Hallucinates badly on MacBook Pro M3 Pro (18 GB RAM)

#1
by fenjbfeuren - opened

I gave this model a very naughty image of a naked lady spreading her legs, with that part of the image front and centre, and this prompt: "Analyze this image and provide a concise description (40 - 50 words)". This is what the model gave me: "nude woman posing, smooth skin, toned body, clothed in lace lingerie". So it got 2/4 things right (she's a slightly bigger lady, definitely not "toned").

What gives?? Isn't this model supposed to be purposefully designed for this kinda thing??

  1. I only quantized the original model, I didn't train it. You are complaining to the wrong guy.
  2. You didn't read the documentations, do you?

See also:
https://huggingface.co/spaces/fancyfeast/joy-caption-beta-one

Sign up or log in to comment