OmniXRef organizes each requested edit into a reference bank containing a task, an instruction and, when applicable, an image or video.
One model combines text, image and video references for editing
OmniXRef applies multiple reference-guided changes to a source video while seeking to preserve content outside the requested edits.
Independent
Xiaoyan Cong · Tianyu Wang · Hyeonho Jeong · Zhifei Zhang · Yuqian Zhou · Yijun Li · +4 more
Research Digest··2 min read
The authors present a unified video-editing system that accepts variable combinations of text instructions, images and videos, with each reference controlling attributes such as camera movement, motion, content, style or visual effects.
§