Google just released Gemini 3 and with it their new image model Nano Banana Pro. The results are crazy realistic and the price is surprisingly low. So I built a full image generation system in n8n that lets you create images through a simple chat interface. The bot can generate three types of images. You can create an image from text. You can use an existing image as a reference. Or you can upload multiple images and combine them into a brand new one. I walk through how the whole system works: 1. How to connect Nano Banana Pro through Kie.ai using HTTP requests 2. How to build a clean n8n form that collects prompts and file uploads 3. How to enhance prompts with AI before sending them to the model 4. How to use Google Drive as a free image host for the uploaded files 5. How to handle the async generation loop until the image is done 6. How to send the final image back through a Telegram chat The bot can make infographics, edits, memes or realistic image composites. And you can run the whole thing from your phone. If you want more builds like this, make sure you subscribe. https://youtu.be/sJsHQ3J1erI?si=8KcVFwNno7R3gFk0

If you are building AI agents in n8n, there are a few features that can take them from basic to genuinely useful. A lot of people only use the default setup, and they miss out on things that make agents more reliable, more accurate and much more powerful. In this video I break down five features that every agent should be using. Some of them are simple. Others are overlooked because most tutorials skip them. I walk through everything using a real example, which is my budget tracking agent that logs expenses into a spreadsheet through Telegram. Here is what I cover: 1. How to use a system prompt the right way 2. How tool calling works and how to let the agent fill parameters on its own 3. How to use structured output so your agent returns clean and predictable data 4. How simple memory works and when to store chat history 5. How to delegate tasks to a sub agent and avoid long messy prompts These features apply to support agents, CRM update bots, appointment setters and any workflow that relies on AI reasoning. If you want more agent breakdowns and practical builds, make sure you subscribe.

Carousels have taken over LinkedIn and Instagram. They are one of the easiest ways to share tips, tool stacks and educational content in a format people actually stop to read. I built a simple system that creates the full carousel content for you using any YouTube video as the source. You just paste the video URL, add optional instructions and the workflow generates every slide, the title, the bullets, the image ideas and even the caption for the post. This lets you turn long videos into clean, high value carousels without writing everything from scratch. All you do after is drop the content into a Canva template and upload it to LinkedIn as a document. In the video I walk through: 1. How to pull a full video transcript automatically 2. How to run it through AI to create the slide content 3. How to structure titles, bullets and image placeholders 4. How to generate a strong caption for the post 5. How to store everything in a simple data table If you create content or you want to post more consistently, this workflow saves a ton of time and removes the hardest part which is coming up with the actual slides. https://youtu.be/E8BZHZJpvsc?si=sbNVPxVca2xDG9M_

I made a full walkthrough on how to build a RAG based support agent using n8n, Firecrawl, Supabase and OpenAI. Sharing it here because RAG sounds complex, but the setup is actually pretty simple once you see the steps. Here’s the quick breakdown of the workflow: 1. Scrape the website with Firecrawl It grabs every page and formats everything into clean markdown. 2. Upload the files to Drive Just to store all the scraped content in one place. 3. Pull the files into n8n Download each file so n8n can process them. 4. Convert markdown into text Supabase needs plain text before embedding. 5. Create a Supabase vector store Run one SQL snippet and your table is ready. 6. Embed and upload n8n chunks the text, embeds it and sends it to Supabase. 7. Connect a Telegram bot This becomes the front end for talking to the agent. 8. Build the agent The agent pulls relevant info from Supabase and answers based on what’s actually in your data. The video walks through every step if you want to build one yourself. If you want more step by step tutorials like this, please subscribe to my channel. I will be posting more in the upcoming weeks :)

I just built a system that creates full UGC ads using AI, and I wanted to share it here because it can save a lot of money for anyone running product based businesses. UGC ads have taken over the marketing world, but hiring creators or agencies to produce them can get expensive. With the Sora 2 model, you can generate realistic UGC style videos using a single product image and a simple prompt. This workflow lets you upload an image, write what you want the person in the video to say, and then it automatically generates a 10 second UGC ad for you. The system uploads your image to Drive, creates a strong prompt with an agent, sends it to the model through the Kie AI API, checks when the video is ready, downloads it inside the automation, and uploads the final video back into Drive so you can keep all your ads organized. Each ad costs about fifteen cents to generate, which makes this a solid option for small businesses that want to test lots of creatives without paying agencies or content creators for every version. I walk through the full setup in the video below, including how to connect the API, structure the prompts, and automate the entire loop from upload to final video storage. Hope this helps anyone who wants to create UGC ads at scale without spending hundreds on production.