DanceTogether: Generating Interactive Multi-Person Video without Identity Drifting

This supplementary package for ICLR 2026 Submission #1215 includes additional materials to support the main paper. The contents are organized as follows:

• ClickOnThisIndex.html: The entry point to the interactive supplementary webpage. It features dual-person interactive video examples generated by DanceTog, human and humanoid robot interaction video examples, and a link to the main supplementary document.
• static/: A folder containing all supporting assets required by the supplementary webpage, including all case videos (compressed).
• DanceTogether-code.zip: Contains the core architecture code of DanceTog, including data processing, training, inference, and evaluation code. Due to the large size of the dataset and pre-trained weights, we have not included these two components. The complete code, pre-trained weights, and dataset will be made publicly available upon paper acceptance.