I'm a bit shocked that GLM 5.2 is not multimodal. Like, how should I use it? I use images all the time.