Meta's first in-house image generator rolls into the Meta AI app, Instagram and WhatsApp
If you use Meta AI, Instagram or WhatsApp, this is the biggest visual upgrade Meta has shipped this year. Muse Image is the company's first image generation model built by Meta Superintelligence Labs, and rather than sitting behind a separate app it is rolling straight into the Meta AI app, Instagram Stories effects and WhatsApp chats. What makes it worth a look is how it works: it reasons through your prompt before generating, pulls in real-time web context, and can blend multiple references together including tagging real Instagram accounts to bring their look into an image. For anyone who creates content for social feeds, that removes a lot of the fiddly prompting normally needed to get a consistent likeness or style.
Muse Video is already in early preview as the natural next step, which makes this look like the opening move in a much broader push to make generative visuals a routine feature of Meta's apps rather than a novelty add-on. Meta surfaced this directly via its About Meta blog before mainstream press coverage, with TechCrunch picking it up the same day.
Grok Imagine now generates 15-second videos, with users reporting a jump in quality
If you use Grok's Imagine tool for video, this update is worth trying straight away. xAI has extended Imagine's video length out to 15 seconds, with early users reporting a genuine jump in output quality alongside the extra runtime. The news surfaced directly through Elon Musk's X account rather than a formal xAI blog post or press release, so some details are still settling, but the update is live in the Grok app now. For anyone using Grok Imagine to storyboard ideas or produce quick social clips, a longer runtime means fewer awkward cuts and more room to tell a story in a single generation, which is the practical gap that was limiting real creative use before.
Vera CPU targets the hidden bottleneck in agent workloads, with Perplexity as an early convert
If you build or depend on AI agents that call tools, run code or query data, this matters: Nvidia is making the case that the CPU, not just the GPU, has become the performance bottleneck. Its Vera chip is designed for maximum single-threaded performance at scale, aimed at keeping every step of an agent's reasoning loop fast even when a system is running many parallel tasks. Perplexity has already tested Vera on real coding workflows and reports jobs completing up to 1.5 times faster than on comparable x86 chips, with concurrent sandboxes starting up to 1.9 times faster. Perplexity says it is now planning to deploy Vera in an upcoming production system.
The practical upshot: if your workflows depend on agents chaining tool calls and data lookups, CPU choice is becoming as strategically relevant as GPU choice for keeping those chains fast and affordable at scale. Nvidia published this directly via its developer blog, with Perplexity's endorsement lending it more weight than a standard hardware announcement.
Bloomberg, citing Reuters, reports that DeepSeek is quietly developing its own inference chip to reduce dependence on Nvidia and Huawei hardware. Nothing has been confirmed by DeepSeek itself, and any production product is likely years away. Worth noting because it fits a wider pattern of AI labs treating chip supply as a strategic vulnerability, not just a procurement line item.
Today's themes
Meta and xAI both pushed out consumer-facing generative media upgrades within a day of each other: still images from Meta and longer video clips from Grok. That is not a coincidence. The competition for attention on personal AI assistants is now as much about what they can create for you as what they can answer, and anyone building products on top of these platforms needs to factor that in. Nvidia's Vera CPU case, backed by real Perplexity benchmark figures rather than theoretical specs, is a reminder that as agents become the default mode of AI use the bottlenecks shift to less glamorous infrastructure choices. Reports of a DeepSeek chip are unconfirmed and years away, but they fit a wider trend of labs treating chip supply as a strategic vulnerability worth solving in-house. Finally, the Grok Imagine update surfacing first through a single Musk post rather than a formal announcement is a reminder that some of the most immediately useful product changes in AI right now are shipping with a social media post doing the job a press release used to do.