Skip to content
AI Gamechangers· AI Gamechangers·· 2026-01-14AI score22

"What it’s allowed us to do is create something you couldn’t without AI"

"What it’s allowed us to do is create something you couldn’t without AI"

Article

Hello and welcome to the latest edition of AI Gamechangers, your newsletter where we quiz people across the games industry who are doing practical, inventive things with AI. It’s been a busy week as we prepare to attend PG Connects London on Monday and Tuesday, where there’s sure to be plenty of AI chat. Many of our recent interviewees will be speaking there in person, so grab your ticket now.

We’re joined in this edition by Susan and Lee Cummings, the veteran founders behind 10six Games. We dig into what it really means to build a “genAI-native” game: runtime character creation, just-in-time asset generation, and the design constraints you need if you want personalisation without breaking balance.

As ever, scroll to the end for a round-up of AI news, links and videos.

Susan and Lee Cummings, 10six Games

Lee Cummings (left) and Susan Cummings (right) are co-founders of 10six Games and Tiny Rebel, bringing together over 25 years of experience in game making, including key roles at Rockstar, 2K and more.

Meet Susan and Lee Cummings, co-founders of Wales-based studio 10six Games and long-time pioneers in games and emerging tech via Tiny Rebel.

Their upcoming game, You vs Zombies, is built around their Infinity Platform, which generates personalised characters, dialogue, and scenario content at runtime (in seconds) while staying within authored constraints for tone and balance.

In this interview, we explore just-in-time asset generation, the practical challenges of making runtime generation affordable, and their approach to ethics and creative ownership in an AI-driven pipeline.

Top takeaways from this conversation:

  1. In roguelike game You vs Zombies, personalisation happens at runtime, with characters, dialogue, VFX choices and scenarios generated in seconds.

  2. Constraints are the point, not a compromise: classes, templates and parameter limits are used to guide the AI and keep the game balanced.

  3. Just-in-time generation changes the content economics. Assets are created only when needed, then reused via a growing library of player-made scenarios.

  4. They’re considering ethics at the production stage, exploring ideas for compensating artists/writers/musicians in a “rock band” structure.

AI Gamechangers: You’ve had a great career in the cutting edge of games and tech. What was the lightbulb moment when you realised that the next step in your career was going to be working with AI?

Susan Cummings: I’ve been married to and working together with my husband, Lee, for 27 years. He’s the creative side of things, and where these ideas come from. Over the years at Tiny Rebel, I’ve seen that he loves new technology – and he gets me excited about new technology.

Someone said something very flattering to us: that we love to find the cool use cases, not the gimmicks. It’s really hard being at the cutting edge of things, because what’s fun doesn’t necessarily always work. But that’s what drives us. We were heavily involved in mobile, and then augmented reality and virtual reality, and a bit in web3. The AI side of things took off while we were in the middle of making the Petaverse.

I remember when we first got access to ChatGPT and the art applications like DALL-E. We started playing there, figuring out what that meant for games. When we decided to put this together, it was already in our heads that there were some profoundly interesting things happening. One of the most interesting things about AI is that it’s quickly become part of the tapestry of our lives, in a way that other technologies haven’t; and as much as there is a sort of hand-wringing about it, I don’t know anybody who’s not using Gemini on their phone or ChatGPT.

You vs Zombies is a hugely personalised roguelike action game. A “protect” level type has you defending an NPC from zombie attack. Many elements of the scenario are unique to each player and are created at runtime.

We had this idea about a year and a half ago: wouldn’t it be interesting if we could allow users to impact a game in meaningful ways? We talked about how you could have the core of a game, and as you move around the world, you could define new areas. User-generated content (UGC) is a defining characteristic of what works right now, whether it’s Roblox or Fortnite or Minecraft. People want the agency to create stuff!

So we started experimenting on our platform, and one of the things from our early experimentation was character creators and how they’ve evolved over the years. They all have internal biases, though. As a woman, I don’t look like Lara Croft. I don’t even want to! But time and time again, you start a male-led game, and you probably have all the associated stereotypes. Now imagine you come from a remote village in Asia – you’re not going to get a relevant experience at all.

We thought, “How do we use AI to change that?” We wanted to make the sort of game that you just simply couldn’t make with previous constraints. There are only so many ways a face can look, only so much knowledge of the world a game can have. But AI doesn’t have that constraint. That became really important to us. “How can you be YOU in a game, quickly?” We set out to prove how efficiently that could happen, and how much creativity could still be dictated by us. I don’t think AI is very creative. We set out to figure out what rules the AI would need so that we could create a fun, balanced, beautiful experience, and let it be malleable for all users. That was what inspired us, the fact that we could give the user agency over their experience.

By January, we had our first proof of concept, with our earliest art style. We have an artist on the team, and it’s his vision when it comes to what our characters look like. So this isn’t a case of us asking the AI to come up with an art style.

AI Gamechangers: Please talk us through the AI platform. What foundation models are you using, and how are you fine-tuning things?

Susan Cummings: We call it our Infinity platform, and the idea is that it mediates with a slew of different LLMs. And in some cases, we have multiple LLMs that we use for things; in other cases, we swap them out.

At the core of it, we use Gemini. We were using ChatGPT, but we recently migrated over to Gemini, which has worked really well for us. It’s been faster and cheaper, and more reliable than what we were getting out of ChatGPT (which Google really liked hearing!).

“Wouldn’t it be interesting if AI could be treated like a member of the development team? Imagine if the AI gave us tickets every week and then followed up.”
Susan Cummings

Underneath that, on the art side, we play a lot. We were using Flux, but now we’re using something called Z-Image, which is the bleeding edge.

You can see how we generate a character – I just give some information about you, tell it a bit about your appearance, some things you care about, hobbies, traits, rivals, things like that. You could put in the name of somebody famous, Donald Trump or Master Chief, something like that – the more well-known that person is, the better the AI will do without any guidance! But we’re hoping the charm is going to be in using you, the player: you describe yourself, and what you want to be wearing, your hair colour and so on.

Once you enter that information, we do our prompt engineering. The character art has been a work in progress over the course of last year: it starts with a custom LoRA [Low-Rank Adaptation module for fine-tuning] that our artist has worked with our engineer to train. And we need enough references and ethnicities and styles to work with, learning how to include a hijab, for example.

The character set-up phase takes your information - about yourself or anyone you can think of - and turns them into unique characters within the game’s style. Here you can see AI Gamechangers’ editor Dave Bradley preparing to fight zombies.

Our art style removes the neck and the legs – the characters sort of look like Rayman. We took that approach because it’s much more friendly in runtime. The characters are 2D, but the backgrounds are 3D, and so removing arms and legs from the body needs to happen consistently so that the user doesn’t have to choose between multiple things. This happens within seconds.

We have custom LoRAs for women, zombies, and more. It also chooses a boss fight (based on the profession or what else you give it; for instance, you’re a journalist, so the AI might make the ultimate boss villain Rupert Murdoch). It also chooses a class for your character, for example, a “speedster”. The reason we’ve created these archetype classes is so that we can help the AI balance out your stats. If we didn’t do that, it might give you 100% [critical hit] and break the game. So we’ve preset some templates because we want to balance the game. This is why I’m so sceptical of companies that are promising platforms where anybody can make a game from a few lines of text – it’s not going to be a balanced game! There are so many issues with that. So ours is on rails, including certain elemental abilities it can pick for you. It has certain abilities to change stats, like cooldown and duration and so forth. We can give it the constraints of how much it can adjust those things, and within what parameters. It picks a utility spell for you. It also picks primary and secondary colours for the VFX, so that gets customised on the fly as well.

We call this a user-generated scenario. And although things like these abilities are unique to this scenario, you might be able to collect them as loot, so when you’re playing another scenario later, you can use this again.

The AI has created war cries too: making hundreds of lines of dialogue in that initial 19 seconds. What do you say when you win? What do you say when you meet an NPC? What do you say to confirm character selection in the front end? What do you say when you encounter a boss? All are correct in character, and specific to our tone – Lee and our writer Finbar work together on tone, and we train the AI with knowledge there as well, using world information and the story bible.

“We’ve been told anecdotally by some big tech companies that no one’s ahead of where we are right now, which is really exciting. We built this nimble, little experimental company. We keep answering questions and coming up with new ones and that’s allowed us to be ahead of others”
Lee Cummings

We use Suno for custom music, and we use ElevenLabs for voice narrations, which are voice-directed by our writer, who previously worked for Aardman.

The story I really want to tell is the fact that we’ve made this with a team of eight people, and we have all the expertise – we could have made this without AI, and had a very rigid experience. We have the artists, we have the writing, we have the engineering. What the AI has allowed us to do is to extend that and create something you couldn’t possibly do without AI. A gamer couldn’t previously say, “I want to be me, and these are the things that matter to me”, and then 19 seconds later, be playing that.

AI Gamechangers: Can you tell us more about the specific game you’re creating using this platform?

Susan Cummings: You vs Zombies is a time-travelling game. In it, Ash and Morgan are the two main characters. It’s a post-apocalyptic world in which everyone other than you, Ash and Morgan is pretty much dead! You agree to sacrifice your physical form to try to change the timeline. Ash and Morgan are in the future by the portal. They’re talking to you from the future when you are outside of scenarios, and are going to be the ones helping you level up and helping you deal with the trauma of the apocalypse. When we create a character, Ash and Morgan have thoughts about it. They comment. Morgan is nicer than Ash, and they have dialogue.

Lee Cummings: We made really hard decisions and then wrote the story afterwards. We backed ourselves into a corner and then thought, “How do we explain all of this?” So we wanted a Quantum Leap game, so we came up with a Quantum Leap-style story.

Susan Cummings: We’ve created a RAG-based system that allows us to populate a database with the levels as they’re being generated on the fly. It will automatically create levels as I play through, and every time a level gets created, the AI knows that. So when you finally get to the boss fight on level eight, it’s going to reference stuff that happened along the way.

There’s a narration line that gets created, and it picks an objective for each level. The first level is always surviving for a period of time. It creates a line for you at the start and the end. And if you fail, then it gives lines for Ash and Morgan to encourage you to enter the map screens.

We’ve pre-created three biomes, and it picks one. This next one could be a horde level. Third comes a protect level, where you have NPCs that you have to protect before they die. And they have dialogue too! Sometimes, you have a level where you have to find something in the environment – that thing also gets generated in our art style by the AI, based on your character. The final level is always a boss fight. What does the boss say as you’re running around trying to kill him? What are his talents? What’s his death like? We can mix and match different stages like bullet hell and poison, more enemies and so on.

“One of the things that’s been really important to us is trying to ask what the economy looks like going forward for freelance. How do we emulate how a rock band works? Musicians come together, and sometimes they even have multiple bands that they’re a part of. That’s interesting for games. Currently, people are siloed behind companies”
Susan Cummings

Every single person who plays this is going to have a completely different experience, in all likelihood. You’re going to pick 10 different scenarios than I would. You’ll be able to share your scenarios, so you might also play something that somebody else made. For the first time, the meta game of this is going to be completely different for everybody. And so I think that’s really cool: there’s a structured story going on, a steel thread that pulls you through this, but you get to make it your own.

And they can share those experiences too. Beyond [the personalisation], there’s everything you’d expect from a game too. There’s a loot system, and there’s a level-up tree as you’d expect from any roguelike. There’s also a trauma system – because of the fact that you’re witnessing an apocalypse, we have a whole system of negative effects, as your character gets something like PTSD from playing through this game.

AI Gamechangers: What are the technical challenges you’ve had to overcome?

Susan Cummings: You have to ask questions that no one’s asked before! We keep coming across use cases and new problems, because the games industry has been kind of slow to adopt this, beyond office productivity. I think 99% of developers are using AI in some way, shape or form, in their pipeline – but in terms of runtime? There are things that we can’t do yet that we wish we could do.

On the art side, for example, we found that if we had one LoRA for all our characters, men were getting boobs no matter what we did. And so we found that we had to create both male and female LoRAs. We had weird issues where if you put a suit on a character, they always had arms, despite how many different ways we told it not to. Eventually, we got good enough that we could separate them and have it consistently stick with the same art style. Initially, we couldn’t get rid of the transparency in any cost-effective way (it was going to double the cost).

Obviously, the gen times and costs are big, big, big factors for us. And my platform engineer is awesome, and he’s done this tremendous job of refining those things over time. Part of that has been about building a platform that has flexibility so that when something new comes up, we can begin slot things in. We have two or three different services that we’ve used to host our models, to deal with costs and downtimes. That’s been a big area of work.

Lee Cummings: When I was a teenager, I worked on building sites. There was a big building site down in Cardiff, and I worked with the lead engineer there, and he showed me the stock they had of equipment. And he said, “Here are the four bolts we have for the whole of the site.” Why has he only got four bolts here? “As soon as I take one, I scan the barcode underneath it, and within 20 minutes, another one is being shipped from Europe over to me, so we only ever need four of each bolt of any size.” This was the amazing just-in-time infrastructure which was built before Brexit! Manufacturing started becoming liquid… and that’s what this does in the creation of assets. You can try to look at games through the angle of what happens when you are literally “just in time”. We’re creating assets when the level hits. When a player reaches the end of the level, we’re triggering assets for the next one. It takes seven seconds to get them, like a loading screen.

“I remember when we first got access to ChatGPT and the art applications like DALL-E. We started playing there, figuring out what that meant for games. One of the most interesting things about AI is that it’s quickly become part of the tapestry of our lives, in a way that other technologies haven’t”
Susan Cummings

How can games change? How can you fundamentally change how a player experiences a game? And the first way is we can make it hyper-customised. You can put in as much or as little as you want to say about yourself, and the system will extrapolate a personal, hyper-customised horror comedy story that you can play through. This is the evolution of when kids were given books in the ’80s, where the publishing company takes your kid’s name and their hair colour, and sends you a book for £25 where the hero is called Dave and has red hair. This is the modern version of that. It’s like Scribblenauts except that instead of hiring 100 artists to sit in a room with a dictionary for three years, we can now future-proof any asset requirements.

Susan Cummings: An important point is that we built everything with the future in mind. They say “great is the enemy of good”. Well, what is “good enough” right now that lets people have these experiences, so that when 3D catches up, for example, we can retrofit? The same with audio. We’d love every single line in this game to be voiced, but that would get really expensive and time-consuming. Eventually, though, probably within the next year, that’s going to become viable. And so this is all about us proving it with this. Everything we’re proving right now is valid. It’s changed so much in the last year. For a while, we thought a minute was great for generation, and then all of a sudden, it’s a few seconds!

AI Gamechangers: Does the just-in-time mentality help from a cost perspective, too? You’re not trying to generate an entire game right from the start, that the player might not see all of.

Susan Cummings: You’re kind of back to when PvP became a big thing. Going back 15 years, I remember we were trying to do a PvE zombie game, and I was getting a lot of pushback from people: “Why do you want to make a PvE game? You’ll have to make so much content. People tear through content so quickly! PvP is better because everyone’s making their own fun.” Here it’s similar: we don’t have to create every conceivable type of character scenario, because we’re letting the users create them. And then eventually, there’s a library of content out there, and scenarios to play if you’re a user who wants to play but not create (and most people really aren’t creators, they’re players; look at Roblox). Most people aren’t creating games. But eventually, you have a critical mass where anything you think of has probably already been created by somebody, and now the scenario already exists. Now, instead of 15 cents, it’s free for us, and it’s just reused content.

Lee Cummings: If one day the servers go down, you’ve already got a massive database of created characters.

Susan Cummings: This was a big initial driving point for Lee: I remember him talking about the Master Control Program in Tron. How do we how do we train up an MCP in everything we know about making a great game, and let a user talk to it directly? That becomes the end-of-life plan, as we suddenly don’t even have to be involved anymore. It knows what it’s doing.

AI Gamechangers: You’ve spoken about how AI can help with representation in character setup. What other ethical issues do we need to consider when talking about AI and games?

Susan Cummings: Lee and I have always had a reputation for trying to approach things ethically, whether that’s the free-to-play [monetisation model], or new tech as it appeared in our Petaverse.

“How can games change? How can you fundamentally change how a player experiences a game? And the first way is we can make it hyper-customised. You can put in as much or as little as you want to say about yourself, and the system will extrapolate a personal, hyper-customised horror comedy story that you can play through”
Lee Cummings

In this case, one of the first things I did as we put together our team of engineers was explain to our artist what we were setting out to do. I said, “I don’t know what this looks like for an artist. I want to make sure we’re not taking advantage of you, because it’s your art style.” In theory, we could do 10 more games on this platform, taking the art style, and what if he wasn’t involved in the company any more?

One of the things that’s been really important to us is trying to ask what the economy looks like going forward for freelance. That’s a big notion. And I think, “How do we emulate how a rock band works?” Musicians come together, and sometimes they even have multiple bands that they’re a part of. That’s interesting for games. Currently, people are siloed behind companies. If I really love someone’s art style, am I going to try and pull them out of EA? What if I don’t need them for the next project? It doesn’t make sense.

You want to be able to pull together creatives for the thing that you’re doing. That’s a benefit for me as CEO, but how do I make sure the rest of the band is benefiting from it? That’s been a big part of this. I don’t have all the answers, but we’re going to figure it out along the way. Does it mean we should set up an SPV [Special Purpose Vehicle] that everyone has a piece of, like stock? Or receives an underlying residual, like with film? Art is the first place where we’ve been trying to do that.

On the writing side, we’ve worked with Finbar Hawkins numerous times now. We are not replacing the need for a writer, because AI writing sucks if it doesn’t understand the story. The next one we’re cracking is how we think about music and musicians and lyrics and things like that. My platform engineer is also a musician, so we’re trying to really think through these areas and make sure we’re doing it in a way that’s sustainable.

The main thing is we’re trying to have honest conversations. I think it’s a much better conversation than simply, “Should we just not use AI?” Come on, it’s not going away. It’s really not – it’s too much a part of the fabric of our lives now, even outside of our jobs. It’s useful if we harness it.

AI Gamechangers: What do you think other game-makers are currently either overestimating or underestimating about AI? Do you have advice for other studios?

Lee Cummings: I think people would be a lot less freaked out if we just talked about it on the maths level. Broadly, it’s the ability to query a massive data set – I know that sounds really boring, but it can do some magical things. I think people need to be realistic about what it can do by understanding how it works a little bit better – that it’s not creepy.

Susan Cummings: I’ve watched how much our engineer has gotten really good at it and how much he enjoys doing it. It was a creative process with him and the artist. I had said, “Why don’t we take some of the results that we’ve liked and feed them back in as more reference?” That didn’t work – it started to dilute the art style, and it moved away from what we had originally set out. It was quite realistic looking, but it wasn’t what we wanted. So we had to go back and hand-create all the reference images the way we wanted them to be. That was a really interesting process. It becomes a very yin and yang thing between the programmer and the artist to get what they want.

Lee Cummings: When you’re working in new tech, there are unknown unknowns everywhere! You need a veteran team which works well together, people who will start iterating on their own, because they bump into weird bits and pieces. We needed to see it as a research project when we started the game. We didn’t know what we were building. We wanted to make a roguelike action game, because that’s what we love.

“You can try to look at games through the angle of what happens when you are literally ‘just in time’. We’re creating assets when the level hits. When a player reaches the end of the level, we’re triggering assets for the next one. It takes seven seconds to get them”
Lee Cummings

We didn’t go in and force a game design onto some tech. We designed the game as we built the really cool stuff to back it up. We went and said, “What fun can we have here? How can we let people express themselves? How can they start seeing themselves in a game?”

Susan Cummings: That’s one of the reasons it’s worked. We put a team together, just eight people who are all very familiar with each other. We’re all very experienced, and we love what we’re doing. There was no question of “Should we use AI?” That’s the reason we started the project! I was on a panel at the Wiggin event [in December]. One of the questions to the panel was, “What was the reaction from your team when you told them they were going to have to use AI?” Well, that’s why we set up the company! And because we’re so small, and we’ve made the process so lean, we’ve unpicked the process.

Lee Cummings: When Tiny Rebel got up to 25 people, people weren’t at every meeting. Now we’ve got a lot to decide, and we’re back to a place where everyone’s at every meeting, largely. It means we can have a hive mind about some of this stuff – we brainstorm, and then almost immediately, we can get a version of that. A day later, I’m seeing it. That’s been a superpower for us.

We’ve been told anecdotally by some big tech companies that no one’s ahead of where we are right now, which is really exciting. I couldn’t decide if that means we’re idiots, or if it’s actually because we built this nimble, little experimental company (I hope the latter is true). We didn’t have a VC or a bigger company telling us what to focus on. Instead, we keep answering these questions and coming up with new ones. And I think that’s allowed us to be ahead of others.

AI Gamechangers: What stage is the game at right now, and what’s your plan for launch?

Susan Cummings: We are starting our private testing – invite only – right now. We’re just getting our Steam page together; we’re going to start wishlisting soon. But the game is totally playable. You can go in right now, create a scenario, play through to the boss fight, and then go back to play more. There are certain things that aren’t in place yet, so it feels like a ’90s video game right now!

Here, the Dave Bradley character is simply tasked with surviving the level. Every run ends with a boss the game invents specifically for you, based on what you told it about yourself.

Lee Cummings: We’ll add skill trees and game levelling up and loops and all the balance, but the core of moving around and shooting, and the story, and the platform that generates assets in real time, it’s there.

Susan Cummings: So now it’s a lot of push and balancing and getting those last core systems into place. But enough of the game is there that we can realistically say, “Yes, we’re launching this this year.”

AI Gamechangers: What’s your vision for the technology’s future? If we speak again in three years, where do you see this kind of game going?

Lee Cummings: In the immediate sense, there are some features that we’re looking at. The real-time aspect of this is really interesting. Our prototype is built around the idea that our game is about a hero (you) versus a villain. In popular culture, there are heroes and villains all the time. Think of Kendrick Lamar and Drake, for instance. There’s conflict everywhere.

Imagine that when a new conflict breaks out in the public eye, our platform works out who the two sides are and then spins up a bespoke adventure for each side, where the other one’s a villain. It can act as a sort of an oracle on what’s going on in the real world, and the game can start reflecting the real world.

When we did our Doctor Who game two years ago, when a new episode [of the TV show] came out, eight hours later, we put out a new level with all the new characters. But we had the scripts in advance, and that would take months to get ready – it was exhausting to do that. Now, the same thing can be done quickly, where games can reflect the real world in interesting ways.

Susan Cummings: Figuring out co-op play and multiplayer is something that’ll be important to us this year.

And one of the next things that we want to do, once this is out there, is start building game two! The next game we want to do, which may well be in the same universe, will extend the platform further. We want to think about the backgrounds and other systems we can build because all of that makes our platform more robust.

“It chooses a boss fight (based on the profession or what else you give it). It also chooses a class for your character. The reason we’ve created these archetype classes is so that we can help the AI balance out your stats”
Susan Cummings

As soon as you try to make a second game on it, you have to start thinking about agnosticising it – so we haven’t decided if we’ll ever make this a platform other developers can use too. First, we have to have a successful game, because no one’s going to care what we build unless it’s fun. A SaaS business is so different from being a content creator. But I do think there’s something very interesting here that could be useful if it’s agnosticised. As a small, eight-person company, we have to pick our battles carefully.

Lee Cummings: Further, I think we’re just scratching the surface of what LLMs can do when it comes to games. There’s an interesting thought experiment. LLMs are great at looking at massive data sets and finding weird correlations and causations that humans just aren’t looking for. What happens when everything you do in a game is all a data set exposed to an LLM? What do you find about how people play through your game? How can you have a self-optimised system based on how people are actually playing in real time, based on the real data?

Susan Cummings: Wouldn’t it be interesting if AI could be treated like a member of the development team? Imagine if the AI gave us tickets every week and then followed up. “These are two problems we’ve seen; now you can fix them within the game.”

Subscribe now! It’s free, and you’ll receive interviews with industry insiders and studio leaders in your inbox. Next up: Kevin Lee, CEO of game creation and sharing platform Verse8.

Further down the rabbit hole

Some useful news, views and links to keep you going until next time:

Source: AI Gamechangers · aigamechangers.substack.com

PCIndieEngines and tools#Opinion#Indie#PC#Engine / tools