This system is designed to automatically turn your long podcast episodes into snappy, short video clips – perfect for social media like TikTok, Instagram Reels, and YouTube Shorts.
The whole system runs on an automation platform called N8N. It uses other AI tools like Visard for the videos, and OpenAI for writing the text.
1. Getting Started with the Video Clipping Service (Visard)
First off, you need to choose and set up a video clipping service that will edit your videos.
We’ve chosen Visard because it’s an affordable option at $29 a month for API usage and allows uploading up to 600 minutes of content. Once you’ve subscribed to Visard, you can find and copy your unique API key from their workspace API settings. This key lets N8N talk to Visard.
2. Setting Up “Scrape and Send” Workflow (Workflow 1)
This first workflow finds new long videos (like your latest YouTube podcast episode) and sends them to Visard for clipping.
In N8N, you’ll use an “RSS Feed Reader” node. Every YouTube channel has an RSS feed (a special link that lists all your new videos). You can get this by using your channel ID in a specific URL format or by using a service like RSS app. You’ll feed this RSS link into the N8N node.
To prevent the system from trying to process too many videos at once, you can add a “limit” (e.g., to process only the two most recent videos simultaneously).
Send Videos to Visard
Connect an “HTTP Request” node (set as a POST request) to the RSS Feed Reader. You’ll paste in the Visard API endpoint for submitting videos. Crucially, you’ll add your Visard API key for authentication.
You’ll also need to specify that the video_type is ‘YouTube’ and feed in the video_URL from the YouTube channel’s RSS feed.
Execute this step to send your first video link to Visard. You should get a “2000” success code and a project ID back. Visard will then start processing, which takes about 5 to 10 minutes.
3. Setting Up “Retrieve and Generate” Workflow (Workflow 2)
This second workflow is the one that waits for Visard to finish, then grabs the clips, makes captions, and organises everything.
Set Up a Webhook to Receive Notifications
Start this workflow in N8N with a “Webhook” node. This acts as a special listening post. You’ll configure this webhook as a POST endpoint.
Crucially, you need to copy the webhook URL that N8N gives you and paste it into your Visard workspace API settings. This tells Visard where to send the clip details once they’re ready. This way, Visard automatically notifies your system when the clips are processed.
Retrieve Project Details from Visard
Connect another “HTTP Request” node (this time as a GET request) to the Webhook node. When Visard sends its notification to your webhook, it includes a project ID. You’ll use this project ID to ask Visard for all the detailed results of the clipping process.
This will give you a list (called a videos array) of all the short clips, including their unique video ID, video URL (where you can download them), transcript (the text spoken in the clip), viral score (how likely it is to go viral), and title.
Process Each Video Clip Individually
Use a “Split Out Items” node to separate the data for each short clip from the list you got from Visard.
Then, add a “Loop Over Items” node. This is super important because it processes each clip’s data one by one, which stops you from running into problems (like getting blocked for sending too many requests too quickly) when talking to other services like Google Sheets or OpenAI.
Generate Captions with OpenAI
Inside the “Loop Over Items” node, add an “OpenAI” node (specifically, “Message a Model”). You’ll connect it using your OpenAI API key. Choose a powerful model like “GPT-4”.
You’ll give it instructions (called a “system message” and “user message”): Tell it to act like a “helpful intelligent social media assistant”. Instruct it to generate “high-quality engaging captions for Instagram and TikTok”. Specify the format you want (e.g., JSON, with a “caption” field).
Give it rules for the captions
“Short and engaging” (around 50-100 words), a “Spartan tone of voice” (meaning direct and strong), “conversational,” and to “use emojis but sparingly”. Crucially, feed it the transcript (the spoken text) of the individual video clip from the Visard data.
Store Everything in a Google Sheet Database
Still inside the loop, connect a “Google Sheets” node (specifically, “Append Row to Sheet”). This is where all your clip data will be stored, and it’s treated like a simple database.
You’ll connect to your Google account. Create a new Google Sheet (you can call it something like “shorts database”) with columns for all the information you want to store, such as video ID, project ID, video URL, title, transcript, viral score, viral reason, clip editor URL, and the generated caption from OpenAI.
Then, you’ll “map” the data from each processed clip (and its generated caption) to the correct columns in your Google Sheet. Add a “Wait” node (e.g., 2 seconds) immediately after the Google Sheets node inside the loop. This helps prevent Google Sheets from getting overwhelmed if you’re adding many rows quickly.
Send an Email Notification
Once the “Loop Over Items” finishes (meaning all clips from a project have been processed and added to your Google Sheet), add a “Gmail” node. Configure it to send an email to the relevant team member or editor, letting them know the clips are ready. Include a shareable link to your Google Sheet database in the email for easy access.
Important Things to Keep in Mind:
- Video Expiry: The video links Visard provides for your clips usually have an expiry date, typically 7 days from when they’re generated. So, you need to use or download them within that timeframe if you want to keep them long-term.
- Pixelated Videos: If your original podcast often includes screen shares, some of the generated clips might appear a bit pixelated. While not part of the basic setup, you could potentially add an advanced step using another AI (like Gemini) to check for blurry images and filter out unsuitable clips.
- API Key Security: Always keep your API keys secure and don’t share them or expose them directly in publicly accessible parts of your workflow.
This complete system automates the grunt work of content repurposing, letting podcasters tap into the massive short-form video network without adding hours to their busy schedule.