The AI frontier just leveled up, more connected, expressive, and personal than ever. Dropbox Dash unifies all your work apps with context-aware search and smart Stacks, OpenAI’s new music generator turns text into melody, and Instagram’s Restyle transforms Stories with AI-driven edits and effects, all redefining how we create, collaborate, and share.

🎬 LTX-2: Open-Source 4K AI Video Engine – LTX unveils its most advanced open-source creative engine unifying audio and video generation in perfect sync. Delivering native 4K at 50 fps and supporting longer cinematic sequences, it empowers filmmakers with an API-first, GPU-friendly system that’s as open as it is powerful.

🌍 Google Genie 3: From Text to Playable Worlds – Google’s next-gen AI world model turns your prompts into fully interactive 3D universes. Describe a world or character, and watch it come alive with physics, memory, and dynamic events, ushering in a new era of explorable, persistent AI environments.

🎥 OpenAI Sora: Create, Collaborate, and Connect – Sora’s upcoming update brings AI-generated cameos, built-in editing tools, and social channels for creators and communities alike, plus an Android release that opens the door to everyone.

From unified workflows and generative music to cinematic engines and living AI worlds creativity has never felt this alive.

Unlock Unified Workflows with Dropbox Dash

Dropbox Dash is a context-aware AI teammate that connects and searches across all your work apps, like Dropbox, Slack, Microsoft 365, and Notion delivering instant, relevant answers and organizing files, links, and updates into shared workspaces called Stacks. Its AI-powered universal search understands natural language queries, provides rapid document summaries, helps teams collaborate from one dashboard, and secures content with robust access controls transforming scattered information into actionable insights, all without switching tabs or compromising privacy. Dash is now broadly available for easy team setup, with deeper multimodal support coming soon for video and multimedia workflows.​

OpenAI’s Music Generator Is Coming

OpenAI is reportedly developing a new generative AI music tool that can create music, tunes, and instrumental accompaniments from text and audio prompts, aiming to let users add music to videos or vocal tracks by simply describing what they want. The system could be integrated with existing OpenAI platforms like ChatGPT or Sora, though it’s not yet clear if it will roll out as a standalone product. OpenAI is collaborating with students from The Juilliard School to annotate musical scores and enrich the model’s training data, leveraging advanced machine learning to enable expressive, contextually appropriate music creation for content creators and digital artists.​

Transform Your Stories with Instagram’s AI Power

Instagram’s new Restyle feature, powered by Meta AI, lets users instantly edit Stories by removing objects, adding artistic elements, or transforming styles via simple text prompts all from within the app. You can use preset effects to restyle outfits, create thematic backgrounds, or apply special filters to both photos and videos, while the ‘Add Yours’ sticker encourages friends and followers to remix and share trending edits. Beyond fun effects, Restyle democratizes pro-level editing, lets users tweak scenes with conversational prompts, and introduces seasonal effects and AI-powered text restyling, making creative story production more interactive while raising new questions around privacy and content ownership.​

Hand Picked Video

In this video, we'll look at Sketch to Thumbnail.

Top AI Products from this week

• AI CMO by Lindy - Lindy is the easiest platform to build no code agents. We shipped marketing actions including Sora 2, Veo 3.1, GPT Imagen, and Nano Banana. Combined them into 3 CMO agents that run campaigns end-to-end.

• Alai - Alai gives 4 options for each slide so that you can discover something that feels right. With Alai’s system of elements, making edits is easy without wasting time tweaking pixels.

• Presentation 2.0 by CubeOne - We make you ready, not just slides. Most AI give you slides to study. CubeOne gives you the story, script, and visuals so you can present right away. Go from nothing to stage-ready in 5 min. Stop making slides, start telling the story.

• Mico - Mico is a new animated persona for Microsoft Copilot's voice mode. It adds visual expressions like smiles and nods to make conversations feel more natural and adaptive. Powered by Copilot, Mico reacts in real-time and can be personalized with different colors.

• Velona AI - Autonomous AI agents that find hidden fleet costs, predict failures, and identify risks in real-time. Works with any telematics provider or OEM. Built on Databricks. Your data stays yours. Get answers in seconds, not weeks.

• Phare Incident AI - Phare's Incident AI digests all the fine details and churns out readable summaries and post-mortem of incidents, saving you from squinting at endless logs like a caffeinated mole.

This week in AI

• Model Stress Tests Reveal Gaps - Anthropic’s new research uses stress scenarios to test model specs, exposing character differences and contradictions in 12 leading LLMs; results show high cross-model disagreement and spec gaps.

• Gemini Unleashes Spooky Visuals - Google Gemini’s Nano Banana and Veo models let users instantly craft hyper-realistic, cinematic Halloween images from prompts; AI art goes viral on social platforms.

• Samsung Debuts Perplexity TV App - Samsung launches the first Perplexity AI-powered TV app, enabling real-time answers, trip planning, and recommendations—plus a free year of Perplexity Pro for users.​

• Google AI Studio’s Vibe Coding - Google’s Vibe Coding in AI Studio lets users build AI-powered apps from simple prompts, automatically connecting Gemini models and APIs. It offers conversational, visual editing, and instant prototyping. Available now for all skill levels.

• PokeeResearch-7B Advanced Deep Research AI - PokeeResearch-7B is an open-source 7B-parameter AI agent using Reinforcement Learning from AI Feedback to perform complex multi-step research with self-verification, outperforming peers on benchmarks.​

Paper of The Day

The paper "AstaBench: Rigorous Benchmarking of AI Agents with a Scientific Research Suite" introduces AstaBench, a comprehensive benchmark suite for evaluating AI agents on scientific research tasks. It covers 2400+ problems spanning literature review, data analysis, coding, and end-to-end discovery, using realistic, reproducible scientific environments and production-grade search tools. Through extensive tests on 57 agents across 22 classes, AstaBench reveals current limitations of AI assistants in science, providing rigorous metrics, cost-accounting, and a leaderboard to drive future advances in agentic AI for scientific discovery. The suite supports standardized, fair comparison with open-source tools and APIs to accelerate reliable AI research workflows.

To read the whole paper 👉️ here