🎬 Create beat-synced AI promo shorts from CSV with Claude Opus and Shotstack
⚡ 0 views · 🎬 Content Creation & Video
💡 Pro Tip — Since TikTok doesn’t have a public API, getting data usually means fragile HTTP scraping. ScraperNode is a community node that gives you clean JSON for TikTok profiles, videos, and comments — drop it into your workflow and skip the parsing.
Description
Quick overview
Turn every row of a CSV into a 32-second vertical promo for TikTok, Reels or Shorts. Claude writes the Shotstack edit: AI images and video shots, AI music, a voiceover and subtitles, all cut to the beat. Self-hosted n8n only: it reads and saves files on disk.
How it works
- Downloads Shotstack’s guide for AI agents once, then reads shorts.csv one row at a time. Each row is one line about a business.
- Claude Opus 5.5 reads the guide and a plain-English brief and writes the whole Shotstack edit: prompts for 8 Nano Banana 2 images, 3 of them animated with Gemini, ElevenLabs music at 120 BPM, a voiceover, subtitles and animated text, all timed to the beat.
- A Code step checks the edit before anything costs money. It fixes small things itself and sends anything else back to Claude with the exact problem.
- One Shotstack render generates every image, video shot and sound, then renders the 32-second short, which is saved to the shorts folder.
- If Shotstack turns the edit down or the render fails, Claude gets the exact error and fixes the edit, up to twice. A wrong API key or an empty balance stops the run with the reason.
Setup
- Install the Shotstack community node from the nodes panel or Settings → Community nodes, then import this workflow.
- Add your Anthropic API key to the Claude and Claude (fixes) steps, and your Shotstack API key to Make the video and Check the render. Free trial: https://shotstack.io/?utm_source=n8n&utm_medium=template&utm_campaign=beat-shorts&utm_content=description
- Save shorts.csv in n8n’s files folder (/home/node/.n8n-files in Docker) with one column, About: what the business is and what the video should say. Make an empty shorts folder next to it.
- On an npm install, change /home/node/.n8n-files to your own files folder in Read the CSV and Save the short. Then click Execute workflow, and try one row first: each short takes 10 to 25 minutes.
Requirements
- Self-hosted n8n, version 2 or later, because the workflow reads and saves files on disk
- The Shotstack community node, any version
- A Shotstack API key with credits for AI generation
- An Anthropic API key. Claude Opus 5.5 works best; other chat models need more fixes
Customization
- Change the length, tempo, number of images or style in the plain-English brief in Claude makes the video
- Ask for fewer video shots to spend less: each is 2.5 credits
- Read the rows from the Google Sheets node and save the shorts to Google Drive
Additional info
About 18 Shotstack credits a short ($3.50 to $3.60 depending on the plan): 7.5 for the 3 video shots, 6 for the 8 images, 3.75 for the music and under 1 for the voiceover and render, plus your Claude usage. Each short takes 10 to 25 minutes. The 10 free trial credits don’t cover a full short, so for a first try change “8 images” to 4 in the brief and delete the line about the video shots, for about 7.5 credits. Write-up with a sample video: https://community.n8n.io/t/321339
🔗 Nodes Used
HTTP Request, Stop and Error, Basic LLM Chain, Anthropic Chat Model, Read/Write Files from Disk, Extract from File
📥 Import
Download workflow.json and import into n8n:
Workflow menu → Import from File