deutrino@Moon was thinking about this after using an 'AI' tool for icon ideas and it had a negative prompt field, do you know how important negative examples are to training art generators? e.g. people in bad poses, poorly-composed photos, inaccurate 'AI' generated images with seven fingers, etc.
does feeding a model bad imagery that's well-tagged with negative terms make its output better?
All GNU social JP content and data are available under the