
Welcome to VP Land! Filmmakers are keeping the director's chair. Editors are holding onto the cut and handing AI the busywork, while the newest video models finally take direction as they generate.
In today's edition:
A LUMIX with real cinema camera chops
Where film editors draw the line on AI
Apple's video models you can steer as they generate
Pika, Krea, Kling, Runway and Comfy in AI Tools
AMD closes its World Labs deal

LUMIX SHX1 Is Panasonic's New 6K Full-Frame Cinema Camera, With ARRI and Blackmagic Color On Board

The SHX1 records Blackmagic RAW straight to an SSD and ProRes RAW to an internal card, so no external recorder hangs off the rig, and it costs $4,699.99 body only in the US.
Panasonic's global announcement builds it around a 6K full-frame partially stacked sensor. ARRI LogC3 and selectable Blackmagic color science sit alongside V-Log, which Panasonic says simplifies matching the SHX1 to an ARRI main camera or Blackmagic cinema cameras.
15 stops, by Panasonic's count. That's in V-Log with Dynamic Range Boost on (14+ with it off), though Panasonic's fine print says Boost limits ISO and adds rolling-shutter distortion.
Set-ready I/O. You get 12G-SDI up to 4K 60p and BNC timecode in/out, per Panasonic's US store page, plus internal 32-bit float audio.
No record limit at 4K 120p. It's the first LUMIX to run on micro V-mount batteries, and Panasonic says 4K 120p records without a time limit within its rated operating temperatures.
A PTZ option later. A Live Production Unit due in the first half of 2027 adds Genlock, PoE++ and IP control for pan-tilt heads.
It's pitched as the B camera beside an ARRI, though Panasonic itself notes the ALEXA 35 uses LogC4 instead, and we'd want to see that 15-stop figure tested with Boost's trade-offs. It reaches the US in January 2027, and our SHX1 breakdown has the Open Gate modes and add-ons.
SPONSOR MESSAGE
The voice-to-text app that built its own speech model
Most dictation apps are a wrapper around someone else's speech model. Wispr Flow trained its own.
It's called Canto, and it was built for the way people actually talk to AI: thinking out loud, changing direction mid-sentence, half-whispering in an open office.
Wispr tested it on 10 hours of real dictations from more than 2,300 speakers. Canto had the lowest word error rate of every model tested, including models from Google, OpenAI, AssemblyAI and Deepgram. That means your prompts to ChatGPT, Claude and Cursor come through the way you said them.
Talk naturally in any app and Wispr Flow types clean, formatted text, 4x faster than typing.

Film Editors Are Giving AI the Prep Work, Avid's ACE Survey Finds

More than three in four of the 120 ACE members Avid polled use AI somewhere in post, yet only 7.5% want it helping assemble shots or rough cuts.
The survey came from Avid, which ran it to test stories it had heard about editors resisting AI. "Nobody grows up saying 'I dream of syncing dailies one day,'" said Matt Feury, Avid's VP of artist relations.
Transcription comes first. It leads at 60%, ahead of audio cleanup at 55% and searching footage for specific shots at 39.2%.
Some days, half a day down to 15 minutes. On MasterChef, Emmy-nominated editor Molly Shock's team used AI shot search to find a contestant glancing at a clock in seconds. "I wouldn't say I'm more efficient," she told Variety; the time goes to "the real creative stuff."
Concerns haven't gone away. Ethical, copyright and security worries made up nearly half the barrier to using AI.
Those chores are what Avid's browser-based Media Composer now handles by prompt, from syncing media to organizing bins, so read this as a vendor's survey. The MasterChef example is still hard to argue with, and our breakdown of the survey has the full results.
Apple Releases FastFrames Video Models That Generate Frames in as Little as 36 ms

Apple released FastFrames, two video generation models built for interactive latency, with the 1.3B FastFrames-S producing a 480p frame every 36 ms on one NVIDIA B200 GPU. Apple says that speed lets you steer the camera with a keyboard or drag objects with physics while the video generates.
Apple's machine learning researchers posted both models on Hugging Face. The 14B FastFrames-L reaches 45 ms per frame on an eight-GPU B200 node and 158 ms on a single GPU.
Latency hid outside the diffusion step. The FastFrames paper finds VAE decoding and KV cache updates account for over half of end-to-end latency, so Apple replaced them with a lightweight pixel decoder and KV refinement network.
Speed without a quality hit, per Apple. According to the project page, FastFrames-S keeps the visual quality of Causal Forcing++, which takes 272 ms, and FastFrames-L is comparable to Krea-Realtime while running over 10x faster.
Built on Wan2.1-14B, checkpoints public. FastFrames-L is distilled from Wan2.1-14B, and Apple released its full training chain plus a LingBot 1.3B it says is 16x faster than the official model.
Running FastFrames takes Linux and an NVIDIA GPU with a CUDA 12.8-compatible driver, and the LingBot demo weights carry a non-commercial CC BY-NC-SA 4.0 license. Our FastFrames coverage has the full latency breakdown.

🎞️ Sisyphos. Mac video editing app that automatically builds first cuts from footage using visual and audio analysis, with local processing and deterministic results.
🗣️ Cutora AI. AI video editor that automates first cuts of talking-head videos
🦸 Prelulu. Generates real-time facial expressions and lip-sync for live, interruptible video calls with custom AI characters.
Pika and Runway Preview Real-Time Generation as Krea, Kling and Comfy Add Video Swap, 8K Stills and Cloud Builds

Pika and Runway are testing generation that responds live, Krea and Kling add footage swaps and 8K stills, and Comfy moves a team's whole ComfyUI setup into the browser.
Real-time research previews
Pika Enact. Pika's Enact research preview lets creators interact with AI actors, re-stage a scene and move the camera in real time, built on Pika H3 Flex, its post-trained version of MiniMax H3. Demos are on request, with no pricing or release date.
Runway Labs Diffusion-UI. Runway Labs' Diffusion-UI is a research app where a real-time model generates every pixel of an interactive interface from data like map routes or weather, laid out by moving a box and edited live by text prompt.
Generation and editing
Krea Video Swap. Krea's Video Swap replaces people, outfits, products or backgrounds in existing footage from up to four reference images, keeping the clip's motion, camera movement and sound. Output runs up to 1080p in the web app and by API.
Kling Image 4.0. Kling's new image model outputs native 8K stills, which Kling calls an industry first, and returns sets of up to 10 consistent images per prompt for storyboards and product shoots.
Team pipelines
Comfy Cloud Builds. Comfy Cloud's custom Builds package a team's nodes, models, LoRAs and private node packs into one setup every artist opens in the browser, on GPUs from RTX PRO 6000 to B200, at $1.75 to $11.23 an hour billed by the second.

Sirio Rebuilds Higgsfield's Character-Swap Workflow With Depth Maps, Seedance 2.5 and a Coding Agent
In a 22-minute video, Sirio reverse-engineers the workflow behind Higgsfield's Genjutsu app, turning footage into depth maps and pitch-shifting vocals so the clip keeps its motion while stripping the faces and music that trigger Seedance 2.5's IP blocks. His point is that wrapper apps package API workflows filmmakers can build and automate themselves, which he packs into an 11-step agent workflow with its own interface.

Stories, projects, and links that caught our attention from around the web:
🎭 UK actors union Equity is recommending members approve a Pact deal with per-use pay and project-specific consent for digital replicas, limits on alterations and synthetic performers, and AI-training licensing obligations, though it did not win replica deletion or limits on routine VFX scanning.
🏜️ Director David Ayer joined Ryan Kavanaugh's ACME AI as a principal and managing partner at the company behind the Grey Box performance-transfer stage in Los Angeles, which ACME says cuts location time and costs by 50 to 75 percent.
🤖 Disney Imagineering and AWS built three AI-powered Star Wars Anzellan droidsmiths that use onboard cameras and mics to react to guests and each other unscripted, with a public debut at re:Invent.
✂️ The New York Times looks at AI rotoscoping in VFX, where artists say the tools cut tedium and time but remain imperfect and raise job concerns.
🧠 AMD closed its $8.2 billion all-stock acquisition of World Labs and named Fei-Fei Li EVP and chief scientist, with her team continuing to build the Atlas world model that turns photos into 3D worlds and Gaussian splats.

👔 Open Job Posts
AI Supervisor
Framestore, London, United Kingdom
AI Lead Artist
Framestore, London, United Kingdom
GenAI Technology Integration Manager, Production Solutions - APAC
Netflix, South Korea
Director, Production Innovation, Production Solutions - APAC
Netflix, South Korea
3D AI Generalist (6 Months Contract)
Rodeo FX - Toronto, Canada

📆 Upcoming Events
Oct 2 to Dec 16
AInVFX Generative AI Bootcamp for Film and TV — Cohort 2
Online
Nov 20 to 23
Flux Festival 2026
Los Angeles, CA
Nov 26
A Practical Approach to Virtual Filmmaking
Online
Dec 1 to 4
SIGGRAPH Asia 2026
Kuala Lumpur, Malaysia
Dec 2 to 3
SMPTE Florida Technology Conference 2026
Orlando, Florida

Thanks for reading VP Land!
Thanks for reading VP Land!
Have a link to share or a story idea? Send it here.
Interested in reaching media industry professionals? Advertise with us.





