# ReactDiff: Fundamental Multiple Appropriate Facial Reaction Diffusion Model

Source: [ReactDiff: Fundamental Multiple Appropriate Facial Reaction Diffusion Model](<https://www.sota2.com/research/paper/68fa7be11e209131cea8b000>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2025-10-26T05:50:18.755Z

Authors: Luo Cheng, Song Siyang, Yan Siyuan, Yu Zhen, Ge Zongyuan

Published: 2025-10-06T11:30:40.000Z

## Abstract

The automatic generation of diverse and human-like facial reactions in dyadic dialogue remains a critical challenge for human-computer interaction systems. Existing methods fail to model the stochasticity and dynamics inherent in real human reactions. To address this, we propose ReactDiff, a novel temporal diffusion framework for generating diverse facial reactions that are appropriate for responding to any given dialogue context. Our key insight is that plausible human reactions demonstrate smoothness, and coherence over time, and conform to constraints imposed by human facial anatomy. To achieve this, ReactDiff incorporates two vital priors (spatio-temporal facial kinematics) into the diffusion process: i) temporal facial behavioral kinematics and ii) facial action unit dependencies. These two constraints guide the model toward realistic human reaction manifolds, avoiding visually unrealistic jitters, unstable transitions, unnatural expressions, and other artifacts. Extensive experiments on the REACT2024 dataset demonstrate that our approach not only achieves state-of-the-art reaction quality but also excels in diversity and reaction appropriateness.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Dialogue-driven Audio-visual Dialogue Generation on DIADG User Study (evaluation)](<https://www.sota2.com/research/sota/dialogue-driven-audio-visual-dialogue-generation-on-diadg-user-study-evaluation.md>): Dialogue-driven Audio-visual Dialogue Generation; DIADG User Study (evaluation); Visual Score: 4.2.

- [Audio-visual Dialogue Generation on DIADG evaluation (mean of 1000 samples)](<https://www.sota2.com/research/sota/audio-visual-dialogue-generation-on-diadg-evaluation-mean-of-1000-samples.md>): Audio-visual dialogue generation; DIADG evaluation (mean of 1000 samples); FID: 21.36.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2510.04712>)

[PDF](<https://arxiv.org/pdf/2510.04712.pdf>)