Runway's Praxis-1 Bet: Video Pretraining as the Shortcut to Robot Brains
Real robot data is scarce and expensive. Runway argues the shortcut runs through video, and it is reporting a 0.95 correlation between simulated and real results.
← All articles
Category
Ai
327 articles
Utopai Wants AI Filmmaking to Remember the Whole Movie, Not Just the Shot
Generating a convincing clip is routine now. Remembering what a production decided across forty shots is the harder problem, and Utopai is betting it is the product.
PixVerse R2 Turns Video Generation Into a World You Can Walk Through
Instead of returning a clip, R2 keeps a world running while you move through it. The trade-off is quality per frame, and AIsphere is upfront about it.
Ideogram 4.5 and the Edit-Drift Problem: Why Changing One Thing Is Still So Hard
Editing an image once is easy. Editing it fourteen times without the picture falling apart is the hard part, and three companies are now racing to solve it.
The Next Fight in AI Video Is Over Understanding the World
When several models can draw a convincing clip, quality stops separating them. The next contest is whether the model understands how events should unfold.
An Open Image Model That Runs in Four Steps Is a Bigger Deal Than It Sounds
Cutting sampling from 50 steps to four is not a small optimization. It changes latency, cost, and whether an image model can run on the machine in front of you.
OpenAI Split Its Flagship Image Model in Two. That Says More Than the Benchmark Does.
Images 2.5 arrives as two models at one price. The fork is an admission that the people buying image generation have split into two different kinds of user.
Domestic Large Models Rush to Open Source in September: This Time, the Competition Is Over “Affordability”
GLM-5.3-Flash is open-sourced under the MIT license and runs on domestic chips, while Meituan serves up the 1.6-trillion-parameter LongCat-2.0. In September, domestic large models rushed to open source, and what they are really competing over is “affordability.”
The Image Model Arms Race: Midjourney, Nano Banana, and the Quiet Rise of Open Weights
Everyone can render a photorealistic image now. The new race is iteration, consistency, and license clarity. Here is where the image models stand.
Wan 3.0 Just Topped the Video Leaderboard. The Open-Weights Story Is Now a Video Story.
An open-weights Chinese model now beats the closed leaders on quality and price. Video generation just got a credible open road.
Kling 4.0 and the New Controllability Race in AI Video
The biggest complaint about AI video was never quality. It was control. Kling 4.0 is betting the future on editing one thing instead of regenerating everything.
Kling 4.0 Is Here, AI Short Dramas Hit 430,000 in a Year: What's Behind Chinese AI Video's Leap Forward
With 30-second clips, local editing, and 10 keyframes, Kling 4.0 turns video generation from a gacha pull into something editable. Behind it: an explosion of 430,000 AI short dramas — and the homogenization debate that came with it.