Image support for non-vision models

It would be awesome for a model, such as Kimi K2, that dose not have vision, to have another model, such as Gemini 2.5/3 flash to describe the image to the non-vision model, consuming an extra standard credit