There's a misconception that resurfaces every time "automating with AI" comes up: the belief that some single tool does the work. You buy the subscription, wire up two boxes, and revenue follows. It doesn't work that way — not because the tools are bad, but because you're looking at the wrong part of the machine.
n8n, the tool this article is about, does nothing on its own. It's a chassis. A wiring board where an event triggers a chain of actions. The value comes from the engines you plug into it: Stripe processing a payment, Resend delivering an email, Claude rewriting a piece of text, ComfyUI generating an image. n8n is the connective tissue between the things that do the actual work.
That distinction — chassis versus engines — is the lens I use throughout this piece. It's also why the cost estimates you see online are almost always wrong: people look at the orchestrator's monthly bill and ignore the real invoice, which lives inside the engines.
What does n8n actually do?
n8n is a visual, self-hostable workflow orchestrator that connects third-party APIs through nodes. It doesn't process payments, doesn't send emails, doesn't generate images: it listens for events (webhooks, timers, triggers from external APIs) and routes them through a chain of calls to external services. The value isn't the orchestration itself — it's being able to write that chain once and let it run forever without touching it.
Its Community Edition is free, self-hostable on any VPS, with unlimited workflows and executions. It isn't AGPL: it's the Sustainable Use License, a fair-code license that permits any use internal to your own business (including automating your creator work), while making it delicate to resell hosted n8n as a service to third parties. I'll come back to this at the end — it matters if you're planning to turn the skill into an offer.
On the Cloud side, 2026 pricing starts at $24/month for the Starter plan (2,500 successful executions), $60/month for Pro (10,000 executions), $800/month for Business (40,000 + SSO). The historical free tier was retired this year. The count is on successful executions: failed runs and tests don't consume quota. Official pricing at n8n.io/pricing.
Self-hosted, the software is free and you pay for the underlying machine: an entry-level VPS for a solo-founder workload sits between $3 and $20/month. The subscription drops by that margin, but — and I'll write this immediately because it's the part guru posts skip — the cost shifts elsewhere: technical time for setup, updates, backups, secret management. It isn't huge if you know what you're doing, but it isn't zero either. An hour a month, minimum.
From here on, the piece is a walkthrough of real chains running in the YUKIMORI workflow, with the engines each one needs.
Chain one: purchase → PDF delivery
The most instructive pattern I know is also the most robust: someone buys a digital product on Stripe and receives the file (or a signed link) by email without me touching anything.
The mechanism, without going into code: Stripe emits a checkout.session.completed event to a webhook. n8n intercepts it, verifies the signature (so it can't be spoofed by someone pretending to be Stripe), pulls the product ID, generates a signed link to the file on Google Drive with a 72-hour expiry, then calls Resend to send the email. The customer receives what feels like a handwritten note but was written by the chain.
The underlying math is what breaks the "sell on autopilot" rhetoric. On a standard European card, Stripe takes 1.5% + €0.25 per transaction, official pricing here. On a €19 ebook (I'm using a generic figure so it doesn't get mistaken for a real price) the fee is roughly €0.54. On a €39 product, roughly €0.84. If the customer comes from the US with a non-EU card in a currency other than your settlement one, the real fee climbs by 2–2.5% — meaning the same sale to a US customer for an Italian business isn't "1.5% + 25 cents," it's ~3–4%. Selling on subscription? Stripe Billing adds +0.7% on top of the base rate.
For Resend, the free plan gives 3,000 emails a month but with a 100/day cap — and that daily cap is the bottleneck that breaks first, not the monthly total. A hundred emails a day works while sales stay below that pace; a single day of magic links, password resets and onboarding burns through it quickly. Above that, you move to Pro ($20/month and up). Resend's quota details here.
That same skill — configuring a reliable Stripe webhook, handling retries, delivering mail from the right domain without landing in spam — is a service you can offer to a small shop or a fellow creator selling manually. You're not selling n8n: you're selling the pattern.
Chain two: Claude inside the workflow
The case where AI becomes an engine inside n8n is text production. I don't mean "I write the post with AI" — I mean batch reworking, translating, spinning variants of the same piece, all triggered by an event.
A concrete example: I publish a long-form article. An internal webhook fires it into n8n, which calls the Claude API to generate three drafts — a short abstract for a newsletter, a condensed social-format version, a translated version. The three outputs land back in Google Drive in separate folders. I read, edit where needed, hit publish. What was two hours of work becomes twenty minutes of review.
The cost of that intelligence is the most under-explained fact in the landscape. Claude Sonnet 4.6 sits at $3 per million input tokens and $15 per million output tokens. Claude Haiku 4.5, enough for most reworking, sits at $1 / $5. Official pricing here. But the lever nobody mentions is prompt caching: cached reads cost 10% of the base input price — a 90% discount on content that repeats between calls (system prompt, few-shot, fixed context). If your workflow has a static instruction block on every call, you pay for it once and it stays at 10% for a while.
Real order of magnitude: thirty short-form reworkings a day with Sonnet 4.6 and prompt cache active sit well below one euro a month in token cost. It's a figure worth repeating because it breaks the prevailing perception: "AI inside automations" for a solo-founder is a marginal cost compared to the infrastructure underneath, not the other way around.
A note worth mentioning once. From August 2, 2026, content generated with AI must be labelled as such under the EU AI Act: I've written about it in EU AI Act, August 2: what actually changes for creators. If you're automating text or image production, transparency about AI origin isn't optional anymore — it's worth getting compliant before scaling the chain, not after.
Chain three: back-office with Google
The slice of work I call back-office — sorting mail, logging events into a spreadsheet, saving attachments, remembering deadlines — is where n8n gives a solo-founder the highest effort/benefit ratio, because these are all small repetitive tasks that steal five minutes at a time and add up to hours by month's end.
The Gmail, Drive, Calendar and Sheets APIs are free for the volumes of an individual business: what you pay, if you pay, is the Google Workspace account for a custom domain (Business Starter plan, order of magnitude $6–7/user/month) — the API calls themselves aren't billed. The real marginal cost of a flow like "an email from platform X arrives → extract the order number → save it to a spreadsheet with date and amount" is: zero from Google, one execution counted on n8n.
The kind of chain I keep running without thinking about it: a payment-platform email arrives in the inbox, n8n reads it, extracts the fields with Claude (a call worth fractions of a cent), writes a row into a Google Sheet that serves as raw bookkeeping, moves the attachment into a per-year folder. At the end of the month, I open the sheet and the bookkeeping is already there.
Chain four: images — where the honest math shows up
Here comes the most instructive case, because it lines up two opposite paths to the same result. For image and video generation, YUKIMORI has two options on the table: a metered service like Higgsfield or ComfyUI self-hosted on RunPod on-demand GPUs.
Higgsfield is an aggregator of generative models (Sora 2, Veo 3.1, Kling 3.0, WAN 2.6, Seedance, Hailuo, Nano Banana Pro and others). Business model is monthly credits: Starter $15/month annualized (~200 credits), Plus $39/month annualized (~1,000 credits, ~€45 at current FX), Ultra $99/month annualized. On full monthly the figures rise by ~$10–15 per tier. Official pricing here. The useful range for a single creator sits between €35 and €50/month depending on the plan chosen — verify current prices, they move.
The real gotcha with Higgsfield isn't the price, it's that credits don't roll over: they expire at cycle end. A slow month and they're gone. It's a use-it-or-lose-it constraint that penalizes irregular usage patterns — which is most creators, not the exception.
ComfyUI is the counterpoint. Open source, free, no subscription. Runs on your own GPU or on cloud GPUs like RunPod. RunPod bills per second: RTX 4090 at $0.34/hour on Community Cloud, $0.69/hour on Secure Cloud; A100 80GB from $1.39/hour. Pricing here.
Math on the same 30 image generations you might run in a month: if each session occupies the GPU for roughly half an hour, that's ~15 GPU-hours/month. On a Community Cloud RTX 4090 that's about $5/month. Against $39–49/month for Higgsfield Plus. A year of irregular-to-normal use lands between $60 and $100 on RunPod, against $470–590 for annualized Higgsfield Plus. On raw numbers, ComfyUI/RunPod wins outright.
But here's the hidden cost the guru posts pretend not to see: ComfyUI is not "install and generate". Setting up a serious workflow (nodes, base models, LoRAs, ControlNet, IPAdapter, upscalers) takes dozens of hours of learning and iteration to reach stable output. For YUKIMORI that skill is already built (I've worked on it for months), so the math above is realistic. For someone starting from zero, those dozens of hours are an investment — and once you add your hourly rate, Higgsfield becomes rational.
The honest comparison is this: Higgsfield is "flip the switch and produce, zero technical time"; ComfyUI on RunPod is "near-zero GPU cost but requires the technical hand." Neither wins in the abstract. The one that wins is the one that fits your time budget.
Chain five: newsletter with Listmonk
Last chain, and the one that closes the commercial loop. Listmonk is a self-hosted newsletter manager, AGPLv3, single binary, latest version 6.1.0 (March 2026). Repo here. It handles the heavy lifting — multi-queue SMTP, SQL segmentation, templates, webhooks, on-board analytics — without a subscription and without per-subscriber costs.
It pairs well with n8n: n8n orchestrates the event (new Stripe payment → add contact to list X with tag Y → schedule the first sequence email at T+2 days), Listmonk sends the volume. Deliverability, which is the real headache with self-hosted newsletters, gets solved by putting Resend as the SMTP behind Listmonk: you get the curated delivery engine of a dedicated service without having to become an SPF/DKIM/DMARC/domain-warming expert.
The cost of this setup: Listmonk zero, n8n zero (already running for the other chains), Resend inside the range of its usage plan. For a solo-founder with a sensible list under 5,000 subscribers, the newsletter infrastructure costs less than lunch.
What does automating with n8n actually cost?
Here's a realistic estimate for a creator running all of the chains above, the way I do, in disciplined self-hosted mode:
- VPS for n8n + Listmonk: ~$10/month
- Resend: free while under the limits, $20/month at the first Pro step
- Claude API (with prompt cache): under €1/month for solo-founder reworking volumes
- RunPod on-demand GPU (if you generate images in-house): $5–15/month on irregular usage
- Google Workspace APIs: free; Workspace account: optional, $6–7/user/month
- Stripe: variable fee as a percentage of sales, no subscription
Order of magnitude: $15–50/month for the entire automation stack. Compare that with the sum of "all-in-one subscription" fees someone would pay to do the same thing buying each tool whole: the gap is 3× to 10× more expensive.
The caveat I keep loud, because it's the part tutorials avoid: this estimate doesn't count your technical time. Self-hosted pays with time, not with subscription. If your time is worth €40/hour and you spend three days setting everything up, you've spent ~€1,000 of time before you've even started. That cost is real — and usually amortizes fast — but it needs to be said. If you don't have the time or the skills for setup, a flat-fee service is teaching you the same lesson at a different price: you pay in euros instead of hours.
From own use to sellable skill — with a licensing note
Everything so far is work I set up for my business. The next step — and I've thought about it at length — is that this skill transfers.
A small brand, another creator, a digital shop still processing orders by hand: the Stripe → PDF chain I describe above you can wire up for them in an afternoon, if you know what you're doing. Same for Google-driven back-office automation, same for the Listmonk newsletter flow. You're not selling a tool, you're selling setup + knowledge of the pattern + maintenance.
Important licensing note, because it's the kind of detail that stops badly-framed projects before they start: n8n's Sustainable Use License does not authorize you to resell hosted n8n as a service to third parties. You can sell consulting and setup (you configure it on the client's VPS, you maintain it, you update it); you can't sell access to your own shared n8n as SaaS without a commercial agreement with n8n. That distinction is worth clarifying before writing a landing page that promises the wrong thing.
The point
Automating isn't buying a tool. It's picking the right chassis (for me it's self-hosted n8n; for you it might be n8n Cloud if you don't want the technical layer) and plugging in engines that do actual work — Stripe for payments, Resend or Listmonk for email, Claude for intelligence, ComfyUI or Higgsfield for images, Google for the back office. The value isn't in the mechanics: it's in picking a stack that doesn't fleece you.
And the way not to get fleeced is to keep the calculator on the desk. Most of the subscriptions floating around assume you won't do the math. If you do it, the cost map changes.
Figures in this article are current as of July 2026 and should be verified before any spend decisions: API prices move, plans get renamed, tiers shift. The method for reading them is what stays.