Rebalancing Reference Frame Dominance to Improve Motion in Image-to-Video Models

Wooseok Jeon1*
Seungho Park1*
Seunghyun Shin2
Sangeyl Lee1
Hyeonho Jeong3
Hae-Gon Jeon1†
1Yonsei University 2GIST 3Adobe Research
*Equal contribution    †Corresponding author
arXiv Code

Dynamic — Qualitative Comparison

For each example we compare Base, ALG, and Ours on the same image condition. Each sub-section corresponds to a figure in the paper.

Continuous Control

Drag the slider to vary the guidance setting g. The video restarts from the beginning each time you change the value, so every clip plays in full. Each sub-section corresponds to a figure in the paper.