Input Video Frames Frame 0s Frame 1.2s Frame 2.3s Frame 3.5s Absolute Timestamp Encoding: <|time_start|> T <|time_end|> Model Input Sequence (Token Stream) |im_start| |vision_start| 0.0 seconds |img_pad| 1.2 seconds |img_pad| 2.3 seconds |img_pad| 3.5 seconds |img_pad| Feature Highlight: • Relative order is not enough. • Absolute time unlocks precise temporal analysis.