Genie 4 world model takes live direction

In partnership with

Welcome to VP Land! AI is showing up earlier and later in the filmmaking process, from world models you direct in real time for storyboarding to generated reenactments in a finished documentary, while a new iPhone app and tool updates from Pika, Apple and Blackmagic hand more control to the people shooting and cutting.

In today's edition:

Google DeepMind's Genie 4 world model takes live text direction
Cocaine Cowboys returns with AI reenactments for its 20th anniversary
Rian Johnson and Steve Yedlin's Chromagraph Camera iPhone app
Pika H3 Flex, Apple Creator Studio and Resolve 21.1.2 scripting
LA passes free micro-shoot permits and $1 crew parking

Genie 4 Is Google DeepMind's New World Model You Steer With Text While It Runs, With Native Audio

Google DeepMind released Genie 4, a world model in research preview that you can steer with text prompts while a simulation runs and that generates its own sound, from ambient noise to dialogue. Trusted testers are pointing it at film prototyping, and a DeepMind engineer cites storyboarding.

Google introduced Genie 4 as a research preview for trusted testers, with filmmaking listed among the creative industries it could serve. You navigate with standard keyboard controls and type prompts to move the camera, change the weather or add a person or object, and Gemini adapts each prompt to fit the surrounding scene.

Audio is generated with the world. Genie 4 produces music, ambient noise and interactive dialogue alongside the visuals, according to DeepMind's announcement.
Scenes hold for up to two minutes. That doubles Genie 3's limit, per TechCrunch's report, which also cites better physics and more humanlike characters.
Gemini directs, Genie renders. Gemini Omni anchors simulations in real-world appearance and behavior while other Gemini models plan steps and check frames. Google research director Tejas Kulkarni described a future of "hybrid world models" in which Gemini writes code and text and Genie simulates it in real time.
Prompts read like shot descriptions. Examples on the Genie 4 model page include a fluid handheld walk backstage at a musical theatre, and an example in DeepMind's announcement asks for a low-contrast 16mm documentary look with the camera gliding over a lake around a saxophone player.
Early-stage limits are on the record. Details and objects can drift the longer a simulation runs, and the team has not worked out accurate audio or legible in-world text. DeepMind engineer Diego Rivas said the real-time element is most useful "in the ideation phase, in the storyboarding phase."

Access starts with trusted testers, then moves to Google AI Ultra subscribers in Project Genie and early access partners in the Gemini API and Google AI Studio. Our breakdown of Genie 4 covers the demos, the Gemini architecture and the open issues in more depth.

SPONSOR MESSAGE

The AI notetaker that gets the hard words right

Your AI is only as good as what you feed it. Feed it a transcript with your product names, acronyms, and numbers spelled wrong, and you spend the afternoon hand-fixing the output.

Wispr Flow Notetaker uses your dictionary and calendar, so product names, acronyms, numbers, and uncommon names come out spelled right. It pops up when your meeting starts and captures Zoom, Google Meet, Teams, and Slack huddles with one click, no bot joining the call.

Then your meetings show up inside Claude or ChatGPT through the built-in connector. No copy paste. Try Notetaker free on Mac and Windows.

Try Notetaker Free

Cocaine Cowboys Gets an AI Overhaul Across Its 4,000-Plus Shots for Its 20th Anniversary

Billy Corben's 2006 documentary "Cocaine Cowboys" is returning for its 20th anniversary with AI work across its 4,000-plus shots, from sharper interview footage and retextured stills to new motion graphics and entirely new AI-generated reenactments. The re-release streams free on YouTube and CocaineCowboys.com starting November 13 at 5 p.m. ET.

Business Insider's Jason Guerrasio reported the re-release as an exclusive. Corben and producing partner Alfred Spellman took on the project after the film's rights reverted to their production company, Rakontur. The original traced how 1970s Miami became the entry point for the US drug trade and introduced audiences to Griselda Blanco.

Upgrades across 4,000-plus shots. The overhaul enhances the resolution of the original interview footage, gives many still photos a richer texture and adds new motion graphics, according to Guerrasio. The film's official site calls the result "a complete visual reconstruction."
New Territory Media handled the facelift. Corben and Spellman turned to the AI studio New Territory Media to rework the film's shots (NTM is the parent company of VP Land).
The reenactments are new scenes, not archival footage. They depict events the former traffickers and hitmen describe on camera, and they play alongside the original testimony rather than replacing it. Business Insider ran an AI-generated reconstruction of Blanco among its images.
Aiming for a film that looks current. "What we hoped was this would look like a movie that was released in 2026," Spellman told Business Insider, adding that AI lets them "scale our style and quality of storytelling in a way that would have been impossible for us to do before." Corben said the work kept improving month to month, and the hard part was "resisting the temptation to start all over again."

For documentary makers, the project shows a full catalog title reworked shot by shot, with generated reenactments built around its original interviews, as Corben aims to reach "a whole new generation of people who might have never heard of it, let alone seen it." We covered the 4,000-shot overhaul and the reenactments in more depth.

Chromagraph Camera Is Rian Johnson and Steve Yedlin's New iPhone App That Replaces Apple's Image Processing

Chromagraph Camera, a $12.99 iPhone app from director Rian Johnson and cinematographer Steve Yedlin, swaps the iPhone's default processing for Yedlin's own color and texture rendering. It builds each image from raw sensor data instead of filtering what the phone already processed.

Chromatic Applications released the app, per its App Store listing, which covers iPhone and iPad on iOS 17 or later. Yedlin built it with Johnson and Eric Cameron, and says the three made it for themselves before making it public.

No filter on top. The Chromagraph product page says the app bypasses iPhone look management and replaces it with the same custom math Yedlin uses on his films, tailored to Apple camera behavior.
The look appears after the shot. You frame, shoot, then open the gallery to see the rendered image, which the developer compares to shooting film. Yedlin said in a statement Newsshooter published that the grading knobs "allow you to make pretty dramatic adjustments while keeping within that overall core look."
Raw originals stay untouched. The gallery keeps full-quality scene-linear raw files and applies the look non-destructively, while exports and an auto-export option send JPEG or HEIC files with the look baked in.

Yedlin calls the rendering "opinionated," covering tonality, chromaticity, grain and halation he, Johnson and Cameron personally like; it is the same approach he used on the Knives Out films. We covered how Chromagraph builds its images in more depth.

🫟 Seamlessify. Seamlessify makes seamless, tileable patterns and textures for design and AI workflows.

👨🏻‍💻 Takehound. Takehound is an AI video editor in the browser for long-form videos where you talk to camera.

💬 Scanfile. Translates text in existing SRT, VTT and ASS subtitle files, plus PDF, EPUB and PowerPoint (PPTX/PPT) documents, in a browser.

Browse the full tool directory

Pika's New H3 Flex Model, Plus MotionVFX in Final Cut Pro and Scripted Timeline Edits in Resolve

Pika's new open-weight H3 Flex model speeds up video drafts, Apple builds MotionVFX into Final Cut Pro, and DaVinci Resolve 21.1.2 lets Studio scripts edit the timeline directly. Together they cover generation, finishing and automation.

Pika H3 Flex. Pika's new open-weight video model is post-trained on MiniMax H3. It generates a 15-second clip with soundtrack in as little as 7 seconds, Pika says, and Pika claims it is up to 1.8x faster (in DiT time) and up to 80% cheaper than H3 Max. You draft at 768p, upscale to 1080p, 2K or 4K to finish, feed in up to 12 references and export up to five audio stems. API pricing starts at $0.01 per second. Other models post-trained on MiniMax H3 include fal's H3 Max and VDN-H3.
Apple Creator Studio. A free Creator Studio update for existing subscribers builds MotionVFX into Final Cut Pro for Mac. It adds templates, titles and transitions, plus rotoscoping, tracking and color grading tools. Timeline images round-trip to Pixelmator Pro, which adds Generate Fill so you can change a selected area from a text description. Beta AI assistants arrive in Logic Pro, Keynote, Pages and Numbers, and new subscribers pay $12.99 per month in the U.S.
DaVinci Resolve 21.1.2. Blackmagic Design expanded the Studio scripting API, according to Digital Production. Scripts can now move, trim and extend clips, remove gaps and silence, copy attributes, and manage effects and grading parameters. They can also grade, render and export Photo page albums. The additions build on the asynchronous APIs that 21.1.1 added for Quick Export, transcription, scene cut detection and Smart Reframe. The scripting improvements are Studio-only, and the update is free for existing Studio customers, though projects opened in 21.1.2 won't reopen in 20.3.2.

GPT-6.1 Terra's Blender Camera Work Outshines Its 3D Modeling in Matt Workman's Seedance 2.5 Test

In an eight-minute video on fal's channel, Matt Workman tests GPT-6.1 Terra in Blender, where its detailed car model comes out rough but its keyframed camera swoop over a low-poly canyon scene is, in his words, "better than I could keyframe it myself." He then feeds that animation to Seedance 2.5 as motion control for a photorealistic driving shot, showing a workable 3D-to-AI pipeline as long as the models stay low-poly and simple.

Stories, projects, and links that caught our attention from around the web:

🎬 The LA City Council passed free micro-shoot permits for sets of 10 or fewer and capped city lot parking and base camp at $1 a day, down from $450, as FilmLA shoot days fall 4.8% year over year.

⚖️ Elvis Presley's music owners sued Anthropic for $48.6 million, saying Claude was trained on 274 Elvis songs without a license, the latest music-catalog suit against an AI lab.

🎮 Epic Games announced an AI game jam, drew criticism and then changed the rules.

🎟️ Sanxingdui: Future Past, which its producers bill as China's first AI-generated theatrical film, opened to weak ticket sales, with trade estimates near RMB1.2 million over its first four days against Bona's stated production cost of more than 10 million yuan.

🧬 Remington Scott, the digital-human pioneer from Lord of the Rings and Spider-Man 2, will lead Eros Imagination, a London unit mixing VFX, AI, digital humans, virtual production and real-time tools that will take work from outside studios, streamers and estates.

CorridorKey 2.0 Takes On the Hardest Green Screen Shots

In Denoised, we break down CorridorKey 2.0, whose keyer now reads the whole clip backward and forward instead of single frames, and Addy reports running a 150-frame 4K clip in about a minute after buying the $50 license that adds a standalone app and host plugins.

We also weigh Kling 4.0's per-second pricing, about half of Seedance 2.0's rate on the same platform, and question when native 4K generation beats upscaling from 720p. The episode recap has timestamps for every segment.

Watch/Listen & Subscribe

Spotify | Apple Podcast | YouTube