ESPHome cameras gain a capture delay and start/end actions for single snapshots. A maker can switch on a light, wait for exposure to settle and then take the image, rather than receiving the frame that was queued before the light came on.
The delay can run for up to three seconds, and frames captured before it ends are discarded. The actions apply to single API and web requests, not continuous streams. On an AI-Thinker ESP32-CAM, a two-second delay produced consistent illuminated images instead of the nearly black first image after idle.