Heroza Zhang of Gorest: “LLMs Will Never Make the Next Elden Ring” — Building the AI-Native Game Engine 18VC Podcast · Episode 15 · July 21, 2026 Guests: Heroza Zhang Source: https://www.18-vc.com/podcast/heroza-zhang-gorest Host: It's an absolute pleasure to meet you here, Heroza. Would you mind doing a brief introduction to our audience about who you are and what you're building? Heroza Zhang: Yeah, sure. Hi, guys. I'm Heroza, co-founder and CEO of Gorest. So we are building a multi-agent for anyone to create interactive AAA games, which is a kind of really high quality and 3D result, and interactive and playable world. Host: And I do see that your first experience listed on LinkedIn was actually already running a video game community, right? Developer community. And then you go to the Perfect World, I think it's another game studio- for your internships. We're just curious, how did you get into the game and game development world? How did you start? Heroza Zhang: Okay. So I think it's back to my initial vision of my life targets. Yeah. Because I would like to understand everything, not only about the theory of how everything's rotated, and also how everyone feels in their mind or something, like almost everything. Yeah, I'm really curious and want to understand at the very beginning. And I was thinking about maybe I can simulate the whole world sometime to understand, to predict the future and understand the past. That is my initial vision. And then I was deep passionate about creating. Like creating stories or drawing or movies or even games. And I find also that game is kind of interactive world. So they are good combination. And I think by making a platform for anyone to make games, to make an interactive world from their mind, so I can not only doing things I'm interested in, and also I can finish my target of understand everything. Host: I think you talk about you want to understand not only the universe, but also every people you meet along the way, right? Heroza Zhang: Right. Host: I'm very curious about the drive behind that statement. Or what prompted you? Or is there anything, any people, any event that kind of shaped how you developed that curiosity? Heroza Zhang: I think... Let me think about this. So, it's more like physically or ph- Host: Yeah, or that, let me-- In other words, what about understanding people? What's the part about understanding people that fascinated you the most? Heroza Zhang: Okay. So I will explain in that way. So, I think for people, understanding other people. So it's more like to not only know why they are maybe talking to you and act in that way, but also know other people and also myself about who I am and why I'm here and where I am. Sometimes maybe people question themselves every day when they wake up from their bed. And, but actually, so maybe everyone has their own identity, like which school they are from or which background they are from. But- Host: For sure Heroza Zhang: yeah. But that is just actually a name or like... And they're just like an NPC from a game. They were just put there. But actually, what is the reason? What is the meaning behind that? I'm not sure, and I guess that most of the people cannot answer this question really frankly. Yeah, so I think that is a part I would like to understand. I mean, about myself- I want to understand why I was put here. Am I an NPC or am I someone? Yeah, so it's kind of like- Host: Are you an NPC? Heroza Zhang: things like that. Host: That's a great question. Heroza Zhang: Yeah, maybe. Yeah. And also, understand why other people act in that way. Host: Right. Heroza Zhang: And why they are put there. But I think if I want to really understand why I'm here- the better way is to understand the whole world. Because, for example, you can't fully understand an NPC from a game- Host: Right Heroza Zhang: because you know there's which kind of game is it by like you-- So you know which position the NPC is put. But in the real world, people are limited by their knowledge and their perspective. Host: Absolutely. Heroza Zhang: But if people learning more, understanding more, and even the whole thing, they will be 100% sure what they are. So it's actually this kind of thing, so like really philosophy. Host: Interesting. And just initially, you enjoy more playing games or- actually developing games? Heroza Zhang: Uh- Host: How do they... Yeah. Heroza Zhang: So, for me, game is not something to make people find, like to play together. For me, game is like interactive world. And for people who is creating the games, so they are not actually showing themselves to their audience. Or maybe some creators think they are compelling with their audience. Like- I want to show my player this thing, and my player need to get over this level, and maybe sometimes they finally understand what I want to show to them. There's also another kind of design. And actually, so for me, I was thinking no matter what the author want to show to the audience- they are actually showing their perspective. They see the whole world. So yeah. So game is actually an interactive world from different people perspective. And if you want to fully understand a person- maybe even though he or she say like really close to you or talking about the things in their mind, but you might not really care about it, or you might not fully understand it. But when you really get into their perspective, you know how exactly how they view the world, then you can fully finally understand them. Or it's another example. So I used to play a game which the main character is blind. So, I mean, the whole screen is black. Host: Oh. Heroza Zhang: Yeah. And you can only mimic. You are using a stick to touch something, and there will be the white line just appear. So you can use it this way to feel, oh, is here a block or is here a table? Something like that. Yeah, so that was a- game really impressed me because I really can understand what blind people experience. Yeah. So wait, I forgot about what is the initial question, actually. Host: Oh yeah, I was just asking, do you feel attracted more to playing games or developing games? Which are you more attracted to? Heroza Zhang: Okay. So get back to it. So my interest is about creating itself. But not creating just some games for people to play or something. I'm more like creating a story, creating an interactive world with average people's stories and perspective, just like a real world. Like combined together and how they work and how the worlds work. So that is thing I'm enjoying about. And yeah, and about playing, I'm not just using my spare time to try to pursue something in a game. I'm more like, okay, so I want to understand this perspective, and I just go into it. Yeah. Host: Interesting. I assume maybe you would not prefer just more commercialized games. Where players are kind of- Heroza Zhang: You pay to win, huh? Host: Yeah, you pay to win. They're just encouraged to pay more or maybe do this ranking. Maybe you're not in that type of game. Heroza Zhang: I'm more like enjoying adventure or narrative games. Host: Okay. Yeah, I can see that. Heroza Zhang: Yeah, but at the same time, I also playing chess. Host: Okay. Yeah. Yeah, so you're definitely competitive- Competitive... for sure. Yeah. Interesting. Yeah. And I think the next question will be, I do remember there was in your LinkedIn post, you mentioned you do your graduate project, which maybe finally turned into Gorest. Is that the story there? Heroza Zhang: Yeah. So when I was pursuing my master degree in Cornell, so just one year, and I put my capstone as a first semester. Yeah, so the capstone is about researching the AI in gaming. So that was the beginning into AutoFAB. And that year, so the AI is not as popular as this year in game industry. And almost 70% of people in game industry hates AI. Yeah. Especially the generative AI. And at that period, there is still a kind of practice or exploration of how AI, like the play mechanism, can be combined in games. So there's some famous one, maybe like One Night- or like AI to Use, or those kind of games. Yeah, so it's just the beginning. And yeah, they're trying to find if they put AI mechanism inside of game, will people feel more different, like feeling than we thought. Because yeah, so actually there is different kind of thoughts, because people may thinking if they put... So we want to experience, for example, when we are playing a game, we want to make our decision really affect the world. And in the past, the designer will design a whole thing about different choice. And make the player thinking their action really affects the result. Host: Yeah. Heroza Zhang: But actually, it is not. It is actually still part of the training. Host: Yeah. It's already programmed. Heroza Zhang: Yeah. So now people are making games. Host: Right. Heroza Zhang: Are trying to make this training to be not be limited by the training. But AI can generate more things and more possibility. And that is a initial thoughts. But the thing is, no matter how many training is it, there's still one line the protagonist or the player will experience. So if it is made by human, they will just try their best to make this line to be as good as enough. But if this line is made by AI, then nobody can make sure the experience is good. Host: For sure. Heroza Zhang: Yeah, so maybe the player did not experience the best part made by the designer, but actually a random mix- made by AI. And also people are trying to making some AI combining product. Yeah, like to maybe talk with the NPC and get their feedback to trying to gathering more attention or any kind of things. And there are many player just reflects the AI part does not make the game to be more interesting than we thought. So I think there is still a part people need to explore about how to making a better- Host: For sure Heroza Zhang: connection. Yeah. But I think for those game are currently famous. They are famous not because it's an AI mechanism, but actually still because their designer have a good talent and really use their heart to put it inside of the games then. Host: It's because they're fun. Heroza Zhang: Yeah. Host: I mean, essentially, if you create games, you want the games to be fun and be playable by- people. Heroza Zhang: Yeah. Host: That's the only way where you can get an immense fan base, right? I think people sometimes think too much about how you can make this game better or more something else, which is kind of not irrelevant to how the game mechanics can actually make it fun. And Heroza, you seem to have a lot of identities and roles, right? You are a novelist, and you are a design narrator. You're engineer. And then you're also a Xiangqi player. I mean, a Xiangqi grandmaster, I would say, right? Heroza Zhang: Yeah. Host: A designer create worlds. Xiangqi create orders. And you also need an engine. Well, you also need engineers that create an engine to support both, right? Out of all those different tendencies that you have, which disciplines do you feel like most directly shaped how you want to build Gorest? Heroza Zhang: Okay. So I think as you mentioned, Xiangqi is a kind of order, so it's more logically. Host: Right. Heroza Zhang: And design narrative stories, it's more emotionally. So, like- Just to make, because what is a word is combined of 100... Half percent of emotion- I'm sorry, and half percent of the logic. Yeah, and I think that shapes my mind. So, yeah. So I use chess to practice my way to thinking logically, and I also using stories to create, to understand about how people react emotionally. Yeah. And I think that will be a best combination- Host: Right Heroza Zhang: of how we define the whole world, and that will be shape about how we should make interactive world. And I think once we fully understand how the world should be made of- Host: Right Heroza Zhang: then when we create a platform for anyone to make something, we are the best person who know how designer, how creators- want to make it and what is the most important element in that. So, it'll be not limited by maybe if I just know programming, then I might- Host: Absolutely Heroza Zhang: I might just build some blog which is black and maybe the UI did not have good taste, and just put them together and say, "Okay, so it has a great design of the coding structure," but that is not really workable for the current era because people are more cherished about the knowhow of about how the people who are- Host: Yeah Heroza Zhang: building games is. Host: Yeah. Heroza Zhang: So, yeah. Host: Has it happened before that your designer persona fight against your, the- Engineer engineer persona and they try to reach a compromise? Heroza Zhang: Do you guys really think I'm an engineer? Host: Well, more about like the older maintainer, right? For example, if the... Regarding gameplay game, mechanics game logic, and you're saying that we should have an order here, but then the designer have inside her heart it's like, "No, we should make it more artistic. Make it more abstract so that people- Heroza Zhang: Oh Host: can enjoy it from there," right? Has that ever happened in the past? Heroza Zhang: So, when everything my game studio with my... Yeah, with other people, they already experienced and tell. Yeah, I was exterior. Yeah, so I was the narrative designer. Yeah. I design the stories. I was not programming it or... Yeah. So it's, I think, it's more about the limitation of time. Because sometimes I might create a big world and- Host: Right Heroza Zhang: stories, and I give it to my teams, and my team say, "The design's too big to be shown in the- Host: Right Heroza Zhang: in the game," and people might... We just have two weeks to finish it- and launch. It's limited by maybe 50 minutes, so I have to cut my design, I have to cut my stories, and I think that's why we need something like Gorest because, like for AAA games, for example- it used to take a lot of time, a lot of cost, and a lot of... Yeah, so it's actually be impossible for small team or individual to make a big enough games and to show- every ideas of their mind. But now, we have the potential and ability to make everything outside of our mind. So, I think the total quality of game will be improved because the idea will not be- limited by technical barrier. Yeah. Host: There's a question in a little later that we can dig deeper into that. Yeah. And talking about this founder journey, I think it's not a very typical path of career pursuit, especially for, I would say, Chinese or Asian families. Is there any people or any event that kind of inspire you or encourage you to become a founder one day? Or, what was the story behind? Yeah. Heroza Zhang: So, actually, when... I was wondering to become, doing my own... Building my own thing at the very beginning, but at that time I was thinking about, I'm not sure that if- that thing is beneficial. I think... Or, actually in my mind it's like if I can gather enough revenue or anything- I can put those money into something maybe not benefit enough, beneficial enough. But what is it? It's actually, I will collect every of the talented people I know in each of the field or industry be together and investigating which kind of things, is investigate their part. So, we can make this world simulator together to understand the whole world. But now, and till today, I find maybe actually I can combine this, my initial startup with this idea together and it is not actually, not non-beneficial. It is beneficial- or maybe... But it doesn't matter, yeah. I'm more curious about the long-term vision, and I think people are also excited about understand everything and predict everything. Yeah. But there's still an event which is really important for me. It was when I was in high school. Yeah, so I joined Hackathon. Yeah, this was 2020. Yeah. And as a leader, I don't... Because my undergraduate is computer science degree, but- before that, I have no experience of how to do programming. And, yeah. And people just asked me to join Hackathon, and I say, "I know nothing," but... I don't know what can I do. I don't even know... So, that was the emergence time of people having Figma. So- actually building a UI is also hard. So I just joined, and I find there's a good role, like for people without any experience to do that. It's team leader. I'm just kidding you. Actually, it's product manager. Host: Oh. Oh my God. Rip to all PM out there. Heroza Zhang: That's for me. For me. Yeah. I'm a product manager, so I'm just joking to myself, so doesn't mind. Yeah. So, yeah, actually product manager also need to learn a lot of things, but yeah- Host: Yeah. Yeah Heroza Zhang: at that time, I think it's easier- Host: Yeah Heroza Zhang: than learning programmer until today. So- So I still need to learn, but- Host: Yeah Heroza Zhang: and anyway, so I just trying to learn and things that period about how to draw Gantt diagram, or how to plan and how to design a product and assign it to my teammates- Host: Right Heroza Zhang: something like that. And at that time, I find if I want to add a feature, every time I need to communicate with my programmer, I need to understand whether this feature going to work. And I think it waste a lot of time, and I think a good programmer... Sorry, I think a good product manager should understand the technical part, or at least they can know whether the feature is workable or not. So, yeah, so it would be also an important quality. So, yeah. So product manager is not a role everyone can do. Yeah. But, yeah. So, yeah. So just ignore my joke. And yeah. And at that time, so even though I did not have any... I don't even know what is venture capital, but I just joined that hackathon. And yeah. And there's the event host called Tanner. Yeah. Host: Ooh. Heroza Zhang: I think he was just graduated from Stanford's graduate business school, yeah? Host: Stanford GSB. Heroza Zhang: Yeah. And I think he's building his own startup in Silicon Valley. Host: Cool. Heroza Zhang: Yeah, so the event is, he hosted. And also, he invite many of his friends- from big company and also from maybe different startups, but those startup already did well, and also venture capital. Host: Right. Heroza Zhang: Yeah, so we have a chance to do the office hour. Host: Wow. Heroza Zhang: Yeah, and I join people's office hour, and I think I met my teacher there, and they just say... I think, yeah, because I'm just a high school student without any experience. Host: Oh, my. Heroza Zhang: Yeah, and I was in China, so I have no idea about how everything going. But I think when I joined that office hour and the guys just told, telling... He's really respects me about- even though I'm just a kid there. Yeah, and when I just telling him about the ideas, and he's still listening and give me some advice, and maybe seeing my products and he say, "Okay, so like good way," or, "Which direction is better?" Or otherwise. Yeah, so I think that was the time I feel... And I just looking about their career path, and I know what venture capital is- Host: Right Heroza Zhang: and what startup going. And I find how to design a career and how to find a good way to solve this problem and to make the things just used to being our mind be real. Host: Yeah. Heroza Zhang: So, yeah. And for the people I feel most impressive joining that hackathon. Yeah, I think the host, Tanner, teached me a lot, and also the people I met in the office hour, like he's building the- Host: Yeah Heroza Zhang: autonomous driving company right now, and yeah. Host: Yeah. That hackathon is your first impression of the entire venture and entrepreneurship industry. Heroza Zhang: Actually, yeah. Host: Yeah. Heroza Zhang: I used to also do founding some small group, for example, building content or- holding event. Host: Yeah. Heroza Zhang: Before joining that hackathon. Host: Right. Heroza Zhang: But that hackathon was actually the important part- taking me to the venture capital industry. Host: For sure. Heroza Zhang: And yeah. Host: Well, I mean, Heroza learned about venture capital back in high school. When did you learn about it? Oh, I learn about it- You learn about it just a year ago, right? One year ago. Yeah. Yeah. So it's important, guys. Yeah. Show up on hackathon. Yeah. I think I was still climbing a tree when I was in high school. I don't know there's hackathon. Yeah. Yeah. And just move on a little bit to your startup. For somebody who may never know, never used Unity or Unreal Engine, how would you describe Gorest to them? Heroza Zhang: I think our product is just for everyone who want to build games. So, if they never use Unity and Unreal, then that is a good segment. So we would just say, actually just everything from your... Bring the triple A kingdom from your mind. Host: Oh, wow. Heroza Zhang: And you will instruct your AI team, and you can just provide your ideas and you instruct. Yeah. That is a vision we are looking for. Host: That's a bold claim. I do want to follow up a little bit, because I remember you mentioned that... Well, I think you mentioned that you trying to build this for prosumers, right? Heroza Zhang: Yeah. Host: Instead of just the beginners. Heroza Zhang: Yeah. Host: So what exactly is the target population of Gorest? For example, is it indie developers, beginners, prosumers, or mature game students? Heroza Zhang: Yeah. So we begin from the small niche. We begin from the prosumer. Host: Okay. Heroza Zhang: Yeah. But there's a question, it's like the trends of every AI products, it's meant to be no barrier. Host: Okay. Heroza Zhang: Yeah, so anyone can understand it. So that would be the better strategy. For example, just plug it in. Host: I see. Heroza Zhang: Yeah, so that is actually the trends. So every tool which still have the barrier- will, I mean, slowly, maybe there's a possible- Host: Got it Heroza Zhang: just dying. So we have to just follow the trends. Host: Got it. Heroza Zhang: But actually, why we targeting to prosumer as a beginning, it's we wanted to make sure we are really improving the total quality of the industry- and to solving the old problem that people are... Ideas and are caught by the limitation of technical. So it's the same, and that is the reason why we want to make sure the prosumer industry to be good enough to be affecting- Host: Interesting. Heroza Zhang: Yeah. Host: So in one sentence, what is the exact problem this prosumer population you're targeting facing today most acutely? Heroza Zhang: Okay. So- Host: Just in one sentence. Heroza Zhang: Yeah. So, let me think about it. So I think it's more like, for the part which is repetitive or tedious or manually all be replaced and simply made, if people have the ideas from their mind, just be put out. Host: What's the kind of monetary and time cost for them to use their own ways? I think you mentioned example of people using Gorest spend five minutes instead of eight hours, if I remember correctly, or some- Heroza Zhang: Yeah Host: some analogy like that. Heroza Zhang: Yeah. Host: Could you kind of walk us through that, like how that's possible, basically? Heroza Zhang: Yeah. So, we know many people hate generative AI. Host: Right. Heroza Zhang: And, yeah. And we also know people does not like AI to do something people are good at. But when we are really interviewing people, we find that actually, because for example, I'm a good storyteller. Host: Yes. Heroza Zhang: I don't want AI to write stories for me. Host: Yes. Heroza Zhang: And if you are an animator, you don't want AI to do animate for me. But actually- Host: Yeah Heroza Zhang: you really be happy to let AI- Host: Do other parts, huh? Heroza Zhang: Yes. Every- Host: I don't care about other parts, just do the general rest. Heroza Zhang: Sometimes this is like that. So yeah, so that's why. And I was thinking about... Yeah. So that's why we are think about we should making sure every part should be done by... Should be workable. Yeah. So that's why we build this features and so we started from... We saw generative AI, but it's a really part, we saw the end creation, and everyone's really hate of. Host: Yes. Heroza Zhang: That's why we started from there. Host: Yes. Heroza Zhang: So, that is a part, no matter you are a designer or you are a programmer, you don't like what is... Actually, in game, or especially trilogy, it's called gray box. So what is gray box? Gray box in... So you can understand, it's like when you are watching Hollywood movies. Maybe the Zootopia- and there's a fox and a rabbit. But when they are first shooting by camera or by the storyboard, it is actually the final 3D model as elaborated as a fox and rabbit. They're actually just without any texture. They're just simple gray fox there, and they're just one of testing the shot sometimes. Yeah. So that is like in game industry, it's called gray box. It is when the game designer have the ideas, but they are not sure how long does this level takes and how they work and whether the gameplay is interesting enough. They will manually just create a gray box. So they will just a cube and a character- Host: Right Heroza Zhang: which is really simple and just put it behind inside or maybe, for example, testing whether the level or the, yeah, elevator is high enough- Host: So it's prototyping to see whether basically it fits the physical world, whether the enemy- it's going to appear, right? Well, if the game mechanic basically asks you to not see the enemy, and enemy has to hide behind a tree, then you can prototyping and actually figure out if you're able to see the enemy behind a tree. Things like that. Heroza Zhang: Yeah. There is something like that. So that is a part we are just helping... Because when game designer are doing that, they are normally using a top-down world map and- Host: Yeah Heroza Zhang: just make it out and manually edit in the engine or something. But now we can just allowing them to just put their sketch right here inside, and we can just directly create this gray box out and makes a really simple way for them to modify based on that. So that is how we save time and also, and maybe build the things they really want, right? Host: Right. I think for Unity and Unreal, they also have their in-house gray boxing tool. I think for Unity, it's called ProBuilder, and I think for Unreal, it's called BSP. What is something that Gorest let prosumers, which is your primary target, do that they cannot do by just adding AI assistant or just a few plugins to their existing tools? Heroza Zhang: Yeah. So actually, those two are more like, there's going to be an easier maybe UI for you to drag a box or something. That is how those two are work. But actually, it kind of belongs to- earlier stage. So just adding a block, but a block need physics. A block need interaction, need mimicking the character, and everything, any storyline. So it is actually not just adding a block or adding a box inside. So I think that is a difference. So for those plugin, it's actually just helping you to maybe build architecture, but we are actually helping you to really fuzzy build a really workable and interactive gray box. So I think that's the difference. Host: Interesting. Interactive. Could you maybe elaborate a little bit more on that part, maybe? Heroza Zhang: You mean like interactive? Host: Yeah, right. Still, I want to get deeper into the differentiation between this and just a few plugins in Unreal or Unity. Heroza Zhang: Yeah. So ProBuilder is actually for you to drag the box and create a box here. So that's it. But we are not only just help you to firstly, from your libraries design and just firstly generate the whole structure, and also your character is ready to play inside, to open the door inside, to interact with monster inside, so that everything is settled down. So, it's huge more than a simple box dragger. Yeah. So I think that's the difference. Host: Interesting. Okay. And what part of data do Gorest own in the workflow? Heroza Zhang: Data, you mean? Host: Yeah. For example, I'm just curious, an example would be... Because basically the Gorest will interpret the intent of the designers- and convert those intent, and sketches, and ideas into the prototype. Right? So for example, if I have my sketches in, but then the prototype out and I don't like the prototype, then I will make a feedback or I'll say, "I don't like it. This is not what I want." Then there could be, is there a self-correction mechanism that you feed the feedback from the users back into the software, and then it generate better and more tailored responses? Heroza Zhang: So I think the first thing is that whether the game is really interesting or not, it will be hard to define by AI. Host: Okay. Heroza Zhang: Or graded by AI. And creator will really hate that. And actually, there's a reason for people to hate that. But, if, for example, we want to create a level maze, but the level maze is blocked, and people actually cannot go out. There's no exit. Then we have a mechanism to alert them that there's no exit, so- Host: Yeah Heroza Zhang: you might need to try to make one. Host: Yeah. Heroza Zhang: Yeah. That is a part we are warning them about, but most of the time, if the level going to be almost 100% exactly of what the creator want to create, and maybe- Host: Okay. Heroza Zhang: Yeah, so it's not like random things. Host: I see. Heroza Zhang: Yeah. Host: I see what you're saying. Heroza Zhang: But if they try and maybe the level in their mind is not fun enough- Host: Right Heroza Zhang: they may need to simply update it, then you just do a simple modification, test it again. Yeah. Host: Okay. I see. Makes sense. And I do see in your previous LinkedIn posts, again, you mentioned that Gorest does precise mesh generation from sketches. No AI-generated assets, no black box, full copyright ownership. What's the specific distinction between, for example, AI-generated asset and the one you are making? Heroza Zhang: Because for AI generative assets, make a computer as example. So AI generate assets. So AI will generate a black computer- with, for example, the keyboards and your screen. And you can see what is shown on your screen. Yeah, but for scenes we are making, there's nothing on the screen, or there's no black out here. It's everything just gray. Or even- Host: Gray box. Heroza Zhang: Yeah. So, that's the differences because it'll be actually not generative AI. Or maybe it's more like two simple cube we put together or something. Yeah. Maybe if you, for example, you really want to connect with generative AI API, you want to make the gray box to be replaced by a- real generated result- then there'll be a problem people considering about whether this... Yeah, so generative study belongs to or something. Host: Right. Heroza Zhang: But actually for gray box, you don't really need to make it to be as real as the scenes you are looking at. So- Host: And you don't have color, so it's just like a- Heroza Zhang: Yeah Host: very- Prototype rough prototype. Heroza Zhang: Yeah. Host: And it just belongs to you. Heroza Zhang: If you can see, there are many posts on Twitter right now about people want to make the video generation to be controllable enough. Host: Yeah. Heroza Zhang: And they are... I think there's an example in my scenes, trying the shot in the blender, and recording it, and put it into video generation- Host: Okay Heroza Zhang: to make it to be more controllable. But actually for the shot in the blender, you'll see they're all gray. So actually, this can pan off important part and which need to be kept in the industry to make things. Yeah. Host: Interesting. You've also written that no LLM is able to make the next Elden Ring. Right? If prompts are not the answer, then where exactly should AI stop and human judgment come in? Heroza Zhang: Actually, so I think that's actually... So we make this sentence to be like, why prompts does not... Prompts means AI builds whole things for you. Host: It can never replace human taste- Heroza Zhang: Yeah Host: the understanding you said. Yeah. Heroza Zhang: I think that is what we want to express. Host: Yeah. Heroza Zhang: Yeah. Host: So how does that shape the product philosophy behind the Gorest? Heroza Zhang: So, I think first of all, we are really solving something. Before we making that- we interviews huge amounts of- people from game industry to find whether they are currently using AI, and whether their attitude towards AI, and what's their current workflow, and what is the most challenging part for them to create or make their game something. Host: Yeah. Heroza Zhang: So I think it's more like a weapon for the creators. So I think the saddest part for people is when people are saying pending work or something, they are saying maybe AI's work is better than human, but actually, there's actually many good ideas inside of humans' mind, but they just don't have time or don't have chance to make it out. And I think we are trying to make weapon for human creator to just release their ideas- and to make every good point out. Yeah. So that is the philosophy, but not just saying, "Okay, so there's no more AAA, and we are just prompt, and get everything work by AI, and one sentence to games, and this game will be huge, more interesting than human." No. So that is not the thing we are currently making. So we are more like-... creating something for a human to create something. Host: And in application, when you're in reality, how do you draw a line between having a tool amplify human creators instead of just replacing them? Heroza Zhang: So, I think, for example, when we are choosing the gray box targets- Host: Right Heroza Zhang: we are more... So that is part, actually, because dragging a box is really boring. Host: Yeah. Heroza Zhang: You know where you want to put. Host: Yeah. Heroza Zhang: But you have to just add in blocks. Yeah. You just need to manually build this by your hand. So, we just make the boring things to be not, and keep your ideas here. But now, we are asking, but it's not like, "Okay. Hi, I want to build a game," but I don't know what game. So I just tell AI, "I want to build a game," and AI gave me a whole design, and AI instructs, and I can now define whether it's interesting or that element is bad. So that is more like trying to replace. And also, especially when you are asking AI to grading the games for people. I think AI should not be the one to grading arts or something. And I think that is about the differences between replacing or for humans. So that is different. Host: And amplifying. Heroza Zhang: Yeah. Host: And Gorest still requires the input, right? For example, an idea or a sketch from the human creators. Does text usually work better, or a sketch usually work better, or a combination of both? For example- Heroza Zhang: I think combination. Host: Combination of both. Heroza Zhang: Yeah. Host: Okay. So, that's kind of the big thing that I want to talk about here, is that because you still require the inputs instead of... You still have to ask the human creators to submit a content or a sketch of the castle. For example, when I create a castle using Gorest, then I have to have my own idea of how the sketch of the castle looks like, and then I tell the Gorest, and the Gorest's going to finish the prototyping work for me. And it's not like I just ask Gorest to create a castle for me. Heroza Zhang: Actually- Host: Because that would just be replacing my idea, right? Heroza Zhang: Actually, I think it depends because if you only have the ideas for building a castle, then I think Gorest can make it for you. Host: Okay. Heroza Zhang: But if you have ideas of how a good castle looks like- Host: Okay Heroza Zhang: then that differentiate your castle than other people's. Host: Ah. That makes way more sense. Heroza Zhang: And if you have a good idea of the castle, then you can have a better castle than other people. Host: Okay. That makes way more sense. Let's talk about something serious. How do you make money? What's your business model? What population do you want them to pay for this software? Heroza Zhang: Okay. So we began from the professional industry. So we started from small MSA studio for game studio- Host: Right Heroza Zhang: and also some related studio who need to build in this kind of simulation. Yeah. And our business model is more like by different team or individual or- Host: Okay Heroza Zhang: especially focusing on B2B, but there's also enterprise, like different version, and for- Host: Okay Heroza Zhang: yearly or monthly subscription- Host: Subscription. Okay Heroza Zhang: and token. Host: And also by usage. Heroza Zhang: Yeah. Host: Got it. Heroza Zhang: And for the long term, we're going to applicable the advertisement. Host: Oh. Heroza Zhang: So it's kind of similar to the Unity- Host: Okay Heroza Zhang: if you can think about their business model. Yeah. That is for where- Host: Makes more sense. Heroza Zhang: Yeah. Host: Yeah. Yeah. And we are moving a little bit to the technical part a little bit, behind the magic beneath your product. Dissect the magic here. Yeah. And I think there was a Stanford project that's called Sketch to Gray Box. Yeah. I think it's around 2022. They basically propose you can draw a sketch on your paper or whatever, took a picture. It will generate a 3D asset for you. Are you using a similar technology, or what's making the difference here that maybe your product can become more commercializable, maybe more reliable product? Heroza Zhang: You mean about the technical feature, or...? Host: Yeah, kind of. Is there any difference between some of the previous technology and the one you are using right now? Is there something happen because of this AI tool? Because I think 2022 is before everything AI happens, right? And now, maybe some new things happen that maybe this whole thing is different, even though it looks like you're achieving some similar results. Heroza Zhang: Yeah. We used to be asked about this question. So there's a most distinguished part is world model or story model. And people saying, for example, video generation, or because they think why those world model generation will now replace your desk work. The traditional way of making games or something. And actually, there are many model company right now are trained their own model to simulate a visual result- of how they look at the whole world. But actually, no matter it is interactive or not, they are still model, and people currently really want a combination. They need an agent to combine those model together. And there'll be a simple 3D editor for everyone to understand, and then use that model to create things what they need. And that is a kind of agent we are currently making. Yeah. So we are not really comparing or afraid of those model generation improvement, but actually we are really acceptable. Also, some games generation model, like Claude or GPT is actually, we are just trying to using their talent- to make this agent to be really combined and work for human, because model is not designed for human, but the agent is. Yeah. Host: So basically, you are saying that even with all these improvements in either Claude, ChatGPT, or maybe just these world models- they can help you basically improve your ability to create better prototypes for the game design, right? Heroza Zhang: Yeah. And we are currently building the structure for making those models to be reliably enough in our platform- and prepare earlier for that. Host: I think you also mentioned that, at least for this current version, Gorest doesn't use AI-generated assets, right? Heroza Zhang: So for current stages, currently, because AAA is really a long-term work, and we're starting from the gray box. Yeah. But I think- Host: Going forward, you will. Heroza Zhang: Yeah. Host: Yeah. Heroza Zhang: But there are going to be applications, people really need that- and there's also a way, like, we use this combination- in the future. Yeah. Host: Yeah. You can see Fable V, it's back, and then ChatGPT 5.6 Soul, right? It's coming up. Do you think your product's still going to be defensible after the next major foundational model update? Heroza Zhang: Defensible? So what do you mean for defensible? Host: I mean defensible. Will they get commoditized or absorbed by the power of foundational model? Because originally, we do post-production of our own podcast, right? I've been using some pretty good editing tools. And then we just heard from a friend who's a influencer herself. And she's been using Claude, just Claude, without going through those intermediary- editing tools for post-production of her own videos. So we were just thinking, if Claude could just replace all those tools... And it's a tool that's commonly considered kind of niche. It's like video editing post-production space. Then there could be so many other spaces or niches that we thought would be niches and would be hard for absorb, but could be easily absorbed by the next foundational model updates. So just thinking, what's your take on it? And if you think game asset, game engine prototyping could be the next one that get absorbed fairly easily, or not? Heroza Zhang: So the first thing just from my mind is about they are fundamental model. And 20 years ago, people actually can use C++ to make games. Host: Right. Heroza Zhang: Yeah. Host: Yeah. Heroza Zhang: They don't need Unity or- Host: Yeah Heroza Zhang: Unreal. Host: Yeah. Heroza Zhang: But they use Unity or Unreal. It is because a new technology appears, and there should be a way to make a good combine and to mix them together to be put out instead of the human being. For example, AAA games cannot be directly generated by Claude, especially about the rendering part. Yeah, so because it would be limited by the tech stock, and you might cannot see the quality to be high enough as AAA games is. So that is a thing. So that's why 20 years ago, people have C++, but people use Unity, and 20 years later, people have the fundamental model, but we need a new engine for that. That is the things I used to think about. Yeah. And I was convincing about. But when I was talking to my game designer friends recently, there's a more interesting point I heard from them. Host: What's that? Heroza Zhang: Because the generation can only make a prototype for people- Host: Yeah Heroza Zhang: and/or maybe interact based on the prototype. But for the real game production pipeline, it can only be for you to make a beginning stage or something to- Host: Okay Heroza Zhang: just show, for example, Tic-Tac-Toe or just jumping stuff. Yeah, something. Even though it is 3D, because I also use Fable, I know how it looks like. But actually, it also have some basic mechanism, but if you really want to put it in the middle of a workflow and- to make the professional iteration, it's just actually not really fit well for the habit people currently have to making games. So yeah, they still need to design a whole workflow- for putting that tech step inside. Host: I see your point. Heroza Zhang: So it is more just helping them to generate codes faster or something like that. So actually, there still need to be have a combination. So yeah, just imagine 20 years ago, so you say want to make a game and you are- Host: Yeah Heroza Zhang: a programmer background, so you use C++. So then- Host: Yeah Heroza Zhang: let's go. Yeah. Host: Yeah. Heroza Zhang: So... Host: Yeah, that totally makes sense. Would Gorest be accumulating any proprietary data while user using it? Heroza Zhang: You mean did we train our own model, or? Host: Yeah. Heroza Zhang: Okay. So currently, we are trying to make the product to be put into customer and makes the interaction to be good enough and workable enough. So we are more combining the API from others- Host: I see Heroza Zhang: and yeah. But yeah, maybe we can have the time or chance to- use our own model in the future. Host: Yeah, once you grow it and have- Heroza Zhang: Yeah Host: accumulated enough data points, then you can feed it and to refine it and to have your own fine-tuned version of the model that could- Heroza Zhang: Yeah Host: serve better purposes. Yeah, and we do want to... Because in the previous episode before this one, actually, we talked to another friend who is doing a product, which he calls it a cursor for Blender. So I think he's also kind of automating this model building process, but I think he's more focused on industrial design- Okay part of the story, instead of game asset generation. And do you see kind of this... I think there are a lot of people in there. We also met another friend who's building a kind of model building but for construction industry specifically. Because they also need to, for example, you have this design for a building, for example. They need to build a 3D model for that, right? And I think- These three examples, including yours, I suppose they maybe have some overlap in their technology, to some extent. Do you see them as your potential competitors, or would you cross field in the future sometime? Heroza Zhang: I think the first thing is about focus. Architecture also need to make something similar to gray box. Yeah. But, not the games gray box. Architecture has their own blockout. But they are more focusing on, for example, each of the centimeter need to be accurate enough to task whether it's an architecture. So they have different things here really need to pay attention to. And for people, for example, we are making games, so it's about how the visual direction, how, for example, the light direct you to which way. So that is our focus, so it's kind of really different. So that is the first thing. And for strategy generation, so or maybe architecture generation, so it's actually just generating model or something, but I think they're actually totally different things. But sometimes, I think for our product, the good thing is that maybe if we work really well in game industry, there also be... For example, Unreal also not only work for game, also simulation. Or space simulation or architecture or those kind of project also do work on Unreal. So actually similar to our product, we can also be applicable in other industry sometime. But actually, if you mention about platform, those are kind of really different. And I think you mentioned beginning about game for Cursor platform, right? Yeah. So, I'm not sure because Cursor for game, from my understanding, is more like using AI to generate code for games- or something like that. Yeah, but game is not just about code, it's about many different details, and many people are working together. So, yeah. So I'm not sure. I think it's a bit more specific maybe. Host: Okay. And I think that one is, he called it Cursor for Blender. You know Blender, the one which you make- Heroza Zhang: Oh Host: basically 3D models, right? Heroza Zhang: Okay. Host: That software. Yeah. For example, he show us some use cases. For example, you want to make a vacuum cleaner. You just prompt it, give your description, and it would just generate a Blender ready-to-export model in his software, right? I think. Yeah, I was- I'm not even sure if you need to export it to the Blender. Okay. I think now it's built on top of Blender. Yeah. So basically you have a finished product, which you can use directly. Yeah. Heroza Zhang: I think it's more like MCP. So actually, people use generative AI to create a model, or they can use MCP for Blender to create a model. But actually, if you compare the token for each of the pipeline, you can feel like Blender MCP actually runs huge more than generative AI. Like few cents for generative AI, and maybe $40 for Blender. So that's actually really costly, but actually the result is similar. Just to build something for you, build a model for you. So it's actually not about the whole process, about- Host: Okay Heroza Zhang: or just question about if those product... Because it's actually a new industry. It's still blue sea. Because when you are questioning, oh, so we have so many tool. So can you work, can me, just make a triple A game right now? So the answer is no. So the industry is really blank and- Host: Well, I want to. Heroza Zhang: People still... You will.