1.
Find the model. Searches the live model list and picks one that matches your request. Name a model if you want a specific one.
2.
Read the parameters. Loads that model's input fields, allowed values and defaults, so the request is valid on the first try.
3.
Upload input files. If the model needs an image, video or audio input from your disk, the agent uploads it to KIE and passes the hosted URL.
4.
Run and wait. Submits the task and polls until it succeeds or fails.
5.
Save the result. Downloads the output to the path you asked for.
Say where to save the result, for example ./output or ./public/hero.png.
Name the model when you care which one is used, for example "with GPT Image 2" or "with Veo 3.1".
Give the key parameters you care about: aspect ratio, resolution, duration, voice. Anything you leave out uses the model's default.
Ask for code when you want integration rather than a one-off file: "Add an endpoint to my API that calls this model on KIE".