Three stories crossed my desk this week. Two are about Hollywood making room for AI. The third is about a new model and a reminder to read the fine print. 1. Culver City is turning into an AI studio town. The LA Times counts a cluster of AI filmmaking companies moving in: Promise, TrueShort, Native Foreign, PhantomX, and Luma AI, which took a 7,000-square-foot space for its Dream Lab. Promise works hybrid: real actors on real soundstages, with AI handling backgrounds, visual effects and synthetic humans. Not "push a button, get a movie." Real crews, new tools. (The unions are watching closely, and they should be.) 2. AI studios now have a manager. Variety is reporting that former UTA agent Dante DeSario launched Collage, a management and investment company that walks AI-first studios through the guilds, the studios, and the copyright questions. Clients include Asteria, Native Foreign, Promise, and Jason Zada's Secret Level. When Hollywood builds representation for something, it has decided that it's real. And his word for what separates the good ones? Taste. 3. A fast, free video model... with a catch. Hao AI Lab released FastH3 Omni, a sped-up version of MiniMax's H3 model. Feed it reference images, clips and a voice, and it gives you video with sound, up to about 12x faster than the original by their own numbers. It already works in ComfyUI. Everyone called it "open." Here's the thing: the license on the model doesn't cover the US, the UK, the EU or South Korea. If you're here, you have to ask MiniMax for permission first. Read the license before the benchmark. Notice Native Foreign shows up in the first two stories. The buildings and the business side landed in the same week, here in LA. Where it goes from here... we'll see.

The catch-up issue is live. Three weeks in one place. What's in it: Adobe put five AI video models on the Premiere timeline. You draw a range on a track, describe what belongs there, and fill it. The model picker sits next to the credit cost. After Effects got an assistant that builds real expression rigs you can still open and edit. OpenAI switches Sora off on September 24. Thirteen days. The replacement column on their own deprecation page is empty on every row. If you have work sitting in a Sora account, go export it. Gemini Omni 1.1 traded spectacle for control. Scene extension to forty seconds, and you can now specify the first and last frame of a shot. Danny Boyle disclosed using AI in Ink. Katzenberg is reportedly building a company to make the tools. Two names you know, moving from opposite ends of the business. And a habit worth stealing: log the exact model name and version with every generated shot. Runway killed two models in July with no grace period. Made in Runway is worthless in two years. gen4_aleph is worth something. The Sora one is the only thing with a clock on it. Everything else keeps. What did you lose three weeks of? Tell me what I missed.

Something quietly tipped over this week, and today's newsletter is all about it. For a couple of years now, the AI-in-film conversation has mostly been an argument. Is it real? Is it cheating? Is it going to stick? This week the argument got a tiebreaker, and it came from about the most credible voice there is. George Lucas said AI is the future of the medium. His actual words: "Artificial intelligence means it's much easier for us to make movies," and "There's nothing you can do about it. That's progress; it's the future." That's the person who built the modern blockbuster telling you the door is open. And it's not just talk from the sidelines. The same week, the biggest streamer on earth put a number on it. In its quarterly letter to shareholders, Netflix said creative partners used generative AI across roughly 300 titles this year, mostly in post-production. Not a demo. Not a pilot program. Three hundred titles, on the record, to Wall Street. Put those two together and the shape of the week is hard to miss. The most famous names and the biggest money landed on the same side, out loud, at the same time. In today's full newsletter, I walk through what that actually means for working creators, plus the honest other half: a director who shipped a 135-minute AI feature almost by himself, Christopher Nolan calling the whole thing a trap from the other side of the same week, and a couple of new tools you can actually put to work. It's a five-minute read, and it's built to help you decide what to use and what to keep your hands on. Read the whole thing here: https://aiography.beehiiv.com/p/george-lucas-says-ai-is-the-future-of-movies Here's the thing I keep landing on. When a tool goes from "maybe" to "here to stay," the interesting question stops being whether you use it. It becomes what you refuse to hand over. The tools are settling. Your judgment is the part that still can't be automated, and this week made that clearer, not less.

A quick craving turns into an action-movie showdown — until Instamart delivers right on time. This AI-generated cinematic short blends sci-fi VFX with a beach sunset showdown between a man and a skull-faced armored warrior, all resolved by 10-minute delivery.

I created multiple AI-generated tsunami VFX shots to test the limits of AI video generation. This was purely a visual AI/VFX experiment — I wanted to see how well AI could create a massive tsunami hitting realistic city environments.