first-class support for dedicated vision provider/model in multimodal flows