In this video, I walk you through an AI image agent I built in n8n that honestly can replace a lot of what people use Photoshop for. I’m using Google’s new image generator model Nano Banana (🍌) to do three core things: generate high-quality images from a prompt, edit existing images with instructions (like removing objects), and remix two images together. I demo everything live through Telegram so you can see the full loop: I send a prompt or photo, the agent figures out what I want, and it sends the final image right back into the chat. Then I break down the exact n8n architecture: a main “AI image agent” plus a separate prompt-enhancer agent. The main agent uses GPT-5 mini just to route tasks and choose the right tool, and I use GPT-5 in the prompt agent because I want the prompt to be super detailed (based on Google’s Nano Banana prompt best practices). For generation/editing/remixing, I call file.ai via HTTP requests, because it makes model access and API setup clean—and it’s cheap (I show my usage: cents per image). I also show how I handle images via Telegram + Google Drive (upload, search by name, fetch IDs, base64 conversion) so the agent can reliably edit and remix with consistency.