Pick an exposure site, an angle and one or more creative agents. Each agent writes the ad copy and a designed image, renders it with gpt-5-image, then a vision pass proofreads the on-image text. Each image takes about one to two minutes.