Today we will learn how (and why) Jussi Kemppainen developed a full game with the help of AI, and what he thinks the impact of AI will be on game development for big and small studios. We will also look at a free powerful gen AI tool for images you might not know about.
Interview with Jussi Kemppainen
Why use generative AI in the development of your game?
Jussi Kemppainen: It began as a kind of harmless experiment to see if I could save some time because I was doing this hobby adventure game project. The project kept getting stuck because making all the art took so long. When DALL-E 2 came out in 2021 and MidJourney in 2022, I saw potential despite the low quality. Initially, I thought I could use these tools to make the game, even if the art wasn't great. Over time, AI technology improved, and during a Christmas holiday, I created a prototype with AI-generated characters and backgrounds. It went viral, which made me realize people were interested. So, I shifted to exploring how AI could be integrated into game development as a kind of public experiment.
How do you evaluate and experiment with new generative AI tools for your workflow?
Jussi Kemppainen: I usually have a goal in mind when I pick up a new tool, like making an icon, a location, a character concept, or a vehicle design. I see if the tool helps me reach that goal quickly, ideally within 15 minutes. If it's cumbersome or doesn't show immediate benefits, I move on. For example, I tried using ComfyUI and local builds of Stable Diffusion, but it ended up being quite time consuming. If a tool works almost like magic and gives good results out of the box, I integrate it into my pipeline. For me It's about minimizing the time spent on asset creation.
Can you explain how AI has influenced the storyline and setting of your game?
Jussi Kemppainen: The futuristic setting of the game was heavily influenced by AI’s capabilities. When AI tools like DALL-E 2 and MidJourney were less advanced, they struggled with realistic contemporary settings but excelled at creating believable sci-fi environments. This led me to set the game in a futuristic world, minimizing the friction between AI's limitations and the game's visual requirements. The story and setting evolved to fit what AI could generate well, which in turn shaped the overall aesthetic and narrative direction of the game.
Have you faced any challenges with AI not being able to produce specific elements you needed for your game?
Jussi Kemppainen: Yes, there have been instances where AI couldn’t make what I needed. For example, I needed a photo booth for the game, similar to the one in the movie "Amélie," but none of the AI systems could produce a photo booth in the style I wanted. I ended up doing it myself in Photoshop, which took about six or seven hours (read the full workflow here). There are multiple cases where I had to adjust my vision based on what the AI could produce, and sometimes I had to do things manually when AI couldn’t deliver.
How do you handle the “loss of control” when using AI in your game development?
Jussi Kemppainen: I’ve had to release control over certain aspects and let the AI guide the development process. For instance, I might have an outline of a story, but when creating a location, I let myself be inspired by the images the AI generates. It’s more improvisational, where I pick the best results and then think of puzzles or story beats that fit those scenes. This approach has helped me stay flexible and open to new ideas, even if it means the final product isn’t exactly what I initially envisioned.
Besides asset generation, how else have you experimented with AI during your game development?
Jussi Kemppainen: I've used AI for voiceovers for side characters, although I hope to get human voices for the main cast due to budget constraints. Recently, I've been using ChatGPT for coding. For example, I needed a function to translate position from world space to local space for coordinates, not game objects. ChatGPT provided a function that worked right away. While I haven't used AI for real-time character responses due to costs, I’ve used ChatGPT to rewrite lines for AI characters, making them sound appropriately robotic and limited, which has worked well.
How has documenting your process on your blog and posting on social media impacted your project?
Jussi Kemppainen: Documenting my process and sharing it on my blog and social media has brought a lot of visibility, though not directly related to my day-to-day work as a game designer or VFX artist. It has helped build my personal brand and opened opportunities to be featured in blogs and magazines. However, it hasn't significantly impacted my professional career or how I use AI at work. The blog has fulfilled its purpose, and I've now shifted focus to developing the game more than posting regular updates.
How do you see the impact of AI on small game developers versus big studios?
Jussi Kemppainen: I have friends who fear big studios will use AI to push out craft people, but I think AI will empower small teams to compete with big studios. AI can make production more affordable, enabling small teams to create high-quality content without massive budgets. This could democratize game development, allowing small creators to produce work that previously required big studios. While there will be a lot of low-quality content, AI can also help uncover hidden gems. Ultimately, AI will likely be a tool that levels the playing field rather than eliminating jobs.
What challenges have you faced when using AI-generated voices for your game characters?
Jussi Kemppainen: Using AI-generated voices presents several challenges. AI voices are great for side characters, but they lack the expressiveness needed for main characters. The game has no budget, so I'm hoping to recruit human voice actors for the main roles. Another issue is the cost of real-time AI voice generation, which can be expensive and isn't sustainable for a free game. Instead, I've used AI to rewrite lines for certain characters, making them sound distinctively robotic, which fits well with the game's theme and adds an interesting flavor to the dialogue.
What are your thoughts on the potential backlash against AI-generated content in the creative industry?
Jussi Kemppainen: There’s a risk of AI-generated content devaluing the perceived worth of art. The flood of AI-created images, for instance, can overshadow genuine human creativity, making it harder to appreciate real craftsmanship. It’s a complicated issue, and while AI has its place, it’s essential to use it in a way that enhances and supports human creativity rather than overwhelming it with trivial, low-value content.
How do you see AI evolving in the game development industry over the next few years?
Jussi Kemppainen: AI is likely to become an integral part of game development. Small teams will increasingly use AI to create high-quality content, reducing the need for large budgets and extensive resources. While big studios might be slower to adopt AI fully, once the technology proves its value, there will likely be a rush to integrate it. AI will help democratize game development, allowing more creators to bring their visions to life. However, we’ll also see an influx of low-quality content, so finding the gems will become a challenge.
Product Spotlight: Leonardo, more than just AI-generated images
Leonardo AI is an AI-driven art generator with highly customizable tools for producing visuals with consistent style.
It offers extensive customization options, allowing users to leverage pre-trained models (to obtain specific styles) or train their own. It supports also diffusion models making it a good choice for a more controlled creative process. The Free tier is quite generous (and especially decreasing the image resolution) you can get quite creative with the 150 daily credits. The other thing I like about Leonardo is that it offers other tools like real time generation, image upscaler and canvas editor. Try it out and check out their tutorial page on YouTube to get a more in depth idea of the possibilities
That’s it for today!
See you next week
