Voyasis

Voyasis is a new AI model that makes open-world games in real time. It was made by Etched and Decart working together. Voyasis is special because it creates videos frame by frame based on what you do with a keyboard and mouse. This makes it very interactive and easy to control. Unlike other AI video models that make videos from text, Voyasis lets you control the action as it happens.

Benefits
Voyasis has several big pluses. It is made to work really well with Sohu, which is Etched''s new Transformer ASIC. This means it can handle next-gen models with over 100 billion parts in 4K video. Voyasis can also handle more than 10 times the number of users compared to H100s. This makes it easy to use and efficient. The model uses a Diffusion Transformer backbone and a new Transformer-based autoencoder. This helps it make fast and high-quality videos. Voyasis also has a Transformer-based variational autoencoder and an accelerated axial, causal spatiotemporal attention mechanism. These features let you interact in real time with almost no delay.

Use Cases
Voyasis can be used in many ways. It is great for making interactive video games where you can control the world in real time. Because it makes videos frame by frame based on what you do, it is perfect for open-world games. Besides gaming, Voyasis can be used for making content and for education. Its real-time interactive features can create fun learning experiences and interactive multimedia content.

Vibes
Voyasis has gotten a lot of attention for its new way of making real-time interactive videos. People like that it makes videos frame by frame based on what you do. They also appreciate that it is made to work well with Sohu, which lets it make high-quality videos at a lower cost and with less power. The teamwork between Etched and Decart was important in making Voyasis work in real time on GPUs. Their inference engine was also key for the live demo.

Additional Information
Voyasis shows a new way to make AI models and products with custom chips. Making Voyasis and Sohu helps make real-time video, voice, inference-time reasoning, agents, search, and more faster, cheaper, and easier. Working with Decart was important in making Voyasis work in real time on GPUs. Their inference engine was also key for the live demo. Voyasis''s code and model weights are being shared, along with a live demo of a larger checkpoint. The model can be run locally using Decart''s special inference platform. This shows that real-time transformer-based video is possible and can be streamed across the web for live experiences.