AlayaWorld brings interactive long-horizon video world models to 24-fps generation
AlayaLab · hf · 2026-07-22
AlayaLab introduces AlayaWorld, an interactive long-horizon video world model that generates explorable environments from text, images, or video.
What it does
- Generates 24-fps video at 540p and 720p.
- Uses a 15B video diffusion transformer.
- Produces short latent chunks autoregressively under camera trajectories and switchable text prompts.
How it stays stable
- Combines a persistent sink frame, compressed temporal history, geometry-aligned spatial memory, and recent-frame conditioning.
- Trains on corrupted histories and prediction residuals from its own rollouts to reduce drift.
- Adds a discrete autoregressive distillation setup that cuts inference from about 30 sampling steps to 4 steps per chunk.
Results
- Reports best performance on iWorld-Bench for long-horizon generation.
- Positioned as an open-source foundation for future interactive video world-model research.
Related event: Alibaba Open-Sources 15B Video World Model AlayaWorld(5 posts)→
More from Multimodal
- Moss Image Turbo 3-Step LoRA Tested: Fast Generation at 6-8 Steps — JadedScorpion · 2026-10-27
- Magnific teases October 28 release: 'We're making it move' — cuenca · 2026-10-27
- Midjourney --sref demo: retro Americana Batman art series — gizakdag · 2026-10-27
- AI video is a new medium, not cinema or animation — and it's perfect for depicting dreams — Kyrannio · 2026-10-27
- Dad turns his toddler's 31-year-old quote into a song via SUNO kids' choir — NathanpmYoung · 2026-10-27
- Band's local-AI music video hits 200k views: 900 generations kept 64, 3-second clips beat face drift — Hwang0500 · 2026-10-27