Video-Reference Video Generation
Overview
The field studies how to create or edit videos while using reference clips to carry over people, objects, actions, styles, or scene information. Common approaches condition diffusion-transformer generators on reference frames or video tokens, using attention, lightweight adapters, visual anchors, or explicit motion and geometry controls; some separate motion generation from identity transfer or refine selected keyframes. These methods support applications such as personalized avatars, character replacement, video restyling, and reference-guided scene editing, evaluated for reference fidelity, instruction following, visual quality, and temporal consistency. A central challenge is preserving fine details while allowing requested changes, especially with similar-looking subjects, viewpoint shifts, complex motion, and long sequences.
Latest papers
29 papers6 papers in the last four weeks, up 100% on the four weeks before. 0.0% of all new papers.