New todaygpt-6VIDEO
AV3D-Harness open-sourced for 3D audio-visual real2sim, inspired by Astra demos
Inspired by GPT-6 Astra 3D demos, a team open-sources AV3D-Harness, which separates sounds from real video, proposes 3D sources and refits acoustics without manual tuning.
Mingfei Chen
@lasiaflyArchived 9 hours agoOriginally posted 2026.10.27
Mingfei Chen @lasiaflyOriginally posted 2026.10.27
The missing piece to bring 3D worlds to life? SOUND! Inspired by the recent @OpenAI GPT-6 Astra 3D demos, we're open-sourcing AV3D-Harness, what we believe is the first agentic harness for 3D audio-visual real2sim. No manual acoustic tuning. AV3D-Harness autonomously separates sounds from real video captures, proposes 3D sources, then simulates → evaluates → refits acoustics to match recordings. 🔓 Open source: https://github.com/yoyomimi/av3d-harness
x.com/lasiafly/status/2114915811785924869Archived 2026.10.28 10:11