Ads Video Generator
Making an advertisement video for a product is expensive and takes a lot of time, because a seller needs a model, a photographer, a video editor, and a voice over, only to promote one single product. The aim of this project is to automate all of that, so a seller only needs to fill in the data of their product and the advertisement video is generated by itself.
My role is as an AI Engineer who makes and concepts the whole flow. The orchestration is built in n8n, and every step is handled by the tool that fits it best.
It starts from Google Sheet, which becomes the interface for the user. All the data of the product is set from there, the progress of every row can be monitored from there, and the result comes back to there as well, so the user never needs to touch the automation itself.
From that data, Gemini 3 Flash generates the prompt for the image creation and for the video creation. The prompt for the image is then executed by Gemini 2.5 Flash Image to generate the product image, and after that Kling v2.1 generates the video based on the prompt together with the image that is attached. For the narration, Gemini TTS (2.5 Pro Preview) converts the script into the voice over, then Whisper converts that audio back into text so the subtitle follows the voice accurately.
All of it is composed by FFmpeg, which merges the subtitle with the video, cuts it, crops it, and arranges it into the final advertisement video. Every asset that is used and every result that is produced is stored in Google Drive.
The n8n itself is not the cloud one, it is self hosted, and I am the one who sets up the whole server from scratch using Docker and Nginx. By self hosting it, the workflow can run without the limit of the execution from the cloud plan, the credential and the asset of the product stay inside our own server, and FFmpeg together with Whisper can be installed directly beside the n8n so both of them can be called straight from the workflow.
n8n Workflow
Master Sheet
