Automating Instagram publishing with n8n and AI: posts, carousels, and Reels

A professional Instagram account that posts consistently gains visibility, but producing a caption, a visual, and the right format (post, carousel, Reel) every day takes time — and that's exactly the resource that runs shortest when the business speeds up. n8n can automate all of the repetitive mechanics (drafting, creating the Instagram container, publishing) while still leaving a human to decide, each time, whether the content actually goes live. That division of labor, not prompt sophistication, is what determines whether the pipeline stays credible over time.

Why automate part of your Instagram presence (and where to stop)

Three concrete observations justify automating the draft, not the publish step:

• Format matters more than topic. A 2022 study by Rahman, Mutum, and Ghazali published in the International Journal of E-Services and Mobile Applications analyzed engagement on brand posts on Instagram and found that videos generate significantly higher engagement than static images, followed by images featuring people and then images without people (see the study on Google Scholar). A pipeline that consistently produces Reels or carousels instead of a single static image starts with a structural advantage, before content quality even enters the picture.
• The bottleneck isn't the idea, it's the formatting. Writing a caption, picking hashtags, generating or cropping a visual to the right aspect ratio — these are mechanical tasks an LLM and an image-generation node can handle in seconds, tasks that easily take twenty minutes by hand.
• Fully automatic content shows — and it costs credibility. The Wortel, Vanwesenbeeck, and Tomas study cited above was conducted directly on Instagram ads: it shows that the mere perception of AI-generated content degrades its perceived credibility, independent of its actual quality. Publishing without human review only saves time in the short run.

The goal, then, isn't to "run the account on autopilot," but to hand the AI the repetitive part — caption, hashtags, formatting — while keeping a human as the sole decision-maker on what actually goes live.

What the Instagram API actually exposes

There's no direct "publish" button: Instagram's Content Publishing API (on graph.facebook.com) works in two steps, with a few constraints worth knowing before building the workflow.

• A professional account is required. Only Instagram Business or Creator accounts, linked to a Facebook Page, are eligible. The Meta app used must hold the instagram_content_publish scope, which means passing Meta's App Review for production use beyond test accounts.
• A container first, then a publish call. An initial POST /{ig-user-id}/media call creates a "container" with the visual's image_url or video_url and the caption — Instagram doesn't accept a directly uploaded binary file as input, only a publicly reachable URL. The response returns a container id.
• The container must finish processing before publishing. A GET /{container-id}?fields=status_code call returns IN_PROGRESS, FINISHED, ERROR, or EXPIRED — this matters especially for a video or Reel, whose encoding can take anywhere from a few seconds to a few minutes. Publishing before FINISHED fails.
• The actual publish step is a POST /{ig-user-id}/media_publish call with the container's creation_id.
• A 24-hour posting quota exists, and its exact value has shifted across versions of Meta's documentation — check it at implementation time rather than hardcoding it.

The pipeline in n8n

1. A scheduled trigger and a source of topics

A Schedule Trigger (every morning, for instance) kicks off the workflow from an editorial calendar kept in Airtable or Google Sheets, following the same principle described in our guide on automated LinkedIn posting. Each row carries a topic, a target format (post, carousel, Reel), and a "to process" status.

2. Caption generation via an AI Agent node

An AI Agent node receives the topic along with a system prompt that fixes the brand voice (tone, length, emojis or not, hook structure) and a set of relevant hashtags. Locking the output down with a Structured Output Parser (caption, hashtags, alt_text fields) avoids downstream formatting surprises.

3. Generating and publicly hosting the visual

The OpenAI node (Image resource) or an API like Flux via HTTP Request generates the visual — see our guide on AI image generation in n8n for details on prompting and retouching with the Edit Image node. A step that's easy to overlook: Instagram requires a public URL, not a binary file. The generated visual must first be uploaded to publicly exposed storage (an S3 bucket, a public Supabase Storage bucket, Google Cloud Storage) before it can serve as the container's image_url.

4. Human review before any API call

The draft (caption + visual) is sent to Slack for approval, reusing the human-in-the-loop pattern detailed in our article on human approval with the Wait node and Slack buttons: a Wait node in On Webhook Call mode pauses the workflow until a "Publish" or "Reject" click. This step, not prompt quality, is what absorbs the risk documented by the Wortel et al. study cited above.

5. Creating the container, waiting, then publishing

Three chained HTTP Request nodes:

1) POST https://graph.facebook.com/v21.0/{ig-user-id}/media
   Body: { "image_url": "...", "caption": "{{ $json.caption }} {{ $json.hashtags }}" }
   → returns { "id": "<container_id>" }

2) GET https://graph.facebook.com/v21.0/{container_id}?fields=status_code
   → poll (with a Wait node between each check, the same pattern used
     for any asynchronous API call) until status_code = "FINISHED"

3) POST https://graph.facebook.com/v21.0/{ig-user-id}/media_publish
   Body: { "creation_id": "<container_id>" }

The submit-then-poll pattern is identical to any slow asynchronous API: the retry and timeout best practices apply as-is, particularly to avoid looping indefinitely if a container gets stuck in ERROR.

Variants: carousel and Reel

• Carousel: first create one container per image with media_type: "CAROUSEL_ITEM", then a parent container with media_type: "CAROUSEL" and children: ["id1", "id2", ...] referencing the child containers. The final publish call targets this parent container.
• Reel: replace image_url with video_url and add media_type: "REELS". Container processing time is notably longer than for an image — the polling step should use wider intervals, or the workflow ends up firing unnecessary GET calls.
• Story: media_type: "STORIES" with an image_url or video_url; only the background visual can be published via the API, not the app's native interactive stickers.

Common pitfalls

• Forgetting the public URL requirement: a Google Drive link or a binary file with no accessible URL causes container creation to fail silently or with a vague error — check that the URL actually opens in an incognito browser tab before passing it to the API.
• Publishing before FINISHED: an overly early media_publish call on a Reel still being processed fails; status polling isn't a formality, especially for video.
• Skipping Meta's App Review: beyond the test accounts configured in the Meta app, instagram_content_publish requires the app to pass Meta's review — a common blocker discovered too late, the day before going live.
• Ignoring the 24-hour posting quota: once you're managing more than one or two accounts at a sustained posting rate, a queue on the n8n side (a Supabase table or Data Table listing pending posts) avoids hitting the limit at the worst possible moment.

Going further

The mechanics described here — scheduled trigger, AI generation, structured output, human review before action — are the same ones structuring the triage and drafting workflows in the Inbox AI Pack (€79), applied here to a social channel instead of a mailbox. Start with a single format (the carousel post is usually the easiest to get reliable) before adding Reels, whose asynchronous processing time demands a bit more rigor in the polling step.