Detect all text within the image (excluding tab text in the case of browser screens). Merge the bounding boxes for each paragraph.
Return JSON with this schema:
{
  [
    {"bbox": [x_min, y_min, x_max, y_max], "label": "(detacted text)"}
  ]
}
If a line break is required, use an escaped line break code (LF) in accordance with the JSON specification. Do not output the literal characters '\n'.
