Some links in this article are affiliate links. Automated Wealth Lab may earn a commission at no extra cost to you. See our Affiliate Disclosure.
- The twelve assets: what “done” means
- The pipeline at a glance
- Stage 1: Ingest and transcription
- Stage 2: The LLM drafting nodes — one prompt per asset
- Stage 3: The human review gate (non-negotiable)
- Stage 4: Clip extraction
- Stage 5: The narrated recap
- Stage 6: Distribution — drafts in, nothing auto-published
- What one recording actually costs
- Variable cost per recording
- Fully loaded: amortizing the subscriptions
- DIY-API route vs all-in-one tools
- Where this pipeline breaks
- FAQ
- Methodology
- AWL verdict
- Sources
Disclosure: This article contains affiliate links to n8n, Kit and ElevenLabs. Automated Wealth Lab earns a commission if you sign up through them, at no extra cost to you. It never changes what we recommend.
Most solo creators do not have an ideas problem. They have a multiplication problem. One good 40-minute recording contains a blog post, a newsletter, a week of social posts and several short clips — but extracting that by hand takes longer than recording it, so most of it never happens.
The fix is not another all-in-one app. It is a pipeline: software does the transcription, first drafts and file-shuffling; a human does the one job that protects quality — approval. This build-along recipe ends in a modeled cost per recording and per asset, from published prices rather than promised savings.
The twelve assets: what “done” means
Define the output first. From one 30–45 minute recording, this pipeline produces exactly twelve assets:
| # | Asset | Format |
|---|---|---|
| 1 | Full transcript, cleaned and timestamped | Text / SRT |
| 2 | Show notes with chapter summary | Text |
| 3 | Long-form blog post | ~1,200–1,500 words |
| 4 | Newsletter issue | Email-ready draft |
| 5–7 | Three short clips, with burned-in captions | Vertical video |
| 8–10 | Three quote / social posts | LinkedIn, X, short-form caption |
| 11 | Narrated recap (audiogram / TTS recap) | ~3-minute audio |
| 12 | FAQ / SEO summary | Structured Q&A text |
Twelve is a useful ceiling: enough to fill a week of channels, few enough to genuinely review. If you cannot review it, do not generate it. This pipeline sits inside a wider one-person system — see the AI automation stack for a one-person business.
The pipeline at a glance
One n8n workflow, six stages, one hard stop in the middle:
- Ingest — a form, folder watch or webhook receives and stores the file.
- Transcribe — an HTTP node sends the audio to a transcription API; the timestamped transcript is saved.
- Draft — language-model nodes turn that transcript into the written assets.
- Review gate — the workflow pauses. Nothing is published or rendered final until a human approves it.
- Produce media — approved clip timestamps are cut and captioned; the approved recap script is narrated.
- Distribute drafts — assets are filed; the newsletter draft is created in your email platform, unsent.
In n8n, one run counts as one execution however many nodes it contains — a 30-node run is not billed as 30 tasks. Start with n8n on Cloud, or self-host the free Community Edition.
Stage 1: Ingest and transcription
Nodes: Trigger (form / webhook / Drive watch) → HTTP Request (transcription API) → storage node.
Keep the original file untouched in a dated folder. Every asset traces back to it and to one transcript — if nodes use different transcripts, timestamps drift and clips stop matching the words.
Two sensible OpenAI routes, both priced per minute of audio (checked October 2026):
| Model | Price | Use it when |
|---|---|---|
Whisper API (whisper-1) / gpt-4o-transcribe | $0.006 / minute | Mature default; timestamped output for captions and clips |
gpt-4o-mini-transcribe | $0.003 / minute | Cheapest; if its accuracy holds on your recordings |
For the worked example — a 40-minute recording — Whisper costs $0.24; the mini route, $0.12. A transcript full of misheard names costs more in review time than that saves, so keep a short custom vocabulary list for the cleanup prompt. One constraint: transcription APIs cap upload size per request (OpenAI’s is 25 MB), so extract a compressed audio track with ffmpeg first.
Stage 2: The LLM drafting nodes — one prompt per asset
Nodes: A sequence of OpenAI (or compatible) nodes, each fed the cleaned transcript — or the structured summary from the first node — plus one narrow instruction.
The expensive mistake is sending the full transcript separately into ten prompts. Use two tiers instead:
Tier 1 — one structuring pass. One call returns: chapter list, strongest claims, three quotable lines with timestamps, and questions a listener would ask. The timestamps become the Stage 4 clip candidates; the questions become Asset 12.
Tier 2 — drafting passes. Each node works from that summary, plus only the transcript section it needs: the blog node gets the full text; social nodes get the quotable lines; the FAQ node gets questions and excerpts.
| Step | Suggested model | Why |
|---|---|---|
| Cleanup + structuring | GPT-4o mini | Pattern work; $0.15 per 1M input / $0.60 per 1M output |
| Show notes, newsletter, social, FAQ | GPT-4o mini | Short, templated outputs |
| Blog draft | GPT-4o mini; optionally one larger-model polish pass | The blog carries your voice — the one step that may justify a pricier pass |
| Clip selection | No separate model | Reuse Tier 1 timestamps; do not pay twice for the same judgment |
GPT-4o is listed at $2.50 per 1M input and $10.00 per 1M output tokens — roughly seventeen times the mini price — so it belongs, if anywhere, in one polish pass. Every prompt carries the same guardrails: use only what is in the transcript, invent nothing, flag anything unverifiable. A pipeline that embellishes is a liability multiplier.
Stage 3: The human review gate (non-negotiable)
Nodes: Wait / approval node → notification (email, Slack or Telegram) linking the drafts → branch that continues only on approval.
This is the stage vendors leave out of demos, and the one that decides whether the other assets help or harm you. AI drafts; a human approves. The gate is a single review document — show notes, blog, newsletter, social posts, FAQ, clip timestamps — read against one question: would I publish this under my name, unedited, today?
Budget ten to twenty minutes of review per recording — less than drafting from scratch, but not zero; any costing that omits it hides the largest cost. Two checks matter most: direct quotes (compare word-for-word with the transcript; models paraphrase while keeping quotation marks) and numbers (anything not in the recording should not be in the output).
Stage 4: Clip extraction
Nodes: Execute Command node (ffmpeg, approved timestamps) → caption step → export folder.
Once a human has approved three timestamp ranges, cutting is mechanical. ffmpeg extracts each segment, reframes it to vertical (9:16), and burns in captions from the timestamped transcript — so captions stay identical to the approved text instead of being re-transcribed with new errors. No per-clip fee.
The alternative is an API clipping tool that also selects the moments — a legitimate trade, but it replaces the Tier 1 timestamp step rather than stacking on it. Do not pay a tool and an LLM to choose the same moments.
Stage 5: The narrated recap
Nodes: HTTP Request to ElevenLabs (approved recap script) → audio saved with the other assets → optional audiogram.
Asset 11 serves the audience that will never watch 40 minutes: a three-minute narrated recap, narrated only after the review gate. Never send an unreviewed script to a voice API — you pay for audio you then throw away.
ElevenLabs pricing is character-based (checked October 2026): Starter is $6/month for 30,000 characters; Creator is $22/month for 121,000 characters — roughly one credit per character on standard models, ~1,000 characters per minute of speech. Effective rates: ~$0.20 per 1,000 characters on Starter, ~$0.18 on Creator, so a ~2,800-character recap script costs about $0.50 modeled. Plans: ElevenLabs — choose on monthly character total, not plan name.
Stage 6: Distribution — drafts in, nothing auto-published
Nodes: Kit node or HTTP Request → create newsletter draft → file all approved assets → final notification.
The newsletter (Asset 4) is created as a draft in Kit for a final read and manual send. Kit’s Newsletter plan is free for up to 10,000 subscribers; Creator starts at $39/month billed monthly ($33/month billed annually) for up to 1,000 subscribers, rising with list size (checked October 2026). Platform choice is analyzed in email marketing for solo founders: Kit, beehiiv and Substack compared; here, Kit simply receives a draft. Every publish step stays manual — see also our faceless content stack cost breakdown.
What one recording actually costs
All figures are modeled from published prices, for one 40-minute recording (~6,000 spoken words). All LLM passes together are modeled at ~45,000 input and ~12,000 output tokens on GPT-4o mini, because Tier 2 nodes reuse the structured summary instead of re-reading the full transcript.
Variable cost per recording
| Line item | Assumption | Modeled cost |
|---|---|---|
| Transcription | 40 min × $0.006/min (Whisper) | $0.24 |
| LLM drafting, all written assets | ~45k in + ~12k out, GPT-4o mini rates | ~$0.01–$0.02 |
| Optional blog polish pass | ~8k in + ~1.8k out, GPT-4o rates | ~$0.04 |
| Recap narration | ~2,800 chars at Creator’s ~$0.18 / 1,000 chars | ~$0.50 |
| Clip cutting + captions | Local ffmpeg | $0.00 |
| Total variable | Per recording / per asset (÷12) | ~$0.80 / ~$0.07 |
Fully loaded: amortizing the subscriptions
Assuming a weekly cadence — four recordings a month:
| Subscription | Published price (Oct 2026) | Per recording (÷4) |
|---|---|---|
| n8n Cloud Starter | €20/month billed annually, 2,500 executions | €5.00 |
| ElevenLabs Creator | $22/month, 121,000 characters | $5.50 |
| Kit Creator (optional; free plan exists) | $39/month, 1,000 subscribers | $9.75 |
| Fully loaded | With Kit Creator | ~$20.75/recording, ~$1.73/asset |
| Lean variant | Self-hosted n8n (~$5–$10/month VPS, directional), Kit free, ElevenLabs Starter | ~$3.00/recording, ~$0.25/asset |
Two cautions: self-hosting is not free labor (updates, backups, uptime are yours), and Kit’s price is list-size dependent.
DIY-API route vs all-in-one tools
| Route | What you pay (Oct 2026) | What you get | What you give up |
|---|---|---|---|
| This pipeline | ~$0.80 variable/recording + subscriptions above | All twelve assets, your prompts, your gate, files you own | Build time; you maintain it |
| Descript | Hobbyist $16/month billed annually ($24 monthly), 10 media hours; Creator $24/month annually ($35 monthly), 30 media hours | Excellent transcript-based editing and clips in one editor | Written-asset chain stays largely manual |
| Opus Clip | Starter $15/month, 150 minutes; Pro $29/month or $14.50/month billed annually, 300 minutes (1 credit = 1 minute of source) | Fast, polished clips with virality scoring | Clips only — one of twelve asset types |
Not “DIY is cheaper,” full stop: Descript is better if your bottleneck is editing itself; Opus Clip, if clips are your only output. DIY wins when you want all twelve assets, weekly, under one gate. (All-in-one rows are subscription figures with usage caps; DIY separates variable from subscription cost.)
Where this pipeline breaks
It amplifies errors. A misheard name appears, faithfully, in the blog, newsletter and captions — hence cleanup and the gate before the fan-out.
It produces plausible filler from weak inputs. A rambling recording becomes a rambling blog post with better formatting. It multiplies a good recording’s value; it creates none.
It drifts without a style reference. Give every node the same style brief plus examples of your best pieces, or each asset sounds generic.
FAQ
How long does it take to build this pipeline?
A focused half-day to a full day if you know n8n; longer if chunking and ffmpeg are new.
Do I need n8n Cloud, or can I self-host?
Either; the workflow is identical. Cloud Starter (€20/month billed annually) removes server maintenance; self-hosting the free Community Edition on a small VPS (directionally $5–$10/month) removes the meter.
Which transcription model should I start with?
Whisper at $0.006/minute for timestamped, caption-ready output; gpt-4o-mini-transcribe at $0.003/minute if its accuracy holds on your recordings. Test both on five minutes of real audio.
Can the pipeline publish everything automatically?
Technically, most steps could. This design deliberately stops at drafts — the gate is a quality and reputation control, not a missing feature.
Why is the ElevenLabs recap only about three minutes?
Character cost scales linearly; attention does not. A ~2,800-character recap costs about $0.50 at Creator’s rate; narrating full show notes costs several times more for an asset fewer people finish.
What does it cost to run weekly for a month?
Modeled: about $3.20 variable across four recordings, plus subscriptions — roughly $83/month fully loaded, roughly $12/month lean. Review time is additional.
Methodology
What we did: Documented a build recipe from published documentation and pricing pages, and modeled its cost for one 40-minute recording. Prices were checked in October 2026 against vendor pages or contemporaneous reporting of them (Sources). Assumptions are stated inline.
What we did NOT do: We did not run this pipeline live, measure an invoice, or benchmark accuracy or clip quality. Every cost is a model from published prices, not a measured result; no first-person testing is claimed.
Conflict disclosure: n8n, Kit and ElevenLabs are affiliate partners for this article. Descript and Opus Clip are not, and were assessed on the same fit criteria.
AWL verdict
For a solo creator recording weekly, the bottleneck was never ideas or drafting — it is the unglamorous chain of transcription, reformatting, cutting and filing, and that chain now costs about $0.80 in variable spend per recording to automate. Build it in n8n, draft with a small model, and spend the saved attention where quality lives: one polish pass on the blog post, and a review gate nothing bypasses. Choose Descript or Opus Clip only if your bottleneck is narrower — editing itself, or clips alone.
Sources
- OpenAI transcription pricing ($0.006 / $0.003 per min, verified October 2026): https://brasstranscripts.com/blog/openai-whisper-api-pricing-2025-self-hosted-vs-managed and https://costgoat.com/pricing/openai-transcription
- GPT-4o mini / GPT-4o pricing, from OpenAI’s official pricing page (verified October 2026): https://www.economize.cloud/resources/open-ai/pricing/gpt-4o-mini/ and https://anotherwrapper.com/tools/llm-pricing/gpt-4o-2024-08-06/gpt-4o-mini-2024-07-18
- ElevenLabs pricing (verified October 2026): https://toolchase.com/blog/elevenlabs-pricing-2026/ and https://apicostcalc.com/elevenlabs.html
- n8n pricing, checked on n8n.io/pricing 15 October 2026: https://dev.to/entechsolutions/n8n-pricing-in-2026-what-each-plan-costs-and-what-you-actually-get-2amg
- n8n self-hosting cost basis (verified October 2026): https://n8nhost.io/n8n-hosting/
- Kit pricing (verified October 2026): https://www.sequenzy.com/pricing/convertkit
- Descript pricing (verified October 2026): https://www.descript.com/pricing
- Opus Clip pricing, checked against opus.pro/pricing, October 2026: http://quso.ai/blog/opus-clip-pricing
