AI Image Generation Has Become Indispensable in Game Development
After over a year of development in large AI models, the author has become accustomed to seeing AI mentioned at industry exhibitions. From AI range hoods to AI wearable armor, AI seems to be everywhere.
However, at ChinaJoy, the author was surprised to find that game exhibitors were not rushing to capitalize on the AI trend at their booths. The displays of major game companies still primarily focused on the game content itself.
Nevertheless, after talking with game developers attending ChinaJoy, the author discovered that AI image generation capabilities have already widely entered the game development process, both in small independent teams and larger companies.
An independent developer who has created nearly forty games on their own told GeekPark, "Currently, AI image generation services applied to game development are very mature. I can generate 4 images at once, run AI 40 times simultaneously, resulting in 160 different directions. I can easily choose one direction to continue development."
A conference attendee who teaches game development also stated that they now teach game developers how to use AI image generation to improve development efficiency.
Another widely applied AI assistance is AI code writing. An independent game developer mentioned that although he had game project development experience, he couldn't write code. With AI code assistance, he was able to create his own game by working hard with AI. "Does this count as half an AI game?" he joked.
The game lead for NetEase's Egg Party also shared how AI generation tools helped users more easily create new levels, adding richer content to the game.
Game developers generally expressed that they look forward to applying AI in video generation and 3D generation in the future.
Developers compared AI image generation and AI video generation: "AI video generation is still not very controllable. Sometimes the first half of the generated video is as desired, but in the second half, the content in the video becomes distorted. Also, AI image generation can produce 160 images at once. AI video can't do that - it generally takes a long time to generate one video, and you can't generate multiple videos simultaneously. You have to wait in line for one to finish before generating the next."
AI 3D generation is another area of anticipation, but there are still some insurmountable problems in directly applying AI 3D generation models to game production pipelines. However, Zhi Ying, a development engineer from Ubisoft China, stated that the technology for AI-generated 3D model textures is currently relatively mature and will be applied to game development processes earlier than 3D model generation.
Innovation in Gameplay Limited to Dialogue-Based Games
In imagining large model technology joining games, a common vision is to let AI take over the game's storyline - originally requiring long branching storylines, but in the future, the game's direction would be completely controlled by AI, with NPCs directly handled by large model dialogues.
However, after talking with game developers at this ChinaJoy, we felt that this future model is still not very mature. On one hand, many games are still 3D games, and because the assets need to be developed in advance, there must be some control over the plot. On the other hand, the capabilities of AI large models themselves are not stable, and it's still difficult to take over the entire storyline of a game.
Under these constraints, we see that current AI-native games still mostly limit gameplay innovation to dialogue-based games - such as the previously popular "Coaxing Simulator," where AI plays the NPC judging whether the game is cleared, while humans use their wit to clear the game through dialogue. The game interface itself is mostly 2D, or even just a dialogue box.
However, even under these constraints, we still saw several interesting AI-native games with some gameplay value.
For example, the independent game "1001 Nights" reimagines the fairy tale background. Players take on the role of a poor bride telling stories to the king, but unknown to the king, the bride has the ability to turn words into reality. Players need to tell a coherent story, planting the weapons they want in the story without angering the king, and ultimately defeat the tyrannical king by turning the weapons into reality through storytelling.
Another independent game, "Yanling Project," is set up as a natural language-driven battle game. Both attacking and defending sides must come up with outrageous attack methods to drive their sprites to battle. AI judges whose attack is more outrageous and whether the attack can be neutralized by the opponent's outrageous solution. It's said that during testing, many players became very addicted to this battle game based on making things up, with an average playtime of thirty minutes.
At ChinaJoy, we saw that AI dialogue-based games are relatively mature. Many game developers also sought to understand related information at this exhibition.
Zhipu AI company also participated in this ChinaJoy. Their game industry expansion personnel told GeekPark that during the ChinaJoy summit, many game developers came to inquire about how relevant AI technologies could be combined with their games. The main product they promoted this time was their super-realistic large model responsible for character role-playing.
This is also consistent with the industry phenomenon we observed: Companies with more sensitive touch have already started developing dialogue-based entertainment apps for overseas markets.
A few days ago, news reported that MiniMax's entertainment dialogue application Talkie, launched last year, has accumulated 3.8 million downloads in the US market this year, ranking fourth among AI applications. At the ChinaJoy conference, Japan's Freakout company also stated that generative AI entertainment applications are dividing the usage time of the game market.
Major Companies Launching AI Companions
During ChinaJoy, many major companies launched mobile game public tests, with AI companions becoming a new selling point in many games.
On July 25th, "Naraka: Bladepoint" mobile game started public testing. One highlight compared to the PC version is the addition of AI teammates that can communicate in real-time voice.
AI teammates have different personality types to choose from. Unlike human teammates, AI teammates are "truly obedient." They really achieve the goal of attacking where directed, helping pick up equipment when asked, and dropping equipment when requested. They also engage in emotional communication with players, calling them "baby" and praising their good performance.
Similarly, during ChinaJoy, Youzu Network also publicly revealed the image of its first AI companion product codenamed "Little You Sauce" for the first time.
Providing game companionship and bringing emotional value seems to be another direction for AI and game integration.
"Bold prediction: in the future, encountering AI characters almost indistinguishable from real people in games will be a high probability event," said Pang Daizhi, Vice President of NetEase.
Visible AI and Invisible AI
An interesting thing is that not many AI logos could be seen in the game exhibition halls at ChinaJoy. The place where the most AI logos could be seen was actually in ChinaJoy's consumer electronics exhibition hall.
It seems that hardware manufacturers are more eager to label themselves with AI than game manufacturers.
In this year's consumer electronics exhibition area, we could see various chip manufacturers emphasizing their AI chip capabilities, various phone manufacturers emphasizing their AI phone capabilities - Huawei highlighted the native intelligent frame insertion capability of its Harmony OS in the exhibition area, and Lenovo brought its AI PC, emphasizing intelligent assistant functions.
In places we can't see, AI is also quietly changing many game-related industries.
Moloco, an overseas advertising company attending ChinaJoy, stated that the precise targeting of its advertisements comes more from its advanced deep learning neural networks behind the scenes. In the future, by conducting group tests on AI-generated advertising materials, Moloco's advertisements will be able to serve game clients more precisely.
Game civilization behavior review, which originally required a lot of manual intervention, is now also being experimented with AI introduction by MOBA manufacturers for more efficient review of uncivilized behavior in tens of thousands of matches every day.
Ubisoft is using neural networks to change the traditional approach to ray tracing, using small neural networks for visualization calculations.
After touring ChinaJoy, we found that the emergence of artificial intelligence has brought the dawn of new game gameplay. However, standing at ChinaJoy 2024 today, new AI technologies have not yet brought revolutionary changes to game gameplay.