Inference-Time Scaling for Joint Audio-Video Generation

Jaemin Jung; Kyeongha Rho; Inkyu Shin; Joon Son Chung

Home/Archive/arXiv

arXiv · 2606.03183

Inference-Time Scaling for Joint Audio-Video Generation

Jaemin Jung·Kyeongha Rho·Inkyu Shin·Joon Son Chung

View original source ↗Read source document ↗Export metadata ↓

Abstract

Joint audio-video generation aims to synthesize realistic audio-video pairs that are both semantically aligned with text prompts and precisely synchronized. While existing joint audio-video generation models often require substantial training resources to improve fidelity, Inference-Time Scaling (ITS) has recently emerged as a promising training-free alternative in single-modality domains. However, extending ITS from a single modality to multimodal domains is non-trivial, as it requires balancing multiple heterogeneous objectives. In this paper, we present the first comprehensive study of ITS for joint audio-video generation. We first demonstrate that a multi-verifier framework is essential to address the limitations of single-objective guidance, including asymmetric performance trade-offs and verifier hacking. Through systematic analysis, we then identify an optimal multi-verifier combination that yields balanced improvements across all quality dimensions. Finally, to effectively aggregate diverse reward signals, we propose Adaptive Reward Weighting (ARW), a novel test-time optimization algorithm. ARW treats reward aggregation as an online optimization problem, utilizing learnable parameters to calibrate reward variances without requiring prior knowledge of reward distributions, thereby ensuring robust multi-objective selection. Experimental results on VGGSound and JavisBench-mini benchmarks demonstrate that our framework significantly enhances semantic alignment, perceptual quality, and audio-visual synchronization of generated outputs. Synthesized samples and code are available on the project page: https://jung-jaemin.github.io/ITS-AVGen-Proj.

Explore related subjects

cs.MM cs.CV cs.SD eess.AS

Keep this discovery

BibTeX RIS

Jaemin Jung, Kyeongha Rho, Inkyu Shin, Joon Son Chung. 2026-06-02. Inference-Time Scaling for Joint Audio-Video Generation. https://arxiv.org/abs/2606.03183

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Multi-Faceted Evaluation and Mitigation of Emotion Hallucinations in MLLMs

Multimodal large language models (MLLMs) have shown strong potential in open-ended emotion understanding, yet they often generate emotion hallucinations. Evaluating such hallucinations is particularly challenging for two reasons. First, emotion understanding spans multiple cognitive facets, from multimodal perception to psychological reasoning. Second, emotional interpretations are expressed in free-form language, making existing closed-ended protocols insufficient for evaluation. To address these challenges, we introduce EHR (Emotion Hallucination Rate), an evaluator that quantifies emotion hallucinations across six facets: expression, action, audio, instinct, logic, and conclusion. Using EHR, we reveal that existing mitigation methods often reduce hallucinations in some facets while aggravating them in others, exposing the limitation of coarse-grained correction and the need for facet-aware localization and mitigation. Motivated by this finding, we propose HMER (Hallucination-aware Memory-guided Emotion Reasoning), a training-free framework for emotion hallucination mitigation. HMER maintains a Hallucination Memory that records localized hallucinated claims and enables targeted logit rectification, together with an Anchor Memory that preserves reliable intermediate reasoning states to stabilize subsequent generation. By selectively suppressing unreliable cues while preserving trustworthy reasoning context, HMER enables fine-grained mitigation across diverse hallucination facets. Extensive experiments on 19 MLLMs demonstrate the prevalence of emotion hallucinations and the effectiveness of our framework across diverse model architectures.

cs.MM↗

arXiv2026

Multimodal Temporal Modeling for Continuous Group Emotion Recognition in Multi-party Dialogues

To realize natural behavior in dialogue agents in multi-party dialogue scenarios, it is important to understand group emotion such as valence and arousal as a whole. Most prior work addressed this task at the utterance level or using a coarse-grained time window, which is not sufficient to capture emotional dynamics. In this study, we formulate continuous recognition of the Group Emotion at a one-second resolution. Moreover, we also introduce the Mixed state, which captures the emotional divergence among participants in the group. We constructed a dataset with frame-level soft labels based on the TEIDAN corpus and propose a multimodal temporal framework that integrates audio and video information using a sliding-window context. Experimental results demonstrate that the temporal Transformer outperforms simple baselines and shows stronger temporal agreement with the ground-truth labels than the LLM-based model. The effect of context length is limited, whereas audio-visual input outperforms either unimodal input on the continuous-label metrics. Additionally, our analysis shows larger Group Emotion recognition errors in intervals with high Mixed values, exposing emotional divergence as a key challenge for group emotion recognition.

cs.MM↗

arXiv2026

MotionCanvas: Learning Implicit Motion Planning from Composable Kinematic Cues

Professional character animation requires both natural motion and precise, versatile control. For example, it is common for the creators to define the timing of a specified action, to control the motion range of the character's arm swing, and the route the character walks through, like specifying various kinematic motion cues on a ``motion canvas''. This motivates us to propose MotionCanvas, a model that supports \emph{cue-conditioned implicit motion planning} to faithfully and coherently connect all cues, dense or sparse, full or partial, into one full-body motion sequence. Specifically, MotionCanvas represents heterogeneous kinematic cues on a shared motion canvas, where position and rotation values are specified across body joints and time. A shared flow-matching model generates motion conditioned on this canvas, with optional language and input motion; cue imputation keeps the specified canvas values fixed in both training and sampling. To learn coherent completion across different cue sets, we train with a compositional cue sampler that varies when cues are applied, which positions or rotations are specified, and how they are combined. Together, these designs enable a single generator to synthesize globally coherent actions that jointly satisfy compatible heterogeneous cues. We test this planning ability with temporal, root, and body-part cues---alone and in combination---and language-guided editing. We naturally extend this evaluation to sequential generation and motion repair, since both require the same ability to organize coherent motion from kinematic cues. Across these evaluations, MotionCanvas establishes state-of-the-art results in controlled-motion quality, mixed-cue adherence, sequential generation, instruction editing, and motion repair while preserving its text-to-motion capability.

cs.MM↗