high-resolution

Definition

More pixels, and therefore more detail — the usual currency in generation, super-resolution, and dense prediction. It is a claim worth making because of cost: compute and memory grow with pixel count, and attention-based models in particular hit a wall well below the resolutions users expect. The common workaround is cascading — generate small, then upsample with a second model — which is why native high-resolution is stated explicitly when it applies.