ByteDance’s Real-Time & Cinematic AI🎦✨

From real-time cinematic video generation to AI-powered operating systems and next-gen chips, the tech landscape is evolving at lightning speed. In this edition:

• ByteDance breaks new ground with Seaweed APT2 and Seedance 1.0, two powerful AI models redefining video creation—from streaming virtual avatars to cinematic storytelling.

• Microsoft introduces Copilot Vision for Windows, an intelligent on-screen assistant offering real-time, context-aware help inside apps.

• AMD strikes back at Nvidia, unveiling the MI400 AI chips and Helios rack system designed for hyperscale AI infrastructure.

Read on for a full breakdown of these transformative announcements.

ByteDance Unveils Real-Time & Cinematic AI Video Models

Seaweed APT2 is ByteDance's cutting-edge model for real-time, interactive video generation. It can stream 24fps videos up to five minutes long, allowing for smooth, continuous, and low-latency output suitable for applications like virtual humans and camera-controlled world exploration. The model uses autoregressive adversarial post-training to efficiently generate long video sequences with a single network evaluation per latent frame, outperforming previous versions in both speed and duration.

Seedance 1.0, also from ByteDance, is designed for high-quality, multi-shot video generation with a focus on cinematic storytelling and visual consistency. It excels in prompt adherence, motion quality, and aesthetics, producing 1080p videos with rich textures and stable subject representation. Seedance 1.0 leverages advanced training and inference optimizations, enabling it to generate a 5-second 1080p video in just over 40 seconds, making it faster and more fluid than many commercial competitors.

Copilot Vision Debuts on Windows in US

Copilot Vision on Windows with Highlights is now available in the U.S., introducing a new way for users to interact with their Windows 10 or 11 PCs. When enabled, Copilot Vision acts as a real-time AI companion that can see what you see, analyze on-screen content, and provide instant insights or step-by-step guidance within apps. Users can share up to two apps at once, allowing Copilot to offer context-aware help—such as showing exactly where to click for a task, giving tips while gaming, or reviewing travel plans. This opt-in feature is easy to activate via the Copilot app and always keeps the user in control. The update also brings Deep Research and file search capabilities, positioning Copilot Vision as a smarter, more integrated assistant designed to help users accomplish their goals efficiently, with more features planned for future updates.

AMD Unveils MI400 AI Chips and Helios Rack to Challenge Nvidia

AMD has unveiled its next-generation AI chips, the Instinct MI400 series, which are set to ship next year and can be assembled into unified “rack-scale” systems called Helios, allowing thousands of chips to work together as a single massive compute engine—ideal for hyperscale AI data centers. Announced by CEO Lisa Su at a launch event in San Jose, the MI400 series is positioned to compete directly with Nvidia’s leading data center GPUs and will be used by OpenAI, as confirmed by CEO Sam Altman. AMD’s current flagship, the MI355X, began shipping last month and is already adopted by major AI players like OpenAI, Tesla, xAI, and Cohere, with Oracle planning clusters of over 131,000 chips. AMD claims its chips offer superior inference performance due to more high-speed memory and lower power consumption, translating to significant cost savings over Nvidia’s offerings. The company is aggressively expanding in the AI chip market, investing in acquisitions and open-source networking technologies, and expects the total AI chip market to exceed $500 billion by 2028, aiming to capture a larger share from Nvidia’s current dominance.

Hand Picked Video

In this video, we'll look at the viral AI image-generation trick that's blowing everyone's minds inside ChatGPT right now!

Top AI Products from this week

• Solar - Solar lets small teams collaborate with AI agents to build and deploy production apps in minutes. Like Figma, but no mockups — just real software with live data and backend functionality, built by our full-stack AI agent, powered by your imagination.

• Keboola MCP Server - Your AI agent just became a data engineer. Keboola MCP Server lets Claude, Cursor or ChatGPT build actual data pipelines—with error handling, logging & built-in governance. Describe your workflow and get production infrastructure in seconds.

• Omi - Omi is an AI-powered wearable that captures your conversations in real time, helping you never miss an important detail. Worn like a sleek pendant, it records meetings and chats, then turns them into searchable summaries and action items.

• Shipable AI by CNTXT AI - Meet Shipable — Go from prompt to production-ready AI agent in under 60 seconds. No code. No canvas. No chaos. Just describe your bot, and Shipable builds it: logic, UI, integrations (CRM, Stripe, Notion...), and deploys it everywhere.

• oghunt - With oghunt you can view products on Product Hunt with ZERO AI Slop. This allows you to find cool new products without the CRINGE AI showing up and destroying the list. It's a safe, free range, and open-source project on github. Please come help.

• ZenCtrl - ZenCtrl is a framework to Generate multi-view, diverse-scene, and task-specific high-resolution images from a single subject image—without fine-tuning. Just drop your product image and start generating images

This week in AI

• AI NBA Ad Shocks on ABC - Disney’s ABC aired PJ Ace’s wild, GTA-style AI ad for Kalshi during the NBA Finals—created in 2 days with Veo 3, it lets users bet on anything, signaling a bold ad future.

• Massive Molecule Dataset Released - Los Alamos, Meta, and Berkeley released Open Molecules 2025: 100M+ DFT simulations for AI, enabling breakthroughs in drug, battery, and materials design. Publicly available.

• Runway Chat Mode Launch - Runway’s new Chat Mode lets all users create Gen-4 images, videos, and references in one conversational interface—generate anything instantly with simple chat prompts.

• Meta Invests in Scale AI - Meta invested $14.3B for a 49% stake in Scale AI, valuing it at $29B. CEO Alexandr Wang joins Meta’s AI team; Jason Droege becomes interim CEO. Scale stays independent.

• Google Debuts AI Weather Lab - Google DeepMind launched Weather Lab, an interactive site sharing AI cyclone models that predict storm paths, intensity, and size up to 15 days ahead—often more accurately than physics-based models.