TokenDial: Continuous Attribute Control in Text-to-Video via Spatiotemporal Token Offsets
We present TokenDial, a framework for continuous, slider-style attribute control in pretrained text-to-video generation models. While modern generators produce strong holistic vide...