how to create a vision model that can originally obey user's...

Prompt

how to create a vision model that can originally obey user's command and use that to look at the picture and return users with good results