MultiRef-Compass: Towards Comprehensive Evaluation of Multi-Reference-to-Audio-Video Generation

2026-07-15

AI Summary

Existing benchmarks struggle to comprehensively evaluate the multifaceted capabilities of multi-reference audio-visual generation models, particularly in terms of consistency, audio-visual synchronization, and instruction following. To address this gap, this work proposes MultiRef-Compass—the first systematic evaluation framework tailored for this task. It constructs 350 synthetically generated samples via a controllable asset synthesis pipeline, covering diverse scenarios such as multi-view subject preservation, multi-entity binding, and human-object-scene compositions. The framework introduces a four-dimensional evaluation protocol comprising 14 sub-metrics, integrating both automated scores and a re-evaluation-enhanced MLLM-as-a-Judge mechanism to enable interpretable and auditable multidimensional assessment. Experiments across eight representative MR2AV systems reveal significant deficiencies in current approaches across multiple dimensions, thereby validating the effectiveness and necessity of the proposed benchmark.

Abstract

Multi-reference-to-audio-video (MR2AV) generation aims to generate coherent audio-video content conditioned on multiple references and textual instructions. Existing benchmarks mainly focus on text-driven generation, single-reference subject preservation, or isolated audio-video alignment, leaving the emerging MR2AV setting largely unexplored. Compared with these settings, MR2AV requires models to jointly reason over multiple references while generating synchronized visual and audio content. Models must not only preserve each reference faithfully but also correctly bind and compose multiple referenced entities into coherent audio-visual events. To address this gap, we introduce MultiRef-Compass, a unified benchmark for MR2AV generation. It comprises $350$ carefully curated samples constructed through a scalable and controllable asset-composition pipeline, covering multi-view subject preservation, multi-entity binding, and human-object-scene composition. To provide interpretable assessment, MultiRef-Compass defines an evaluation protocol with four dimensions: Basic Quality, Reference Consistency, Audio-Visual Consistency, and Instruction Following, using 14 sub-metrics. MultiRef-Compass integrates automatic metrics with a rejudging-enhanced MLLM-as-a-Judge framework, enabling scalable and auditable evaluation of both perceptual fidelity and reference-conditioned composition. Extensive experiments on eight representative MR2AV systems reveal substantial room for improvement across multiple evaluation dimensions, underscoring the need for a comprehensive benchmark and positioning MultiRef-Compass as a foundation for future MR2AV research.

Problem

• multi-reference
• audio-video generation
• reference consistency
• audio-visual alignment
• compositional generation

Innovation

• multi-reference generation
• audio-video generation
• comprehensive evaluation
• MLLM-as-a-Judge
• reference consistency