I wired a no-LLM captions API into n8n — here's the free template

Every social post I write used to start the same way: stare at a blank caption box, burn five minutes, write something mid. Multiply that by every post, every platform, and it's a real tax on anyone who publishes consistently.

I tried the obvious fix — an LLM in the loop. It worked, but every caption cost me tokens, latency, and a bill that scaled with volume. For something as structured as a social caption (hook, body, CTA, hashtags), that felt wrong. So I built a small API that generates captions deterministically — template-and-variant engine, no LLM anywhere near it — and then I wired it into n8n so the whole thing runs hands-off.

The n8n template is free and MIT-licensed. Here's what it does and how to use it.

What the workflow does

Four functional nodes:

Form trigger — you fill in topic, audience, tone (warm / bold / professional / playful), platform, and how many captions you want (1–10).
HTTP Request — calls POST /v1/caption-pack on the Caption-Pack API with your API key.
Code node — formats the returned pack into readable text and surfaces the credit balance.
Form ending — shows your captions on a completion page. Copy, post, done.

Each run costs exactly 1 credit, no matter how many captions you request. Failed calls never cost credits.

Try it in five minutes

Get a free key — 25 calls, no card: open https://caption-pack-api.onrender.com/v1/free-trial in your browser. The key is shown once, so save it.
Import the template into n8n: grab caption-pack-n8n-template.json from the repo below, then Workflows → ⋯ → Import from File.
Create a Header Auth credential: Name Authorization, Value Bearer plus your key. Select it in the Call Caption-Pack API node.
Hit Test workflow, fill the form, submit. Captions land on the completion page.

Check your balance anytime with GET /v1/balance and the same header.

Make it yours

This is where it gets fun. Swap the form trigger for:

A Schedule trigger + Google Sheets — a daily caption pipeline that appends ideas to a sheet while you sleep.
A Webhook trigger — call it from your own app or AI agent with {topic, audience, tone, platform, count}.

I built it for my own posting workflow first; the template is just the packaged version.

Why no LLM?

Two reasons: cost and determinism. A caption pack is structured output — I know exactly what shape I want back. A template engine gives me that in under 100ms with zero per-call AI bill, and the output never surprises me. If you're generating captions at volume (agencies, schedulers, agents posting daily), the math gets obvious fast.