Seeing and Hearing: Open-domain Visual-Audio Generation with Diffusion Latent Aligners

TL;DR

The paper tackles open-domain cross-modal generation of video and audio by exploiting pre-trained single-modality diffusion models bridged through a multimodal embedding space. It introduces a diffusion latent aligner that performs multimodal guidance in the latent diffusion process using ImageBind, enabling joint-VA, V2A, A2V, and I2A tasks without large-scale retraining. The method relies on multimodal guidance losses across visual, audio, and text embeddings and optional guided prompt tuning to ensure temporal and semantic coherence. Experiments on VGGSound and Landscape demonstrate superior audio-visual alignment and content fidelity compared with baselines, highlighting practical potential for industry-scale content creation.

Abstract

Video and audio content creation serves as the core technique for the movie industry and professional users. Recently, existing diffusion-based methods tackle video and audio generation separately, which hinders the technique transfer from academia to industry. In this work, we aim at filling the gap, with a carefully designed optimization-based framework for cross-visual-audio and joint-visual-audio generation. We observe the powerful generation ability of off-the-shelf video or audio generation models. Thus, instead of training the giant models from scratch, we propose to bridge the existing strong models with a shared latent representation space. Specifically, we propose a multimodality latent aligner with the pre-trained ImageBind model. Our latent aligner shares a similar core as the classifier guidance that guides the diffusion denoising process during inference time. Through carefully designed optimization strategy and loss functions, we show the superior performance of our method on joint video-audio generation, visual-steered audio generation, and audio-steered visual generation tasks. The project website can be found at https://yzxing87.github.io/Seeing-and-Hearing/

Figures (7)

• Figure 1: Overview. Our approach is versatile and can tackle four tasks: joint video-audio generation (Joint-VA), video-to-audio (V2A), audio-to-video (A2V), and image-to-audio (I2A). By leveraging a multimodal binder, e.g., pretrained ImageBind, we establish a connection between isolated generative models that are designed for generating a single modality. This enables us to achieve both bidirectional conditional and joint video/audio generation.
• Figure 2: The proposed diffusion latent aligner. During the denoising process of generating one specific modality (visual/audio), we adopt the condition information (audio/video) to guide the denoising process. By leveraging the pretrained ImageBind model, we calculate the distance of the generative latent $\mathbf{z}_{t}^{M_1}$ with the condition $\mathbf{z}_0^{M_2}$ in the shared embedding space of ImageBind. Then we backpropagate the distance value to obtain the gradient of $\mathbf{z}_{t}^{M_1}$ with respect to the distance.
• Figure 3: Compared with baseline on the video-to-audio generation task. SpecVQGAN fails to generate realistic and aligned audio with the input video. Our method can produce aligned audio with the input video rhythm.
• Figure 4: Compared with baseline on the joint video-and-audio generation task. Our method can produce better text-aligned visual content than the vanilla model. Besides, our generated audio is also of better quality and better alignment with the generated videos.
• Figure 5: Compared with baseline on the audio-to-video task. Given the input audio, the generated videos by TempoToken are not aligned with the input audio and the generation with poor visual quality. Our method can produce visually much better and semantically aligned content with the input condition.