InterCMDM: Block-Causal Diffusion for Autoregressive Human Interaction Generation

2026-07-02

AI Summary

Existing methods for generating two-person interactive motions struggle to simultaneously preserve long-range temporal causality of individual actions and mutual coordination between agents: bidirectional denoising disrupts temporal causality, while autoregressive strategies suffer from coordination degradation due to temporal drift. This work proposes a block-causal latent diffusion framework that employs dual-stream causal diffusion Transformers to model each agent’s motion separately and introduces a unified multi-task attention mask to jointly capture diverse interaction patterns—including synchronization, response, leader-follower dynamics, and independence. By adopting block-level diffusion targets, the method avoids redundant encoding-decoding and enables mask-controlled interaction-type specification during inference. It achieves state-of-the-art performance on the InterHuman and Inter-X datasets, significantly improving text-motion alignment, motion realism, and long-term temporal coherence.

Abstract

Text-conditioned human interaction generation must capture both long-range temporal causality within each individual and tightly coupled coordination between partners. Existing interaction diffusion models typically denoise full sequences using bidirectional attention, which obscures causality and hinders streaming and long-horizon generation. Autoregressive alternatives enforce causality but often suffer from temporal drift, leading to coordination degradation and unstable interaction dynamics over time. We propose InterCMDM, a block-causal latent diffusion framework for autoregressive two-person interaction generation. InterCMDM introduces a Dual-Stream Causal Diffusion Transformer that maintains separate causal streams for each person while modeling inter-person dependencies via unified dual-stream attention with multi-task attention masks. These masks unify interaction modeling within a single attention mechanism and support diverse coordination behaviors, including simultaneous actions, reactive responses, leader-follower dynamics, and independent motion. By training a single model across these mask configurations as a form of data augmentation, InterCMDM enables controllable interaction generation by simply selecting the desired attention mask at inference time. Finally, a block-wise diffusion objective enables stable latent rollout over long sequences without repeated decode-encode cycles. InterCMDM achieves state-of-the-art performance on InterHuman and Inter-X, improving text-motion alignment, realism, and long-horizon continuity.

Problem

• human interaction generation
• temporal causality
• coordination dynamics
• autoregressive generation
• long-horizon continuity

Innovation

• block-causal diffusion
• dual-stream attention
• autoregressive generation
• multi-task attention masks
• latent diffusion