That would probably be better suited with a vision model of some sort or some other model specifically trained on the task at hand. It's not a problem with AI in general but rather using a general AI for a special use case.