SPW-Nav streams one minute of 2K 360° video in real time from a single panorama
AlayaLab · hf · 2026-10-09
AlayaLab proposed SPW-Nav, a language-guided streaming panoramic world model that interprets movement instructions and streams one minute of 2K 360° video in real time from a single panorama.
Method: each instruction is interpreted as camera motion in the previously generated panorama; spherical rotation decoupling applies rotation exactly on the sphere, pose-aligned conditioning keeps translation inputs bounded over long streams, and multi-term memory with a few-step generator continues scenes as instructions change. The authors also release SPW-NavSet, panoramic videos with camera trajectories and verified instructions.
Applications include interactive 3D scene exploration, VR, and embodied agent training. SPW-Nav outperforms prior panoramic generators in camera-following accuracy and video quality, and supports on-the-fly instruction switching.
More from Multimodal
- Qwen-Image-2.1 Outfit-Swap Consistency LoRA Trends on Hugging Face — ausboss · 2026-10-09
- How Seedance V2 shocked the industry: 1,000+ data specialists and licensed cinema-grade footage, not scraped Douyin — creatoroff · 2026-10-09
- Running MiniMax H3 on a single RTX 5090: 0.8MP is the ceiling for 15s clips — Realistic-Fennel-190 · 2026-10-09
- Higgsfield Katana lets AI influencers remake viral edits without a reshoot — SimplyAnnisa · 2026-10-09
- AI video's new market: short-shelf-life viral clips, not feature films — bendee983 · 2026-10-09
- Fable 5.1 Simulation Video Goes Viral With 16.1M Views — FinanceYF5 · 2026-10-09