Current models are much too keen on aligning things where even amateur photographers would separate the edges. In this case: the distant beach, with the rim of the glass and the posts on the dock.
But because it’s all 2D-to-2D, no network has any incentive to clarify or separate parallel edges. There’s significant pressure to ensure things connect. It doesn’t know which cows are small and which are far away.
Current models are much too keen on aligning things where even amateur photographers would separate the edges. In this case: the distant beach, with the rim of the glass and the posts on the dock.
But because it’s all 2D-to-2D, no network has any incentive to clarify or separate parallel edges. There’s significant pressure to ensure things connect. It doesn’t know which cows are small and which are far away.