Not to mention the uniform tilt (1) isn't necessarily what they have in mind and (2) would likely be incidental on a case-by-case basis rather than having the exact same tilt applying to everything and (3) presumably there are lots of other incidental variations in presentation that are no more or less plausible than tilted buns (e.g. stray shred of lettuce or overextended drizzle of sauce or like 20 other things). You would think any given image would be a motley combination of those things rather than an exact repetition of one.
Like the sibling said about plastic models, if the generated image does faithfully look like the food then I guess it's fine. Over time anyway customers won't even know it's AI generated or a real photo as all the hallmarks of AI images are rapidly disappearing.