Image Labels Outperform Prompt Tweaks for Vision Models

Adding explicit labels for start and end points significantly improves accuracy, especially in grid-based puzzles. Generic annotations like row/column numbers or grid lines offer little benefit and sometimes degrade performance. The most effective modifications resolve specific visual misreadings and link image content directly to task language.

Summarised from the original article by an open-source model, then checked automatically: every number and English name must appear in the source. The original article is authoritative.