Welcome, humans.

Imagine grabbing an actor in the middle of an AI-generated scene and moving them across the screen. The video keeps playing, and the character actually swings and reacts as they're moved. Corey notices it instantly: "I noticed they swing."

That's the wild little detail from our latest conversation with Pika CEO Demi Guo and creative leads Matan Cohen-Grumi and Lindsay Brillson. They're developing Enact, an experimental interface for staging scenes, talking to AI actors, and changing camera moves as video takes shape. Less "rewrite the prompt and hope," more "stand over here, then try that line again."

Here’s our favorite parts:

  • (06:19) The actor actually swings: Corey spots the odd little physical detail in Live Staging. Lindsay explains that the person responds as though they're being picked up, even while the scene continues.

  • (33:44) Get the perfect take, then move the camera: Matan explains how Enact aims to separate the actors' performance from camera work. You can get the scene you like before trying new angles.

  • (23:31) Your AI cast can rehearse: Lindsay describes actors taking spoken notes, trying alternate deliveries, and incorporating feedback before the final "Action."

  • (19:15) Pika admits the speed tradeoff: Demi says the responsive, real-time model is slightly lower in image quality than their non-real-time version. More control doesn't make the engineering limits disappear.

  • (40:34) Okay, but how do you direct a dinosaur? Demi offers the wonderfully impractical example of something you cannot easily do with a real film crew, but might direct inside an AI-generated scene.

The demos are the grabber, but the discussion underneath is about a more familiar creative frustration: the difference between telling a tool what you want and working with something that responds while you figure it out.

Why watch this? Because if you've ever generated the almost-right AI video ten times, you'll appreciate the possibility of directing a performance instead of constantly starting over. The guests also get into how far the technology has come, what still looks rough, and why creative control matters more than just producing more clips.

Watch and/or Listen now: YouTube | Spotify | Apple Podcasts

P.S. At 24:11, Lindsay calls prompt-based generation "the old way." Corey points out that the old way was basically Q1. AI video already has ancient history, apparently.

Keep scrolling for how Enact figures out which actor you're talking to, our November 18 San Francisco meetup, and four recent episodes worth catching up on.

Additional Resources: How do you direct an AI scene in real time?

A normal AI video generator works like placing an order: describe the scene, wait, then see what comes back. Pika's Enact research is trying to make that process interactive. Here's what Demi and the team describe:

  1. Start with a scene and an action. Instead of writing one giant prompt, the director can move a character, issue spoken direction, or make choices about the camera. The important difference is that these are inputs while the scene is unfolding, not just instructions before a render.

  2. Train for interactions, not only pretty frames. Demi says the team built a synthetic-data pipeline for this unusual task. They first construct 3D scenes, then turn them into more photorealistic video examples using a slower, higher-quality video model. This helps train the system to handle changes in a scene.

  3. Stream the result with low delay. The Enact research builds on Pika's H3 Flex video-model work. Demi says the model needed to handle different kinds of interaction input and respond continuously, rather than deliver five-second chunks that would feel too slow for directing.

  4. Keep track of who got the note. An agent layer provides scene context. Demi gives the example of Sarah and Darren: if you tell Sarah to move closer or deliver a line with more emotion, the system needs to apply that instruction to Sarah, not Darren, and keep it in mind for the performance. That is software context, not evidence that the characters are conscious.

  5. Rethink the camera after the performance. Matan describes locking in the actors' performance first, then trying different camera movements without asking the cast to recreate the moment.

THE CURRENT LIMIT: Demi explicitly says the interactive real-time approach produces somewhat lower image quality than Pika's non-real-time version. And Pika's H3 Flex availability is not the same as announcing broad public access to Enact.

🎙️ UPCOMING EVENT: The Neuron IRL in San Francisco

We're getting together in San Francisco on November 18, starting at 4:30 PM PT / 6:30 PM CT / 7:30 PM ET. Come hang out for drinks, bites, a live Neuron podcast recording, and conversations with the people building and using AI. It's free, but spots are limited.

🎙️ In Case You Missed It…

Four recent interviews and episodes we think you’ll love.

1. For anyone wondering how fast AI really changed: AI Changed More in 4 Years Than You Remember

Corey and Grant recap four years of AI progress

TL;DW: Corey and Grant revisit the last four years of AI in roughly six-month jumps, from the early ChatGPT shock to voice, reasoning, tools, and agents. The fun part is how quickly yesterday's jaw-dropping breakthrough becomes ordinary.

Why you should watch: It is a useful reset for your sense of AI history before you start predicting what comes next.

Watch / Listen: YouTube | Spotify | Apple Podcasts

2. Napster Is an AI Company Now. Seriously.

Napster CEO John Acunto podcast episode

TL;DW: CEO John Acunto explains why Napster shifted from music into visual AI agents for businesses and how its "capture, learn, act, own" philosophy connects customer interaction, AI memory, and data ownership.

Why you should watch: Come for the surprise that Napster builds AI agents now. Stay for a practical conversation about interfaces, businesses, and who controls the resulting intelligence.

Watch / Listen: YouTube | Spotify | Apple Podcasts

3. For factory teams tired of programming robots: Robots That Learn Jobs by Watching You

Standard Bots CEO Evan Beard podcast episode

TL;DW: Standard Bots CEO Evan Beard explains why machines that learn a task from demonstration can handle the variation that defeats older rigid robot programs. He gets into vision, factory reliability, and keeping humans involved where it matters.

Why you should watch: It's a grounded look at what AI actually improves in industrial robotics, minus the humanoid backflips.

Watch / Listen: YouTube | Spotify | Apple Podcasts

4. Can Living Neurons Make AI 5x Faster?

Alex Ksendzovsky and Jon Pomeraniec discuss living neurons and faster AI

TL;DW: Researchers Alex Ksendzovsky and Jon Pomeraniec study living brain cells to discover computational patterns they can put into software. Their reported speed and cost gains run on normal GPUs, not a brain in a laptop.

Why you should watch: The conversation makes a bizarre-sounding idea concrete and explains how biology could inspire more efficient video generation.

Watch / Listen: YouTube | Spotify | Apple Podcasts

🎙️ BONUS: AI News, Rumors & the Week Ahead

Corey broke down the GPT-6.2 Astra and Fable 5.5 chatter, how the midterms might affect AI policy, and where Anthropic IPO speculation stands. Catch the latest AI News, Rumors & the Week Ahead here.

Subscribe to our YouTube Channel for more!

Subscribe to The Neuron on YouTube

More interviews with builders, creative people, and researchers every week.

Stay curious,

The Neuron Team

What'd you think of this podcast episode?

Pick an answer below, then tell us why with the "additional feedback" option.
Mario Vuksan

Mario Vuksan

CEO at ReversingLabs

Mario founded ReversingLabs in 2009. He serves as CEO driving all aspects of the company's strategy, operations and implementation. Mario has held senior technical positions at Bit9 (now Carbon-Black), Microsoft, Groove Networks, and PictureTel (now Polycom). He is the author of numerous research studies, speaking regularly at FS-ISAC, RSA, Black Hat and other leading security conferences. Mario is a graduate of Boston University and Swarthmore College.

The Neuron Logo

Don't fall behind on AI. Get the AI trends & tools you need to know. Join 700,000+ professionals from top companies like Microsoft, Apple, Salesforce and more.

Property of TechnologyAdvice. © 2026 TechnologyAdvice. All Rights Reserved

Advertiser Disclosure: Some of the products that appear on this site are from companies from which TechnologyAdvice receives compensation. This compensation may impact how and where products appear on this site including, for example, the order in which they appear. TechnologyAdvice does not include all companies or all types of products available in the marketplace.

Stay in the loop

Get notified when we publish new articles.