Idea Favor

If you upgrade to the experimental NEXT version of the Vercel AI SDK that does support multimodal Gemini 2.0 Flash EXP, will it support multimodal output, including base64 image generation?