Today in Generative Media

Midjourney's AI video model; Animaj raises $85M; What Ancestra means for AI video

News and Opinion

• Midjourney launches its first AI video generation model, V1 (TechCrunch)

• HarbourView Equity Leads $85M Investment in Animaj, the AI-Powered Kids and Family Media Company (EXCLUSIVE) (Variety)

• Ancestra actually says a lot about the current state of AI-generated videos (The Verge)

• Why Hollywood Is Training for Jobs That Don’t Yet Exist (IndieWire)

• Google's AI Is Actively Destroying the News Media (Futurism)

• AI copyright anxiety will hold back creativity (MIT Technology Review)

• As ChatGPT Linked to Mental Health Breakdowns, Mattel Announces Plans to Incorporate It Into Children's Toys (Futurism)

• What’s Happening to Reading? (The New Yorker)

• Authors Are Posting TikToks to Protest AI Use in Writing—and to Prove They Aren’t Doing It (Wired)

• AI Scraping Bots Are Breaking Open Libraries, Archives, and Museums (404 Media)

• Google’s Gemini panicked when playing Pokémon (TechCrunch)

• Fears about AI push workers to embrace creativity over coding, new research suggests (PsyPost)

Software

• LightX2V: Light Video Generation Inference Framework (GitHub)

• Wan2.1-T2V-14B-StepDistill-CfgDistill is an advanced text-to-video generation model built upon the Wan2.1-T2V-14B foundation. (HuggingFace) This approach allows the model to generate videos with significantly fewer inference steps (4 or 8 steps) and without classifier-free guidance, substantially reducing video generation time while maintaining high quality outputs.

• Self-Forcing Video Generation: Real-time video generation with distilled Wan2-1 1.3B (HuggingFace)

• Develop Custom Physical AI Foundation Models with NVIDIA Cosmos Predict-2 (NVIDIA)

• Build and Deploy a Remote MCP Server to Google Cloud Run in Under 10 Minutes (Google Cloud blog)

• Anysphere launches a $200-a-month Cursor AI coding subscription (TechCrunch)

Research

• ImmerseGen Agent-Guided Immersive World Generation with Alpha-Textured Proxies (project page)

• AlignHuman: Improving Motion and Fidelity via Timestep-Segment Preference Optimization for Audio-Driven Human Animation (project page)

• JAFAR: Jack up Any Feature at Any Resolution (project page)

• Evolutionary Policy Optimization (project page)

• Self-Adapting Language Models (project page)

• Don't Pay Attention (arXiv)

• Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task (arXiv)

Misc

• How do you teach taste? Style? Things without clear "right" answers? Art schools figured this out centuries ago through structured critique and feedback. … We combine human preference data with AI critics that grade responses using predefined rubrics. The magic? Critics don't need endless new data - we can iterate on rubrics offline, making the whole system more flexible and powerful. (X)

• I built an MCP-powered RAG over video files (X)

• Our very own Alexander Chen created "Hello, world," a short film inspired by the technology of his childhood, with Google Labs’ Flow. (X)