Watch the video with chapters & transcript →
AUTOMATION
In short: A schedule trigger fires the workflow daily at 11:00, and an AI agent writes a fresh Sora 2 video prompt each run — checking a Google Sheet of past titles and prompts so it never repeats an idea.
This n8n workflow produces faceless YouTube Shorts on autopilot: it invents a video idea, logs it, sends it to Sora 2, waits for rendering, and downloads the MP4 — every day, without a human in the loop. Ani Björkström, a tech consultant based in Stockholm, shares the full template free to anyone who comments on the video.
An Edit Fields node holds the creative brief as variables, and an AI agent turns that brief into a concrete video prompt. The variables are the general idea (stories about a little cat), a sample prompt, the aspect ratio (9:16 for Shorts), and the number of videos per run — one. The agent's system message sets the ground rules: guidelines, an example of good output, the required JSON notation, a "think" tool for harder reasoning, and access to a Google Sheet of previous ideas. The user message simply drags in those variables, so nothing is hardcoded.
The sheet is the workflow's memory: it stores title, caption, video prompt and aspect ratio for every clip generated, and the agent reads it before writing a new prompt so ideas never repeat. Connecting it requires credentials from console.cloud.google.com: enable the Google Sheets API under APIs and services, then create an OAuth client (web application) to get a confidential client ID and secret for n8n. The workflow then uses an append-row operation, so each new video lands as a new row on sheet one.
Through a separate sub-workflow that does nothing but take the prompt and size and post them to OpenAI — a deliberate split, so the idea-generation half can be reused with other video APIs later. The HTTP node sends a POST to the create-video endpoint from platform.openai.com's documentation, authenticated with a header credential whose value is "Bearer", a space, then the API key. The JSON body carries the model, the prompt, the length (8 seconds) and the size suited to Shorts and TikTok; the size can be a variable rather than hardcoded.
By polling. Sora 2 first returns a video ID with status "queued", so the workflow waits one minute, calls the retrieve-video endpoint with that ID, and routes the result through a switch node.
| Switch outcome | What the workflow does |
|---|---|
| Completed | Calls the content endpoint with a GET request and downloads the MP4 |
| Not completed | Returns to the wait node for another minute, then checks again |
| Fallback (anything else) | Saves the outcome to an error node for logging |
Once downloaded, the video can be saved locally or published straight to social media accounts.
Checking every second wastes computational resources. This wait-then-check pattern is called polling, and one-minute intervals are enough here.
Yes. Set the number-of-videos variable higher and the split-out node separates each result so every video gets its own Google Sheet row and API call.
The schedule trigger is set to daily at 11:00, but the interval can be hourly, weekly or monthly, and you can set the exact minute.
Read the full video transcript, with chapters, on the video page →
Want this working inside your finance team?