One Photo, a Walkable World: InSpatio-World 1.5 Turns Images Into Spaces
The demo does not show a clip with a chosen camera path. It shows a space that keeps rendering as you explore it, which is a different kind of model.
← All articles
Category
Ai
327 articles
Distill Once, Plug Into 54 Models: NVIDIA's LongLive-Plug for Video
Video distillation used to be a per-model tax. LongLive-Plug turns it into a reusable component, the way LoRAs turned fine-tuning into a portable file.
ChatGPT Now Dresses You: Image Models Move Into the Shopping Cart
The feature is free, powered by an image model built for editing accuracy, and a sign that image generation is sliding into other workflows as plumbing.
OpenAI's DevDay Math: GPT-6.1 Sol at One-Fifth the Cost, With Astra Held Back
Cheaper and capable inference is the input to agents that run all day. The same week, the company declined to ship its most capable model at all.
NVIDIA's $4,999 DGX Spark 64GB Puts a Petaflop on the Desk for Always-On Agents
A year ago, the models that fit in 56GB needed a rack. Now they fit in a box the size of a lunch, which is what makes the price cut matter.
Grok Imagine Video 1.5 Lite Lands on fal and Venice, and the Pitch Is Anonymity
The model is a cut-down version of one that already exists. What is worth watching is where it landed and how one host is selling it.
Anthropic Will Spend $100M Training 10,000 Engineers, Because the Model Was Never the Bottleneck
The scarce resource in enterprise AI is not model access. It is the person who can make a model safe and useful inside a specific process under specific rules.
Decision Models Are Quietly Replacing the LLM in the Agent Loop
Routing, classification and tool selection do not need fluent text. A small model that returns a probability over a fixed set of answers is faster, cheaper and, for those jobs, often better.
TaoMate-H3 Sells a Metric Nobody Else Was Measuring: Time to First Segment
Most video models compete on clip length. This one competes on how long you wait before anything exists, and it publishes the code instead of a demo reel.
China's New Open Models Are Being Trained to Be Agent Substrates
None of these releases was built to be good at conversation. They were built to be the thing an agent runs on, and the training choices follow from that.
Runway Built an Ad Engine Because Creative Volume Was the Bottleneck
Runway built the system for its own paid marketing and then packaged it. The internal numbers are impressive, self-reported, and silent on how they were measured.
Agora-2 Makes the Case That a World Model Is a Multiplayer Machine
A clip has no state. A world has to resolve what several participants write to at once, which is why the four-versus-sixteen test is harder than a demo reel.