See llms.txt for all machine-readable content.

Back to Templates

Create reviewed product video variants with Seedance, Creatomate and OpenAI

Created by

Created by: Andy || banandyy
Andy

Last update

Last update 8 days ago

Categories

Quick overview

Turn one approved product photo into two or three separate video concepts with exact approved captions. Use OpenAI for motion planning, Seedance 2.5 for image-to-video, and Creatomate for overlays. Verify actual files, review every variant, and download an approved handoff ZIP.

How it works

  1. Reads a unique, explicitly approved Google Sheets Queue row containing a product name, trusted immutable PNG URL, approved photo hash and two or three short fact captions.
  2. Downloads the photo without credentials or redirects, validates its portrait PNG structure and checks its SHA-256 against the human-approved hash before any model or generation request.
  3. Asks OpenAI for motion ideas while enforcing exact caption text and fact IDs. Each variant uses the same approved photo, with a separate image-to-video request rather than unrelated story scenes.
  4. Stores a durable intent before each request, saves returned fal/Creatomate IDs, polls with bounds and resumes one operation per scheduled or manual execution. Ambiguous paid submissions require verified-ID reconciliation.
  5. Creates a separate five-second, 720x1280 Creatomate render for each variant, with the product name, exact fact caption and AI-concept label. Audio is optional.
  6. Downloads each actual MP4, checks supported container/codec metadata, duration, dimensions, optional audio and byte hash, then holds the complete batch for human review.
  7. The authenticated companion verifies and packages the original photo and all videos. An allowlisted reviewer approves the exact batch hash or rejects it. Approved handoff downloads files without scheduling or posting them.

Setup

  1. Import setup.json, the worker and review-and-handoff.json inactive in the same n8n project. Connect Google Sheets credentials to setup, create separate test resources and copy their spreadsheet/Data Table IDs into both configurations.
  2. Set real reviewer emails in the companion allowlist and keep n8nUserAuth with user metadata enabled. Copy identical host, model, limit and estimate settings between worker and review configuration.
  3. Prepare a rights-cleared 9:16 RGB/RGBA PNG on trusted immutable public HTTPS storage. Configure its exact hostname and run the separate read-only photo preflight to inspect the image and obtain its SHA-256.
  4. Add the real photo URL/hash, product name and distinct approved fact captions to Queue. Set variant_count to two or three and complete rights, immutable-storage and generation confirmations only after reviewing the input.
  5. Connect OpenAI, fal Header Auth and Creatomate Header Auth. Enter current cost estimates and a per-batch estimate limit before intentionally enabling generation. Run the credential-free demo first, then a separately budgeted live test before production. Use the exact current version/hash in the authenticated review form. Download and inspect every image/video, check all claims and optional audio, and approve only the exact batch you reviewed.

Requirements

  • Designed for n8n 2.28.6 with built-in nodes, native Data Tables and authenticated Forms. Native workflow tests use synthetic authenticated-user metadata; actual browser login has not been validated. Google Sheets, OpenAI, fal.ai Seedance 2.5 image-to-video and Creatomate accounts are required for live use; API/hosting charges are separate. Source images must be approved immutable public HTTPS PNGs on configured trusted exact hosts, portrait 9:16, RGB/RGBA8, noninterlaced and within the documented limits. The companion and setup files are included in the product package. Only synthetic local/provider-mock tests were run; live services and n8n Cloud are not validated.

Customization

  • Change the product photo, approved captions, audience, motion style, variant count and optional audio in Queue. Configure trusted photo hosts, model, estimates, polling bounds and file limits in both workflows before starting a new job. Overlays can be adapted in the RenderScript builder; changes to schemas, model contracts, aspect ratio, duration or review rules require code changes and fresh tests. Exact facts remain human-controlled and existing jobs reject changed configuration.

Additional info

This is a locally tested beta. No paid generation or posting was performed during package tests. AI can distort a product or invent visual details; identity and factual imagery are not guaranteed. The photo hash identifies bytes downloaded by n8n, not the provider's later independent fetch. Use trusted immutable storage and inspect every final variant. Estimates are not billing caps. MP4 checks inspect metadata/sample ranges and bytes, not decoded visual truth. Save review ZIPs promptly because provider URLs can expire. This workflow does not publish ads, certify claims or promise conversions.