Project Genie and the Future of Interactive Media
An insider’s view on world models and what Friday’s $64B market crash misunderstands.
Last Friday, gaming stocks lost $64 billion in market cap. Unity dropped 22%, its worst day since 2022. Roblox fell 12%. Take-Two shed 10%. The culprit? Project Genie; a research prototype my team at Google Labs has been building alongside Google Deep Mind. I’ve spent years building toward interactive AI experiences, and I watched the market panic over something I know intimately. They got some things right. Other things very wrong. Let’s talk about it.
TL;DR: Genie is real. It’s a genuine leap. But “game engines are dead” misses what matters. The technology is there. The product isn’t. When it arrives, it won’t replace human creators. I believe it’ll hand them better paintbrushes.
What is Project Genie?
If you haven’t tried it yet, here’s the quick version.
Project Genie is a research prototype from Google Labs and DeepMind that lets you create interactive 3D worlds from text prompts or images. Type a description, upload a photo, and the model generates a world you can actually explore in real time.
It works in three steps:
World Sketching. Describe what you want—a mountain range, a medieval castle, a red blood cell floating through a vein. The model generates a preview image, you refine it, then you enter.
World Exploration. This is the part that matters. As you move, the world generates ahead of you. There’s no preloaded map. The model creates the path in real time based on your actions, handling physics, lighting, and consistency all on the fly.
World Remixing. Take someone else’s world and modify it. Change the time of day, swap the character, adjust the prompt. Build on top of what others created.
The limitation: sessions are capped at 60 seconds, and some Genie 3 features like promptable events aren’t available yet. It’s an incredible demonstration of a research prototype, not really a full product. Yet.
But within those constraints, people are creating things that couldn’t exist last week. And that’s exciting.
Examples Worth Seeing
I made one myself: a hike through the Eastern Sierras with afternoon light hitting granite peaks. One of my photographs from a trip there became an explorable world.
A few others that stopped me scrolling:
Free Solo: The Game. Steren recreated what looks like Alex Honnold climbing El Capitan—third-person view, sheer vertical rock, stomach-dropping exposure. 2,000+ likes in a day.
Surfing physics. Ziyang Xie captured something important: “Genie 3 can simulate splashes, foam, and their interaction with the surfer that are almost impossible for traditional graphics engines to render in real-time.”
Working mini-map. Bilawal Sidhu was “absolutely floored” to see a handheld mini-map that actually functioned. The model implicitly understood world-space to screen-space transforms.
Lost plane crash. Max Weinbach prompted: “A beach at sunset like Lost, plane crash in the distance with flames, luggage and parts thrown everywhere.” And it worked.
Sundar’s own demo is worth watching too. The gallery at labs.google/projectgenie has many more and we’re working on a way for people to contribute their favorites to the gallery as well. Spoiler alert.
The First Time I Saw It
I saw early versions of Genie 3 last summer and was immediately struck by what this could become. The demo was amazing of course, but the reason I’m so excited about this is connected to one of my core theses. I’ve believed for years that the future of media will be personal, interactive and immersive. So much of my role at Labs has been in pursuit of that vision, and Genie 3 feels like the single greatest leap toward it so far.
What made it feel like a leap rather than incremental progress? The depth of realism. Physics, lighting, motion, aesthetics—intricate and cohesive in ways I wasn’t ready for. The model was making things I had never experienced, even though they were shaped like familiar concepts from modern games.
That’s when it clicked.
Why World Models Matter
There’s a reason people like Yann LeCun have been talking about world models for years. His argument: current LLMs can’t truly reason or plan because they don’t understand how the world works. “If you can predict the consequences of your actions,” LeCun says, “you can imagine whether a particular sequence of actions will fulfill your goal.”
That’s what Genie 3 is attempting. Not just generating images or text, but simulating environments that respond to actions in real time. It’s the difference between describing a world and being in one.
We’re not there yet. But this is the direction.
What the Market Misunderstands
Friday’s crash reflected three misunderstandings. Let me address them directly.
Misunderstanding #1: This is a finished product.
Not yet. What we’ve shipped is a genuine breakthrough—real-time world generation with physics, lighting, and consistency that didn’t exist a year ago. The team has done remarkable work to get here.
But there’s a gap between where we are and where we’re going. Control latency. Long-term memory coherence. The full suite of actions, interactions, goals, and fun that make experiences meaningful. Sessions are capped at 60 seconds for now.
Andrew Ng put it well:
“All of AI has a proof-of-concept-to-production gap.” We can generate stunning worlds. We can’t yet make them into games people want to play for hours. That’s the work ahead.
The breakthrough is real but the journey isn’t over.
Misunderstanding #2: Rendering pipelines are what matter.
Unity’s CEO Matt Bromberg responded on X Friday, calling world models “a powerful accelerator for creative processes.” He’s right; and that framing matters.
The market heard “Google can generate game worlds from text” and panicked. $64 billion evaporated. But here’s what the selloff missed: the best games companies have always differentiated on understanding people and storytelling, not merely rendering pipelines. Genie may be a revolution in available capabilities, but it’ll require connecting it to the workflows of creators to make it truly useful. It doesn’t change what makes games great, we’ve merely added another tool to the arsenal of the creative.
Misunderstanding #3: Better tools replace creators.
History says otherwise. Photography didn’t kill painting; it freed painters to pursue impressionism. Photoshop didn’t replace designers. It expanded what they could imagine. Digital audio workstations didn’t eliminate musicians. They democratized production.
The pattern holds. Better tools give more power to the most creative.
I’m more bullish about possibilities than fearful about changes.
The Fear Is Real. Is It Valid?
The GDC 2026 State of the Industry survey landed Friday too. 52% of game developers now say generative AI is harmful to the industry—up from 18% in 2024. The rate nearly tripled.
One anonymous dev: “I’d rather quit the industry than use generative AI.”
Visual artists are most opposed at 64%, designers and narrative teams at 63%, programmers at 59%. Executives? Only 19% positive, but still the most optimistic group. I understand the fear. A lot is changing, and it’s hard to feel like we’re keeping up.
But I think much of that fear is misplaced. The pattern from every previous creative tool holds: “creatives are hired for their vision, not the tools they use.”
If genAI helps creators share stories with less tedium and more craft, we’ll see an abundance of new experiences. My belief is rooted in optimism about human creativity. I won’t pretend the anxiety isn’t warranted. Change is hard.
What 3 Years Looks Like
I’ve made AI predictions for years, and I tend toward optimistic timelines. Take this with skepticism. Here’s where I think we’re headed:
World models will become teacher systems. A creator describes a world, and the model captures it in a way that’s reliably reproducible—maybe through Gaussian splats, maybe geometric representations, maybe the model itself retains embedding memory or can consume more conditioning signals at runtime. The winning approach isn’t clear, but we need precision and consistency.
Then similar techniques for characters, objects, places, puzzles, quests—the full vocabulary of interactive experiences.
Creators will work at a higher order of abstraction. People will still design the details that matter, but through more powerful interfaces. We’ll operate at a higher level of abstraction, and often in ways that feel new. Perhaps more like directing where you coach a character to perform a role in an experience.
Characters will have looks, wardrobes, voices, backstories, biases; everything needed to emote in interesting ways. Creators will cast these characters into experiences, crafting scenarios with guardrails and motivation that play out organically as players interact.
W'e’ll have fewer “pick option A or B” moments. And as much as I appreciate the innovation of Bandersnatch, I’m not a believer in pre-canned choose your own adventure. Instead, I’m excited to see more experiences that are truly novel, bespoke and new. Maybe we’ll need to build trust before coaxing information from an NPC cast as a defendant. The player side of these experiences will be amazing.
We’ll build environments and situations within which stories unfold. Unlike before, more of these stories will come to life in expressive ways that once required years of 3D training and specialized technical expertise. Higher abstraction will lead to better stories.
It’s not clear how long this transition will take to come about, but the possibilities are exciting and the pace is quick. I get why it’s scary, but it’s that same unknown that has me excited for what’s coming.
What Stays Human
There’s an overhyped experiment playing out on platforms (MoltBook anyone?) where AIs playact as people. There are dozens of companies trying to generate fully autonomous AI narratives. I’m skeptical these lead to meaningful experiences for people.
Fantasy author Mark Lawrence ran experiments comparing AI and human writing. His conclusion: human work “felt more organic, varied and, well, lived in.” AI stories had “dialogue so undistinguished between characters that you could reassign the names and it would still make sense.”
That resonates with what I believe. Humans are essential to telling the human narrative. Each of us has a story—unique feelings and perspectives shaped by actually living a life. AI has no experience of the world. It can mimic, but it hasn’t struggled, failed, or overcome anything.
People create art to express themselves and share in human experience. As long as there are people, there will be stories to tell.
We’re building tools to help that happen. Genie is a new capability in that ultimate human odyssey. But it’s us who will craft the worlds and stories that move each other.
The Bottom Line
The market saw an incredible product demo and priced in disruption. Analysts called it “overblown panic.” Developers expressed fear. Everyone reacted to something different.
Here’s what I see from inside:
Genie is real. Not vaporware. The leap in realism and consistency is genuine—a breakthrough the team should be proud of. But the journey isn’t anywhere near over.
The best game companies have always differentiated on understanding people and helping creatives, not merely shipping rendering pipelines. That doesn’t change.
And what’s coming isn’t replacement. It’s elevation. Creators working at higher abstraction, with less tedium and more craft. Directors instead of pixel-pushers. Many more stories to be told.
I’m still in the arena on this one. Building, shipping, learning. Some of what I believe is probably wrong.
But I believe this: the future of media will be much more personal, interactive and immersive. With Project Genie we just took a massive step toward that coming true.
Think I’m too optimistic? Too bullish on human creativity? Naive to believe game companies remain essential even as the sands shift? I read everything—hit reply and let me know why I’m wrong. And of course everything I write on trond.ai are my own personal points of view. Google neither endorses nor reviews what I write here.