LiveTalk: Real-Time Multimodal Interactive Video Diffusion via Improved On-Policy Distillation

TL;DR

This paper introduces LiveTalk, a real-time multimodal interactive video diffusion system that achieves high-quality, low-latency video generation conditioned on text, image, and audio, enabling seamless human-AI interaction.

Contribution

The paper proposes an improved on-policy distillation method for multimodal video diffusion, significantly reducing inference latency while maintaining visual quality and enhancing multi-turn interaction capabilities.

Findings

Matches the visual quality of larger models with 20x less inference cost

Reduces response latency from 1-2 minutes to real-time

Outperforms state-of-the-art models in multi-turn video coherence

Abstract

Real-time video generation via diffusion is essential for building general-purpose multimodal interactive AI systems. However, the simultaneous denoising of all video frames with bidirectional attention via an iterative process in diffusion models prevents real-time interaction. While existing distillation methods can make the model autoregressive and reduce sampling steps to mitigate this, they focus primarily on text-to-video generation, leaving the human-AI interaction unnatural and less efficient. This paper targets real-time interactive video diffusion conditioned on a multimodal context, including text, image, and audio, to bridge the gap. Given the observation that the leading on-policy distillation approach Self Forcing encounters challenges (visual artifacts like flickering, black frames, and quality degradation) with multimodal conditioning, we investigate an improved distillation recipe with emphasis on the quality of condition inputs as well as the initialization and schedule for the on-policy optimization. On benchmarks for multimodal-conditioned (audio, image, and text) avatar video generation including HDTF, AVSpeech, and CelebV-HQ, our distilled model matches the visual quality of the full-step, bidirectional baselines of similar or larger size with 20x less inference cost and latency. Further, we integrate our model with audio language models and long-form video inference technique Anchor-Heavy Identity Sinks to build LiveTalk, a real-time multimodal interactive avatar system. System-level evaluation on our curated multi-turn interaction benchmark shows LiveTalk outperforms state-of-the-art models (Sora2, Veo3) in multi-turn video coherence and content quality, while reducing response latency from 1 to 2 minutes to real-time generation, enabling seamless human-AI multimodal interaction.