G-Images

The second model in the family. You point a camera at something, say what you want changed, and it changes the frame locally — no service in the middle, no upload. It answers to the voice assistant, which is where the request usually comes from.

Like the rest of the family it is trained from nothing rather than adapted from an existing model, which puts a hard ceiling on how good it can be at this size and makes the interesting question how much can be had from seventy million parameters.