Welcome back to AI Gamechangers, your regular chat with people across the games industry who are doing genuinely interesting things with AI. Every edition, we sit down with a senior practitioner to talk tools, pipelines, and real-world results.
This week’s guest has worked on some of the biggest franchises on the planet, from World of Warcraft to Monopoly Go, and she’s used that experience to build something she believes the industry has long needed. Irena Pereira, founder of Infinite Realms and Unleashed Games, has created an AI platform that transforms source material like books, scripts and lore bibles into structured, queryable world models. The pitch: making IP adaptation faster, cheaper, and actually faithful to the source.
As ever, scroll to the end for a round-up of the latest AI and games news (and remember to browse our archive for free interviews with CEOs and studio heads).
Irena Pereira, Infinite Realms
Meet Irena Pereira, founder and CEO of Infinite Realms, a platform that uses AI to transform written source material into structured world models for IP adaptation across games, comics, streaming and more.
With a career spanning Blizzard, Scopely and a string of independent ventures, Irena brings deep experience in user experience, IP licensing and cross-platform game development.
In this interview, she explains how Infinite Realms helps teams work with complex IP (from Jenn Lyons’s The Ruin of Kings to RA Salvatore’s Demon Wars) without requiring everyone in the process to be experts on the books first.
Top takeaways from this conversation:
Infinite Realms converts source material (novels, scripts, lore documents) into a structured, relational world model that teams can query in plain language, enabling writers and designers to work quickly with complex IP.
The platform addresses a structural problem: approval and adaptation processes currently used to manage major franchises rely on email chains and individual experts, creating bottlenecks in licensed game development.
Irena’s team built a game and a three-episode webcomic based on a fantasy novel in two months, with a budget under $200,000 (and no one on the production team was familiar with the book beforehand).
The system is text-only and positions AI as an accelerant for human creative teams rather than a replacement, with Irena arguing that fast-follow transmedia execution is the real competitive advantage IP holders are missing.
AI Gamechangers: How would you describe what the Infinite Realms platform does and who it’s for?
Irena Pereira: Our platform essentially introduces the capability to be able to take any source material and turn it into a walled-garden LLM.
It makes something like a custom GPT for a book or a movie or a television series, to enable it to be adapted into multitudes of different formats. It can be used anywhere from production, content, ideation, go-to-market, marketing… It is essentially the ultimate lore master of your source material.
What was the fundamental problem you were trying to solve? What did AI bring to the table that couldn’t be done before?
The way people have always done IP adaptation is by having one person in-house that is the true expert on a particular property.
I’ve worked on Game of Thrones, I’ve worked on World of Warcraft, I’ve worked on Monopoly Go – and the processes for not only the adaptation, but also the approval chains (running something by George RR Martin, for instance) are difficult. You have an idea, you send it into the agent’s inbox, and they’re on vacation for two weeks, and then when they come back, you’ve already developed three different alternatives, spinning wheels because you have to keep your team busy while waiting for George RR Martin, who is supposedly trying to write his books… We’re pulling him aside to be able to approve whether or not this or that shade of red is appropriate! The processes currently used to manage the adaptation of the most valuable franchises in the world are the same systems that my parents used to manage their book club. It’s email!
“If I can build a game in eight to ten weeks, there’s no excuse for not starting now. And you can do it in a way that’s consumer-friendly around AI: our AI isn’t generating the art; it’s accelerating your team and reducing the communication and approval lag”
Irena Pereira
Unfortunately, we’ve never thought of IP as core infrastructure, especially in games, and this is actually where it’s the riskiest. One of the reasons why we don’t adapt IP as a first choice (you see a lot of funding going towards original IP) is not because people don’t want to adapt great IP; it’s because it’s too friction-filled. It’s too onerous. There’s too much headcount in simply managing the relationships with the IP holder. There are massive minimum guarantees.
The system we’ve built doesn’t necessarily use AI the way you typically think about AI. AI is a companion to the technology. People aren’t necessarily experts, and maybe they’d have to refer to a 500-page novel before they can start developing. So we use AI, at first, as a processor, to comb through the pages of the book and transform source material into a structured, queryable world model that helps teams understand canon, relationships, timelines, tone, and adaptation opportunities. It has relational connections to each other, including what page numbers they’re referenced on, who they look like and their family relationships, if they change their name, if they die… all of the things that can be present in fictional universes.
Then we can later use any off-the-shelf AI to query the database, and have engaged conversations with the data, and find lines of dialogue, or ask it to explain something in a way that will appeal to Gen Z. We have the ability to query this as if it is the expert in the room, and everybody on the team has access to it, so that a writer who isn’t, for instance, a Warhammer aficionado can upload the quest they’ve designed, and have the system tell them whether it’s accurate or not. It can say what details might need to change. Or tell me what portion of the 137 books of Warhammer that this applies to! And it can do so with fidelity and accuracy in a way that was never possible before.
Please tell us about yourself and how it is that you came to get started on this. How did you identify this problem?
I’ve been making games now since 1999, and I’ve worked on many games that have failed and some that have been wild successes!
Many of them have been IP adaptations. I’ve been in the room in those difficult moments where there’s been frustration, or people who didn’t have the expertise were making decisions and pulling the story off the rails. I’ve seen one of the most beloved board game franchises being taken in a direction that was so misaligned with canon and with the board game experience that we had to reboot the entire thing, almost from zero, in order to make the breakout hit it became. I’ve seen this time and time again.
I was frustrated with my experience at a major publisher because there was resistance to some of the work I was doing there. I’m a UX designer, so my passion is the player and making sure that the player resonates with the content. IP is a very fast track to get amazingly connected and engaged audiences.
When I decided to go entrepreneurial, I looked at my bookshelf – I was a great reader before I had children! When I had time, I read voraciously. I saw all the books I had by Ray Bradbury, Robert Heinlein, David Eddings, Tolkien, Anne McCaffrey, Jennifer Roberson, Melanie Rawn, Mercedes Lackey… and (especially with the female authors) I worried these books are never going to get adapted. It’s too hard. You hand a book to somebody to read, and typically they either weren’t a reader to begin with, or they might not actually read it but claim they did, and then they’ll decide it shouldn’t be adapted. We’re going to lose some of the greatest stories ever told.
And now, with an AI future where stories will be diluted, and entertainment experiences become more and more generic, that’s going to be the default. IP is going to matter. Cohesive, quality storytelling is going to matter much more. I think IP is actually the next great asset class in the information age. AI gives us the superpower to codify it, to turn IP into software, and that’s what I wanted to build.
How do you make a great IP world model that can be replicated across transmedia, whether that’s games, a webcomic, a new television series, or a streaming pitch for Netflix? We were able to build a system to do that effectively, and the only reason we could was because we had prior content knowledge around World of Warcraft. My co-founder, Bryan Winge, and I worked together on World of Warcraft and interacted daily with the database system that drives that world. So the design of our system came from asking: what if we could take a book and modulate it into a world model, the same way we ran WoW? When we did that, it clicked. It became magic.
Are you working with any specific well-known IPs right now? Can you talk about them?
We’ve signed with six specific IPs. We are interested in working with properties in the public domain, too, and giving them a way to maintain the coherence and consistency of their brands and carry them into the future.
“I know there’s a deep well of content just begging to be adapted, that wouldn’t be possible at all without a technology like this. We want to turn beloved books into living, breathing worlds, and that’s the ultimate end goal”
Irena Pereira
The first author I can publicly mention is Jenn Lyons. She actually became a Sunday Times bestseller about two weeks ago for her latest novel, so she’s now an international bestseller! We used her work as a proof point for the capabilities of our system. We ingested her book, Ruin of Kings, and in two months, we made a game experience and a three-episode webcomic that tells the story of how one of her characters gets trapped in a dungeon. The game experience is about playing to escape.
What’s remarkable is that nobody on the teams had read the book. Nobody except Jenn herself, who was approving everything. The system wrote the dialogue, described the rooms and locations within the world, identified the character’s motivations, and even identified which character would be the best fit for the adaptation in the first place. It was able to help us parse the books in a way that someone with zero prior knowledge could essentially have a dynamic “CliffsNotes” of everything necessary for the adaptation.
We are also working with RA Salvatore on his Demon Wars IP. We have a mandate to drive profits toward charities, and we think there’s a wonderful game in there. But he is a busy man, and we can’t have him 24/7. Our instance of Demon Wars is there - available - in the meantime.
Tell us about the technology stack. What is the core LLM you’re using, and what choices are you making there?
Currently, we’re working in Gemini on GCP [Google Cloud Platform], because the larger the context window, the better. It matters that we ingest multiple books at once, though we can also do it piecemeal for larger loads.
Our system isn’t that heavy in terms of compute cost, because we only process the content once in its pure form. When we assign it to our world model, we’re working with pieces of it. We designed the system to minimise unnecessary exposure of source material and to keep rights-holder content protected.
That approach also means we can have runtime engagement with our world model. It can deliver dialogue in the way we’ve always wanted: think of active NPCs within a game that can communicate dynamically about world events. You can do that with our system at a fraction of the cost, without having to upload or process massive amounts of data all at once.
[We have a] deployment for public domain works, and my favourite to work with is Peter Pan. It’s owned by Great Ormond Street Hospital, and we’ve donated use of our platform to support their efforts to activate the IP and raise money for their children’s cancer centre. I’m fully behind any efforts there.
We break down the content: by chapters, by timeline. What happens on page one: they’re in London, then they travel to Neverland, then they come back to London. Off-the-shelf AI doesn’t have a sense of temporal action. It doesn’t know the difference between page five and page 500, so we have to install that understanding.
I tried to use AI to help me write a crime story once, and it just told me who the murderer was in the second paragraph! Not because I’d decided yet, but because it made the assumption that the obvious suspect was the villain, regardless of whether he’d be revealed much later in the story. Our system addresses exactly that kind of problem.
It can break down artefacts, characters, locations, and organisations. And one of my favourite things it does is identify the meta-themes of the work. That’s something that requires a deep understanding of the content to glean, and our system surfaces it automatically.
“In the kind of AI-assisted world I imagine, you don’t want it to be perfect – you still want a human on the other side to fix it, curate it, make sure it’s appropriate. But you want to solve the blank page problem. It enables teams to move a hundred times faster”
Irena Pereira
We can output certain default reports, some of them like academic reports. What are the lost MacGuffins, who are the “lower decks” characters? What are the moments that happen off-camera that might be an opportunity to develop into a side story or webcomic script, or a premium streaming pitch for Netflix? Which parts might suit a 10-page children’s picture book or a cinematic trailer? We have already preset some suggestions, like default outputs.
Your platform is described as “canon safe” – that’s got to be important, because creators will be concerned about where LLMs can be inconsistent, if they hallucinate, if they go off-brand. How do you enforce tone and ensure it doesn’t get off track?
It just does it by default, and what’s incredible is that it can maintain the voice of the author as it processes the content. Let me give you a direct example – this is my party trick. I can ask it to give me 25 insults from Peter Pan to Captain Hook, categorised by his insecurities. It takes a little while because what it’s doing is collecting everything it knows about Peter Pan (all the page references, everything it knows about Captain Hook), grabbing those snippets to understand the relationship and then producing the output. It does that every single time, and that’s exactly why we work that way: it maintains fidelity.
But then my other favourite prompt, if you want to modulate the tone of the brand – I could say, “Give me 50 insults from Peter Pan to Captain Hook, but make it Gen Z.” Here’s what came out: “Your whole crew is just a bunch of NPCs. This ship is not the vibe – needs a deep clean, low key.”
We also have the Iliad and the Odyssey loaded into our system. One of my favourite authors (someone we’re currently in contract negotiation with) loves the Odyssey on an academic level. I asked him to come up with a question for the AI, and he said: “What if the gods aren’t real?” And it output this collegiate-level dissertation. It was incredible.
It has the concept of voice, but still applies it to the content faithfully. And in the kind of AI-assisted world I imagine, you don’t want it to be perfect – you still want a human on the other side to fix it, curate it, make sure it’s appropriate. But you want to solve the blank page problem. It enables teams to move a hundred times faster. That’s why we were able to build a game in eight weeks: we knew exactly what the character was, what the location was, what the villain’s motivations were, because we had the context of the books in our system and could just query it directly. It gives page references better than any collegiate student I know!
Let’s look at a big topic: the future of storytelling. How important do you think transmedia and IP are going to be to the industry?
I think transmedia is the future – and doing it in a way where you can have parallel execution across mediums is where the IP play really matters. You need some kind of organisation or infrastructure built around the IP.
Take KPop Demon Hunters as an example. Massive sleeper hit! Nobody saw it coming. They’d been working on it for years, and then it just went gangbusters. But then everything went quiet after release for maybe four to six months. Now, a year later, there’s a McDonald’s campaign, which is cute, but it hasn’t had the cultural draw or impact they were aiming for. Why? Because they took too long.
“Our system doesn’t necessarily use AI the way you typically think about AI. AI is a companion to the technology. We use AI as a processor, to comb through the pages of the book and transform source material into a structured, queryable world model”
Irena Pereira
Fast-follows matter. You need a coordinated ramp to capture that zeitgeist and ride it like the wave it is. With AI, there are no excuses anymore; it’s just a matter of how quickly you can catch that wave, because if you don’t, it peters out and becomes a one-hit wonder. But if you do, it can stay at the top of the charts and spawn TV series, spin-offs, comic books, board games, video games. And you can de-risk a lot of that far more than before.
If I can build a game in eight to ten weeks, if I can build a webcomic in six, there’s no excuse for not starting now. And we did it all for under $200,000, hiring professional teams who executed beautifully. The capability is so powerful. And you can do it in a way that’s consumer-friendly around AI: our AI isn’t generating the art; it’s accelerating your team and reducing the communication and approval lag.
You mentioned that the platform is a helper. As well as helping people understand the IP, does it also, for instance, write scripts or create images for you?
I believe strongly in AI not replacing human developers. Our system right now is intentionally text-only.
We still encounter resistance from certain dev teams: there’s sadness because their narrative designer or game design lead, the person who’s the expert, who loves the property they’re working on, is being replaced. Quite literally, in some cases. But here’s the thing: that person is a bottleneck. So how do you reconcile that?
What you can do is essentially operationalise their knowledge base into a piece of software that other people can access. That person is still there, still necessary: they set the direction, the tone, they navigate the team toward the product vision. But they do so from a position of genuine expertise, and at the same time, they democratise that expertise to all 50, 60, 200 team members in a way they simply couldn’t do physically themselves.
“IP is going to matter. Cohesive, quality storytelling is going to matter much more. I think IP is actually the next great asset class in the information age. AI gives us the superpower to codify it, to turn IP into software, and that’s what I wanted to build”
Irena Pereira
Our intention is to leave the tools and the execution layer up to the game teams. How they output is up to them. That said, we are talking to some partners about API hooks to drive things like in-game dialogue. And what does that do for the product? It makes it more reliably faithful to the source material, in a way that human opinion, personal perspective, and the desire to put your own stamp on things can sometimes get in the way of.
I know there’s a deep well of content just begging to be adapted, that wouldn’t be possible at all without a technology like this. Our first tagline was: we want to turn beloved books into living, breathing worlds, and that’s the ultimate end goal.
Further down the rabbit hole
Some useful news, views and links to keep you going until next time:
Last week, Anthropic launched Fable 5, its state-of-the-art Mythos-class model… and took it down again by the end of the week under pressure from the US government.
Korean AI lab Miso AI released Miso One, an open-source text-to-speech model that promises more expressive speech than anything currently available, with 110ms latency and one-shot voice cloning. Model weights are already available, with API access coming soon.
HoYoverse, the studio behind Genshin Impact and Honkai: Star Rail, plans to invest up to $14.6 billion in AI over the next three years, with a focus on building its own in-house AI ecosystem rather than relying on external models. The tech will feed into NPC systems, content generation, and live-service tools — with upcoming life sim Petit Planet the first title to feature AI-powered NPCs.
Saga, which we featured last year, has sold off its blockchain network and relaunched as Saga AI Labs, doubling down on AI-powered digital characters and autonomous agents.
Artificial Agency, the generative behaviour engine we featured in the newsletter last year, has hired Greg Canessa as COO. Canessa brings 25 years of senior experience at Activision Blizzard, Microsoft Xbox, Sony and Google, and will focus on scaling studio adoption of the company’s AI behaviour engine.
Barn Cleave, CEO of TILKI, a new game creation platform, has been discussing the company’s approach to AI-driven world-building: creators describe a world, and TILKI expands that into a fully structured canon of over a thousand pages which then runs as a living simulation. We’ll have a full conversation with Barn in the newsletter very soon.
X&Immersion has shut its doors. You may remember them as the AI character platform which was one of the first interviews AI Gamechangers posted.
Asha Sharma, CEO at Xbox, took the AI copilot out of consoles in her first 100 days because gamers just weren’t into it. Now you can watch an interview with her here in which she discusses it with Bloomberg’s Emily Chang.