FlashReFocus: Video Spotlight
Keep one person sharp and blur everyone else, frame by frame. SAM 3 tracks the people, Depth Anything V2 gives each frame's depth, and the FlashReFocus renderer renders the edited depth.
Ching-Heng Cheng1* · Chia-Ming Lee2,3* ·
Ming-Ching Chang2 ·
Xin Li2 ·
Yu-Lun Liu3 ·
Chih-Chung Hsu1,3†
1National Cheng Kung University · 2State University of New York at Albany · 3National Yang Ming Chiao Tung University · * equal contribution · † corresponding author
1National Cheng Kung University · 2State University of New York at Albany · 3National Yang Ming Chiao Tung University · * equal contribution · † corresponding author
A single focal plane cannot do this when the people around the subject stand at nearly the same
depth: they would stay as sharp as the subject. The spotlight edit moves everyone but the subject away from the
focal plane in the depth map, so only the subject stays sharp. The renderer was trained on single photographs with
one focal plane; this use is zero-shot.
① Upload a short clip
Examples (5 s clips from Pexels)
② Click the person to keep sharp
1 5
0.5 2
0.2 1
③ Result
Upload a clip to start.