The inputs that matter

• api_key - paste your sk_... key and hit the Activate button. It's remembered after first use, so you can clear the field.
• project - pick from the projects your account returns. Shows "(activate your API key first)" until a key works.
• new_prompt_on_run - on by default, and it's the important one. When on, ComfyUI is forced to re-fetch a fresh prompt on every queue run (the node tells ComfyUI its inputs changed even when they didn't). Turn it off if you want to lock a prompt and stop burning API calls.

The prompt field is both the display and an override - the generated text lands there, and with new_prompt_on_run off, that's exactly what the output returns. The single output is a STRING, named "CLIP Text Encode" - it feeds the text input of a CLIP Text Encode node, which is the standard way to condition a sampler.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/rabbitprompt/ComfyUI-Rabbit-Prompt.git

Then restart ComfyUI. Or just search ComfyUI-Rabbit-Prompt in ComfyUI Manager. There's no requirements.txt beyond what ComfyUI already ships - it uses aiohttp, which core ComfyUI has, so this is genuinely light: no model downloads, no torch versions to fight with.

What goes wrong

• The "No API key" error means you never activated one, or you're on a fresh machine - the key only lives in that node's config.json, so every new install needs the key pasted again (you get one look at it when you generate it, so keep it somewhere).
• Generation needs outbound network to Rabbit's cloud functions. Behind a strict firewall or fully offline, you'll get a "Could not reach rabbit" error, and that's the whole ballgame - unlike a local prompt enhancer, there's no offline mode.
• This is an API wrapper, so the usual rules of that category apply: your prompts (and your key) leave your machine, and there's no local fallback. It's a text-composition node rather than an image-generation one, so you're not shipping reference images around - but the security posture is the same one you should hold for any pack that holds a credential and phones home by design.

It's a young, niche pack - essentially zero community footprint yet - so treat it as vendor tooling: fine if you're already a Rabbit user, skippable if you're not.