Like a paparazzo or street shooter on a long lens from far away: the subject and background squashed flat, layers of background stacked behind him, a blurred object in the near foreground (leaves, a car's tail, a railing, a window frame) and a subject who has no idea. It feels like watching.

How to prompt: the trigger is tlcandid (not the file name). Start with tlcandid,. The training captions are tag-style: person, clothes, action, place, time of day, and always say what the blurred foreground object is.