Today in Generative Media - 10/17/25

Promise launches VFX branch; Sora bans MLK Jr. deepfakes; Altman played Hollywood

[This newsletter may appear less frequently next week, while I’m at ICCV!]

News and Opinion

• This Deep-Pocketed New Studio Aims to Bring GenAI Filmmaking and VFX to Legacy Media (Hollywood Reporter)

• OpenAI’s Sora bans Martin Luther King Jr. deepfakes after his family complained (CNN)

• How Sam Altman Played Hollywood (Hollywood Reporter)

• AI’s Copyright War Could Be Its Undoing. Only the US Can End It. (Bloomberg)

• AI content now outnumbers human-written articles on the internet, but the good news is that the slop seems to have plateaued, for now (PC Gamer)

• Spotify partnering with multinational music companies to develop ‘responsible’ AI products (The Guardian)

• Is art dead? What Sora 2 means for your rights, creativity, and legal risk (ZDNet)

• Google’s viral Nano Banana AI image tool has already hurt Adobe, data suggests (Business Insider)

• Anthropic is giving away its powerful Claude Haiku 4.5 AI for free to take on OpenAI (VentureBeat)

• How ByteDance Made China’s Most Popular AI Chatbot (Wired)

• The AI Industry’s Scaling Obsession Is Headed for a Cliff (Wired)

• Alibaba says its AI spending in e-commerce is already breaking even (CNBC)

• Method teaches generative AI models to locate personalized objects (MIT News)

Software

• Sa2VA: Marrying SAM2 with LLaVA for Dense Grounded Understanding of Images and Videos (GitHub)

• Introducing Claude Haiku 4.5 (Anthropic)

• Anthropic turns to ‘skills’ to make Claude more useful at work (The Verge)

• I replaced NotebookLM with a self-hosted alternative because I don’t want Google having more of my data (XDA Developers)

Research

• FlashWorld: High-quality 3D Scene Generation within Seconds (project page)

• USplat4D: Uncertainty Matters in Dynamic Gaussian Splatting for Monocular 4D Reconstruction (project page)

• Editable Physically-based Reflections in Raytraced Gaussian Radiance Fields (project page)

• InfiniHuman: Infinite 3D Human Creation with Precise Control (project page)

• Advancing End-To-End Pixel-Space Generative Modeling via Self-Supervised Pre-training (project page)

Misc

• These ICCV 2025 papers are the Open Access versions (Computer Vision Foundation)

• changing anything (face, clothes, device, sky, surroundings, .etc) in the photo by hovering over it and moving mouse wheel, without prompting: (X) [concept video, not a demo]

• Me: Can I upscale these GTA5 gifs so they look real? Wan2.2: Hold my beer (X)