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:

1. Free to use model

2. Correct images (at least one out of 3 first drawn) for at least 8 out of 10 prompts

3. Prompts consisting of pairs of English nouns (only those representing objects/living organisms, etc; not concepts, ideas) connected with prepositions ("a"/"the" optional)

4. Nouns randomly selected from list of 5000 most common English words filtered to nouns (if selects noun not conforming to #3 , select another pair)

5. 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.