VIS-Ground: Video Interactive Storytelling with Contextual Grounding
VIS-Ground turns messy interactive video context into constraints a generator can execute.
The paper frames the problem as contextual grounding: combining the source material, the rendered video history, and the viewer’s new request before continuing generation. VIS-Ground does this through structured context abstraction, constraint induction, and constrained video generation with planning, verification, revision, and rendering. Across three video generation backbones, the authors report the best composite scores, with a 10.3-point average absolute gain over the strongest baselines. They also flag remaining weaknesses in dependency extraction and faithful rendering. HF Daily Papers' note
The paper frames the problem as contextual grounding: combining the source material, the rendered video history, and the viewer’s new request before continuing generation. VIS-Ground does this through structured context abstraction, constraint induction, and constrained video generation with planning, verification, revision, and rendering. Across three video generation backbones, the authors report the best composite scores, with a 10.3-point average absolute gain over the strongest baselines. They also flag remaining weaknesses in dependency extraction and faithful rendering. HF Daily Papers' note
score 4