Prompt Quill Sailing

This is the fun one. Prompt Quill Sailing isn't a prompt rewriter - it's a prompt explorer. Instead of improving what you typed, it drifts through the Prompt Quill vector store the way you'd wander a gallery: each run picks up where the last one left off and fetches a prompt as far from it as possible, so consecutive generations float off in unexpected directions. The author calls it "sailing the vast ocean of data," and honestly that's the right metaphor. You'll get prompts you would never, ever type yourself. That's the entire point.

It's part of the same ComfyUI_Prompt-Quill pack and, like its siblings, it's a thin client for a separate local server - an LLM plus a 3.2-million-prompt vector store. The node itself is one file that POSTs JSON to /get_next_prompt. If you only install one node from this pack, make it this one.

How it works

Every call sends your seed prompt plus a pile of flags to the server's /get_next_prompt endpoint. The server keeps a "journey" history on its side, picks the furthest-away related prompt from the store, and returns a new prompt and negative. The node has IS_CHANGED hardcoded to NaN, so it never caches - every queue fires a fresh sail. That's deliberate: each run is a new step in the journey.

The inputs that matter

There are a lot, but a beginner sets four and leaves the rest alone:

• prompt - your starting point. The whole trip radiates from here.
• reset_journey - the important one. Set it to true for exactly one image, then back to false, to restart the journey from your current prompt. The README stresses this because ComfyUI is stateless: the history lives on the server, and flipping this is the only way to tell it "new voyage."
• distance - how far each jump reaches into the store (1–10000, default 20). Bigger = weirder, faster drift.
• use_all_trip - keeps the generated prompts forced to stay around your seed prompt, no matter how far you sail. On if you want a themed trip; off for true exploration.

Then there's the "add" family, each a toggle plus a text box: add_style/style (e.g. "cyberpunk"), add_search/search (bias toward a topic), add_negative/negative, rephrase/rephrase_prompt (have the LLM reword the result your way), plus summary and unload_llm. Flip one on, type something, done.

Outputs are two strings: Prompt and NegativePrompt. Feed them into a CLIP Text Encode, or pick the Conditioning variant to skip that step.

Install

ComfyUI Manager → "Install via Git URL" → https://github.com/osi1880vr/prompt_quill_comfyui, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/osi1880vr/prompt_quill_comfyui

No pip deps, no model downloads - the pack only needs requests. Then run the Prompt Quill server from osi1880vr/prompt_quill on a reachable host; it's the part that actually consumes RAM/VRAM (people run server + ComfyUI on one GPU at ~16GB).

Troubleshooting

• It keeps sailing from the old journey. You forgot the reset_journey dance - true for one run, back to false. It's easy to miss and confusing until you know it.
• "Something went wrong, did you install Prompt Quill?" - the node's error string for a failed HTTP call. Server down, wrong URL, unreachable host.
• Your negative text shows up even with add_negative off. This node compares the boolean to the string 'false', which is never equal, so your negative text is always appended. Leave it empty unless you want it.
• Every queue is different. Because it never caches, re-running the same workflow gives a new prompt each time. That's the feature, but it makes reproducibility awkward - if you want a fixed result, generate once and pin the output string.
• Manage expectations on longevity. The project went quiet in 2025, so the store and server are roughly as shipped. It still sails fine - it just isn't growing.