Excited to share our new work, SonoWorld, led by @DerongJin and Xiyi Chen!
Silent worlds are incomplete. We turn a single image into an explorable 3D audio-visual scene, enriching visual worlds with immersive spatial sound.
The framework is modular and training-free and can be built entirely with open-source models while also supporting proprietary models such as Marble for enhanced visual fidelity.
Kudos to the team for making this happen. Thanks as well to @theworldlabs and to @drfeifei for the early feedback and encouragement.
Explore the 3D audio-visual world below (🎧 headphones recommended) 👇
Silent worlds are incomplete. We turn a single image into an explorable 3D audio-visual scene, enriching visual worlds with immersive spatial sound.
The framework is modular and training-free and can be built entirely with open-source models while also supporting proprietary models such as Marble for enhanced visual fidelity.
Kudos to the team for making this happen. Thanks as well to @theworldlabs and to @drfeifei for the early feedback and encouragement.
Explore the 3D audio-visual world below (🎧 headphones recommended) 👇
Generated 3D worlds look amazing — but they’re silent.
Excited to share SonoWorld at #CVPR2026: from one image, we generate an explorable 3D world you can walk through and hear with grounded spatial audio.
📍Poster: Jun 7, 11:45–1:45 MDT, ExHall F
🌐humathe.github.io/sonoworld/
1/7View quoted post on X
