
Sam Rowe
2.3K posts

Sam Rowe
@Sam_Rowe
Technologist, Artist, Enterprise growth advisor, Investor, Father, and Husband. All views my own.


Apple has just released details of a new AI video generator experiment: STARFlow-V is a big deal because it proves that Apple has found a way to build a top-tier video AI without using Diffusion (the technology behind Sora, Runway, and Kling). Most current video AIs work by taking a fuzzy static image and slowly "denoising" it into a clear picture. STARFlow-V instead uses Normalizing Flows, a method that smoothly transforms simple data into complex video in a single, reversible stream. Why this is interesting: It’s Reversible: Because the math works in both directions, you can run the model forward to create video, or backward to understand existing video. This makes it naturally good at editing (Video-to-Video) without needing the complex, "hacky" workarounds that Diffusion models require. It’s Streamable: It generates video frame-by-frame in a strict timeline, making it much better suited for streaming or potential real-time applications (like video games or "world models") compared to models that generate a whole clip at once. It’s Competitive: Historically, this technology was considered "too hard to scale," but Apple just showed it can match the quality of today's best models while being potentially faster and more consistent.


Italian narrow roads? How can Tesla FSD handle them???😉



Claude Opus 4 —> 4.1 The latest and greatest model from @AnthropicAI is now available in the GitHub Copilot model picker. ⚡




GitHub Copilot Free for VS Code has arrived.











