Pošljite SMS: Inducing high energy-latency of large vision-language models with verbose images