The #trump crowd picture looked like a fun test for one of the LLMs that can interrogate images. I've found it to be pretty accurate. It says, " There are hundreds of people in the image."
Do try this at home:
./llava-v1.5-7b-q4.llamafile --cli -ngl 35 --image ./crowd.png --temp 0 -e -p '### User: Approximately how many people do you see?\n### Assistant:' --silent-prompt 2>/dev/null
https://github.com/Mozilla-Ocho/llamafile
/nosanitize