Back to n8n: Building an AI-Powered TikTok Carousel Automation
I recently started re-learning n8n after a year away from the tool. About a year ago, I played around with it but never really dove deep. This time, I decided to actually build something. My first project was an automation to generate product ideas and turn them into carousel sliders for TikTok.
The Workflow Breakdown
The automation starts by reading product information from Airtable, including reference images and specifications. It then calls an LLM (Claude 3.5 Sonnet) to generate ideas, captions, and image prompts. The goal is to convey a short story across five slides.
Once the ideas are generated, the automation handles the image creation through Nano Banana via API calls. The process follows a specific structure:
- Slide 1: Generates the first image containing the character.
- Slide 2: An image showing neither the character nor the product to set the scene.
- Slide 3: Shows the character again. To maintain character consistency, the automation passes a reference of the first image back to the API.
- Slide 4: Creates a slide featuring both the character and the product.
- Slide 5: A final image showing the character in a resolution of the story.
Everything is automatically saved to Google Drive, and the Airtable fields are updated with the new URLs. The final part of the flow uploads the assets to Blotato to generate the actual slider and post it to TikTok.
Learning Through Debugging
Initially, I downloaded a TikTok carousel workflow from a YouTube tutorial. It didn't work because the Blotato node was outdated, and several other elements weren't working out of the box. Looking back, that was actually a good thing.
I had to recreate the entire workflow from scratch to make it work. I ended up changing and improving many parts of the original design. I couldn't pinpoint one specific issue as the "hardest" part, but the debugging process was where the real learning happened.
The "Learning by Doing" Reality
Learning by doing is truly the only way to master these tools. Once you connect enough nodes and see the same errors repeatedly, you develop a gut feeling for how the logic works. There is no magic pill to get good at n8n.
Before I really understood the tool, I thought it was cool but didn't know what to use it for. Now that I am proficient, I see simple and useful use cases everywhere. My advice is simple: if you want to learn something like n8n, just jump straight in. As the saying goes, you can't learn to swim by reading about water.