See llms.txt for all machine-readable content.

Back to Templates

Create reviewed Instagram carousel PNGs with OpenAI and Bannerbear V5

Created by

Created by: Andy || banandyy
Andy

Last update

Last update 8 hours ago

Categories

Quick overview

Create five to seven branded carousel PNGs from approved facts. Review the source-bound text plan before Bannerbear V5 rendering, verify actual ordered image files, and privately archive a complete ZIP with caption and sources. A second approval accepts the entire final carousel; nothing publishes automatically.

How it works

  1. Reads an approved factual brief from Google Sheets, validates unique IDs and supported text limits, and binds it to an actual owned-template snapshot in a native Data Table.
  2. OpenAI proposes the exact requested number of ordered slides, caption and references to supplied fact IDs. Source IDs, coverage, number use, text and conservative line limits are checked before human text-plan approval.
  3. An authenticated allowlisted n8n user can inspect or replace the unrendered plan and must approve its exact version/hash before rendering. Changes after rendering require a new brief ID.
  4. Before each paid render, the worker fetches the actual template and checks its saved configuration hash. It sends the documented nested V5 modifications object, records a separate original render UID for each slide and polls finitely.
  5. Downloads actual PNG bytes, checks CRC/chunk structure and exact1080x1350 dimensions, and stores verified images in a permission-checked private My Drive folder. It downloads them again to verify hashes/order before compression.
  6. Builds one complete ZIP with numbered PNGs, caption, approved plan, source mapping and manifest. The Data Table remains authoritative while Google Sheets mirrors status.
  7. A separate authenticated exact-version review accepts the whole final ZIP after visual inspection. The workflow does not publish to Instagram.

Setup

  1. Import setup, worker, authenticated review and the local credential-free demo. Keep production inactive with drafting/rendering disabled while configuring.
  2. Connect Google Sheets/Drive OAuth2 and Bannerbear V5 Header Auth in setup. It creates your own full layout from the bundled definition, plus fresh queue, Data Table and private My Drive resources. No seller-private template UID is required.
  3. Copy returned resource/template IDs into matching worker/review policies and update the queue template ID. Set actual signed-in reviewer emails, exact permitted folder-user emails, trusted HTTPS output hosts and brand settings.
  4. Connect OpenAI and Bannerbear V5 credentials to worker HTTP nodes, and Google OAuth2 to their respective queue/archive/review nodes. V2 Bannerbear keys are incompatible with V5. Never paste keys into Sheets or forms.
  5. Add a unique approved brief with3–12 confirmed facts and5–7 slides. Deliberately enable drafting, inspect/correct the text plan through the signed-in review form, and approve its exact version before enabling paid rendering.
  6. Run successive worker operations manually first, inspect every final actual PNG and caption, then separately approve the complete ZIP. Activate the minute schedule only after deployment-specific checks. Local mocks do not prove live provider typography or Cloud behavior.

Requirements

  • n8n with native Data Tables, Forms with n8n User authentication, Code, HTTP Request, Compression, Google Sheets and Drive; Google OAuth2 access to Sheets and private My Drive; OpenAI API access; a Bannerbear V5 key with image-template read/write and image read/write permissions, usable account capacity and render credits; approved facts/content rights and human visual review. The complete original text-first layout is reproducible through setup and needs no photo/logo asset. Hosting, tokens, render credits and storage cost extra. No runtime shell, SDK or community node is required.

Customization

  • Set your textual brand, colors, own returned template/resource IDs, actual reviewers, permitted folder users, exact output hosts, OpenAI model and documented finite limits in matching policies. Use a new brief ID for changed registered input or creative changes after rendering. Layout geometry/font/footer changes require updating the bundled definition and validator, then retesting. V1 supports plain Latin text and accents in en/de/es/fr. Automatic publishing and photo/logo ingestion are outside this package.

Additional info

Includes worker, authenticated review, bootstrap and local demo JSONs, a full owned Bannerbear V5 layout definition, examples and tests. Exports are inactive and credential-free with drafting/rendering disabled. The build used no paid model or render call. Strict isolated native mocks test requests, durable state, approvals, actual PNG/ZIP bytes and ordered membership; they do not prove live Bannerbear text-to-pixel correctness, Roboto metrics, OAuth permissions, real reviewer login or n8n Cloud. Local previews are explicitly labeled LOCAL and use Arial. Human plan/final visual review is required. Uncertain mutations stop for original-UID investigation; no Bannerbear idempotency behavior is invented. No social publication occurs.