Tex-to-image AIs learn to generate correct compositional images.
Burden of proof on those betting YES.
Resolves to YES if all conditions are met:
Free to use model
Correct images (at least one out of 3 first drawn) for at least 8 out of 10 prompts
Prompts consisting of pairs of English nouns (only those representing objects/living organisms, etc; not concepts, ideas) connected with prepositions ("a"/"the" optional)
Nouns randomly selected from list of 5000 most common English words filtered to nouns (if selects noun not conforming to #3 , select another pair)
Prepositions randomly selected from the list: [on,under,in,behind]
Example: "a table on a vase", "a table in a vase"
Clarifications and discussions in the comments.
As of now web search found e.g. that list of common words: https://github.com/filiph/english_words/blob/master/data/word-freq-top5000.csv (may be filtered by "Part of speech"=="n"