Vision support detection is added in version 1.54.0, automatically recognizing when a model can process images. When vision is available, prompts can include image references without extra configuration.