Agentic data science with the Data Agent Kit
cloud.google.com·7h ago
AI Research
arXiv CS.AI·24d ago·5 min read
Astronex-World 1.0: Real-Time Interactive World Model Foundation
TL;DR
Astronex-World 1.0 is an open video world-model foundation that generates future visual states based on text prompts or initial observations. It achieves real-time video generation at 24 fps, outperforming several existing models in key benchmarks.
✦ Why It Matters
Explore Astronex-World 1.0 for real-time video generation in your AI projects today.
Key Takeaways
- Utilize Astronex-World 1.0 for interactive video applications.
- Leverage the model's real-time capabilities for autonomous systems.
- Experiment with the bidirectional and causal models for diverse outputs.
- Incorporate camera parameters for enhanced video realism.
How It Works
Astronex-World 1.0 employs a bidirectional model for generating full-context video and a causal model that uses block-causal attention to maintain coherence over time. The model integrates camera intrinsics and extrinsics, allowing it to accurately simulate camera movements and actions.
A 64-dimensional action stream modulates the output at each Transformer layer, enabling dynamic interaction with the generated content.