AV3D-Harness open-sourced: first agentic harness for 3D audio-visual real2sim
Scobleizer · x · 2026-10-27
Inspired by OpenAI's GPT-6 Astra 3D demos, the team open-sourced AV3D-Harness, billed as the first agentic harness for 3D audio-visual real2sim—tackling rooms that look right but sound wrong.
The pipeline is fully autonomous with no manual acoustic tuning: it separates sounds from real video captures, proposes 3D sound sources, then simulates → evaluates → refits acoustics recursively to match the recordings. The Apache-2.0 repo covers 3D scene perception, real-to-sim reconstruction, and recursive fitting, with agent instructions and tool contracts built in.
More from Multimodal
- Redditor shares eerie dreamlike AI video of a dim attic bedroom — Sweaty-Ad4488 · 2026-10-27
- MatAnyone-E: 7.4x faster 4K video matting with 7.6% of the GPU memory — ccloy · 2026-10-27
- Single-illustration real-time anime character demo built on See-through and Anime2.5DRig — bdsqlsz · 2026-10-27
- Best video-generation workflow for RTX 5050's 8GB VRAM, eyeing Wan 2.2 5B — RemarkableManager135 · 2026-10-27
- "Glass sponge neuron" AI-generated deep-sea creatures are delightfully eerie — Sauers_ · 2026-10-27
- "Best AI-made creation I've seen so far": netizen rates it 10/10 — sebpaquet · 2026-10-27