Here are my standardised images. Compare them to the historical versions manually gpt-image-1.5 (gpt-image-1-mini, first in series gpt-image-1) until I have time for a more detailed response
“A fruit cart tumbles down some stairs. The fruit cart sign reads ‘Apple-a-day'”
Quality setting: low
Quality setting: medium
Quality setting: high
“realistic photo – a person calms a rearing horse”
Note: I also ran "a person calms a rearing horse" and it was broadly the same. That is with forcing term 'realistic photo' removed.
Quality setting: low
Quality setting: medium
Quality setting: high
