September 6th, Thursday.

In the Tardy Studio, employees were busy working on the development of the new VR game “Real Estate Agent Simulator.”

After being forged and polished through “Animal Island,” the entire development team had made significant progress, with more seamless collaboration and smoother development processes.

Even Cai Jiadong, the lead designer, had become more adept.

He seemed to gradually grasp Tenda’s design philosophy and methods, becoming increasingly aligned with Lin Wan’s thinking, and gradually developing into a qualified lead designer.

After working on his computer for a while, he took some printed documents and went to report his work to Lin Wan.

At this moment, Lin Wan was concentrating intensely on her computer screen, with slightly furrowed brows, as if pondering something very important.

Seeing Cai Jiadong’s arrival, Lin Wan temporarily shifted her gaze from the computer screen.

“Director Lin, have you looked at that experimental mini-program sent by the Slow Horse AI Laboratory? What do you think of it?”

“If we can use it, it would not only significantly shorten the development time of our entire project but might also influence the specific design plan of the game.”

Lin Wan paused for a moment: “Oh, I’ve been thinking about other things lately and haven’t had time to look at it. How do you find the results?”

Cai Jiadong answered honestly: “I think… the results are surprisingly good.”

Lin Wan hadn’t expected Cai Jiadong to give such a high evaluation.

This mini-program was just sent two days ago by the Slow Horse Laboratory to both Tenda Gaming and Shangyang Games departments, described as an experimental artificial intelligence program still in the testing phase, for them to try out and provide feedback.

Since Lin Wan had close ties with Shangyang Games, this program was naturally also sent to the Tardy Studio.

After all, they were all part of Tenda’s internal industries, family members speaking with one voice.

Lin Wan had been a bit busy these past two days, so she hadn’t had time to look at it.

The reason she hadn’t paid much attention was mainly because she thought this thing probably wouldn’t be very reliable.

Currently, interactive movie-type games, or some RPGs with lots of dialogue that need to show character expressions and speech up close, all used motion capture technology with actors performing the roles.

Motion capture technology could capture not only body movements but also subtle facial expressions. During the process of actors speaking their lines, audio was recorded simultaneously with facial capture, and then this data was applied to models created based on the actors, resulting in very realistic effects in the game.

Of course, if the actor’s delivery wasn’t satisfactory, someone else could be hired for voice-over to achieve better results.

However, this approach had a problem: the workload was too large, and it wasn’t very flexible.

Once the plot dialogue was set, it couldn’t be changed casually. Even adding a single line would require bringing the motion capture actors back for re-recording.

According to reports, this program from the Slow Horse AI Laboratory was an alternative solution to facial motion capture technology. The specific principle was to automatically generate characters’ facial movements through intelligent analysis of voice content, combined with past motion capture data in the editor. In other words, it could be considered automatic lip-syncing.

It was roughly similar to some software that made photos sing, but reportedly with better results.

Lin Wan was somewhat skeptical about this.

How could something automatically generated by a program be compared to what motion capture actors performed in person?

Mr. Pei had very high requirements for the “Real Estate Agent Simulator” game—it had to have high-precision modeling like “Struggle,” while also using real human motion capture for all actions, to give players an immersive and realistic experience.

If the NPCs’ movements were stiff and expressions distorted, it would break immersion and flaw the entire game.

It might even trigger the uncanny valley effect, giving players a scare.

After all, the immersion of VR games was far beyond conventional games—a minor error could lead to major problems.

But now hearing Cai Jiadong say that the results were surprisingly good?

Lin Wan couldn’t help but become interested and checked on her own computer.

The “Real Estate Agent Simulator” game had been in development for a month, and many character models had been created so far.

Due to the existence of the official platform simulator, creating these character models was actually a character customization process, not requiring everything to be built from scratch. Moreover, the “Struggle” game that Mr. Pei had previously made, although mostly featuring Westerners, had many characters that could be directly used with minor modifications.

As soon as he received this program, Cai Jiadong had arranged for designers to use the existing models to create a few cutscenes and recorded them.

Lin Wan quickly watched the cutscenes and after a moment of silence asked, “This was automatically generated using the new program? Not motion capture?”

Cai Jiadong nodded: “Completely automatically generated. However, the support for foreign languages isn’t particularly good at present.”

Lin Wan fell silent.

Whether the support for foreign languages was good or not didn’t matter; after all, the “Real Estate Agent Simulator” game was originally only targeted at domestic players. Even if the game’s reception was good and they planned to release a multilingual version, that would be something for several months after the release.

These results were truly impressive!

The expressions of the characters in the game were very natural, without any sense of stiff movements or distorted expressions.

Of course, it couldn’t be as perfectly seamless as real human motion capture, with perfect coordination between voice and facial movements.

But unless one was intently focusing on the character’s lips, the discrepancies were basically unnoticeable.

Actually, achieving this level was already sufficient, considering that many foreign films and TV shows with redubbing have completely mismatched lip movements, yet audiences get used to it and don’t find it strange. The facial expressions automatically generated by this program were much more natural than those redubbed films and TV shows.

“Based on the current situation, this technology can be directly applied to the ‘Real Estate Agent Simulator’.”

“And…”

Lin Wan’s eyes suddenly lit up: “Doesn’t this perfectly solve that problem that has been troubling us?”

“No wonder we always felt that this problem couldn’t be solved well and were uneasy about it. Looking at it now, could it be that Mr. Pei was already prepared, knowing that this new achievement from the Slow Horse Laboratory could solve this problem?”

The emergence of a new technology often influences game design thinking.

Previously, Lin Wan and Cai Jiadong had been very worried about the design details of the core gameplay in “Real Estate Agent Simulator.”

According to the initial setting, one of the core gameplay elements of this game was selling houses.

The basic process of this gameplay was: players receive a new house type, renovate and rearrange the house, use their funds for repairs, replace furniture, then receive customers, answer different questions during the introduction, and finally make some price adjustments to close the deal.

This gameplay mechanic ran throughout the game, although at the beginning, players might not have enough funds to make changes to the rooms and could only lower the rent to close deals after customers pointed out flaws. In later stages, players could identify problems with the house in advance and resolve them before customers came to view.

But in designing this core gameplay, Lin Wan and Cai Jiadong encountered a small problem.

Obviously, this was a repetitive gameplay mechanic, and if not done well, players would quickly get bored.

If every customer looked similar, asked similar questions, and had similar dialogues, actions, and behaviors, players would likely get bored quickly, and this core gameplay would not be able to sustain the game experience.

So, how to make this game maintain freshness like a MOBA game despite repetition, at least making players feel that each customer encountered would be brand new, and looking forward to receiving the next customer with anticipation—this was crucial to the game experience.

This problem was multifaceted.

To solve the appearance issue, the customer’s appearance could be automatically generated by the program.

Many games’ character creation systems had similar functions, which wasn’t difficult to implement.

Of course, faces automatically generated by character creation systems might be somewhat inconsistent. For example, a woman dressed in business attire might have a mohawk hairstyle, earrings, and an exaggerated tattoo on her forehead, which would look strange from any angle.

Therefore, this system would need some patches, using algorithms to define this person’s identity and correspondingly lock certain specific elements in the character creation, preventing the character’s appearance from deviating too far from the original character design.

With the appearance issue solved, next came the actions.

This was also relatively easy to solve, as NPCs in a room could only do a few types of actions: sitting on a sofa, drinking tea, gazing out of a window, bending over to examine a corner, etc.

These actions could all be arranged in advance, making a few sets of actions according to the character’s gender, body type, and age, and automatically triggering them in the room according to algorithms.

Additionally, NPCs would have corresponding body language when speaking, which was also easy to handle.

In short, these were problems that could be solved by accumulating a large quantity of assets.

As long as enough actions were accumulated and automatically extracted through random algorithms, combined with different appearances, body types, and other factors, it could be done fairly well.

The most critical problem was the content of the speech.

In the core gameplay, each house had corresponding issues, such as poor ventilation, small bedrooms, unclean bathrooms, poor window soundproofing, and so on.

Of course, each also had its corresponding advantages.

Before the customer arrived, players could view the house themselves. After viewing, the system would provide several options for players to choose from, such as whether the layout was regular, whether there was good ventilation, how the lighting was, and so on.

The player’s choice at this time represented the accuracy of their understanding of the house and would also affect the options when answering NPC questions.

When showing the house to NPCs, they would ask questions about the house, and players needed to choose corresponding options to respond. If they chose correctly, there would be no problem; if they chose incorrectly, it would lower the customer’s evaluation of the house, leading to either a price reduction or the deal falling through.

These questions were, of course, predetermined and involved many professional aspects.

This content wasn’t difficult to create either; with more communication with Sloth Apartment, standard answers could be established for each house type.

But the biggest problem was… how to handle the specific wording used by NPCs when expressing these questions?

The same question would definitely be asked differently by different people.

For instance, some might say “the house has poor lighting,” others might say “it’s too dark in here,” and still others might say “the building across blocks the view,” and so on.

If all NPCs said “the house has poor lighting,” it would seem very rigid.

But this problem couldn’t be solved by just accumulating assets.

If still using the previous motion capture method, changing each phrase would require a new motion capture session. Additionally, people of different body types have different facial expressions when speaking, and motion capture couldn’t be mixed and matched. This meant that multiple expressions of the same idea might need to be recorded several times.

For other plot-driven interactive games, this wouldn’t be a major issue—it would just mean more work, requiring a few more recording sessions.

But “Real Estate Agent Simulator” was different; there were too many questions, the volume was too large, and yet it didn’t provide any critical improvement to the game experience.

Not doing it would mean it wasn’t perfect or extreme enough, easily breaking players’ immersion;

Doing it would cost too many resources, to an unbearable extent, with extremely low cost-effectiveness.

So they were stuck in a dilemma—neither option was suitable.

But now, with this artificial intelligence program from the Slow Horse Laboratory, this problem was immediately solved!

This way, they no longer needed to motion capture every line, but only needed voice-overs.

The workload of pure voice-over was much smaller, and they could even collect voices online, allowing enthusiastic players to help with voice-overs. After importing these voice-overs, the program would automatically generate the character’s facial movements based on the audio, paired with certain body language and actions, creating a sense of unique individuality for each character.

Cai Jiadong sighed: “That’s what I thought! When designing this detail earlier, I was wondering, this problem was simply unsolvable under the existing conditions! Did Mr. Pei not think of this when designing this game?”

“Now it seems that Mr. Pei clearly thought of it early on, definitely knowing about the recent research results from the Slow Horse Laboratory, which is why he had us make such a game!”

Lin Wan nodded: “That’s definitely the case.”

“Then there’s no need to hesitate, let’s proceed with this!”

“By the way, Sloth Apartment has already sent over the new batch of house type information, please check and pass it to the designers.”

Cai Jiadong nodded and said: “Alright.”

After starting the development of this game, Lin Wan directly approached Liang Qianfan for assistance.

When it came to this knowledge, who could be more professional than Liang Qianfan?

Sloth Apartment had many architects and designers specializing in how to renovate house types. They just needed to send people to collect materials on various house types online, then sort out the advantages and disadvantages of these house types, and send them to Tardy Studio to be directly implemented in the game.

These matters, of course, were not for Lin Wan to worry about; they were the job of the lead designer, Cai Jiadong.

Watching Cai Jiadong leave, Lin Wan leaned back in her chair and let out a long breath.

“Phew… good thing Liang Qianfan wasn’t taken away.”

“If he had been taken away, although we could find another architect at Sloth Apartment, they definitely wouldn’t be as reliable as him.”

“Speaking of which… what’s the situation with Mr. Pei suddenly taking the department heads for special training…”

Yesterday, Lin Wan had felt something was off.

Although she had moved to Tardy Studio, she was still in the chat group for Tenda’s department heads.

Usually, everyone would chat in the group, exchanging information about the status of various departments, and the group was quite lively.

But yesterday, the entire group suddenly became quiet, and several people who were usually very active didn’t say a single word.

Lin Wan felt something was wrong and inquired from Wang Xiaobin today, only to find out that ten core department heads, including Ye Zhizhou, had been taken away…

Lin Wan was startled, initially thinking that these people might have been caught by Mr. Pei doing something wrong behind his back. But upon further inquiry, she learned they had been taken to a special training base…

Even the head of the Slow Horse AI Laboratory, Shen Renjie, who had recently provided significant help, was taken away like the other department heads.

Fortunately, Liang Qianfan from Sloth Apartment survived; otherwise, the game development at Tardy Studio might have been affected to some extent.

Lin Wan’s instincts were very sharp, and she immediately felt that this matter was no small thing.

After asking a bit in the group, she successfully obtained the list of the first batch at the special training base—a total of ten people: Huang Sibo, Hu Xianbin, Xiao Peng, Rui Yuchen, Ye Zhizhou, Shen Renjie, Guo Licheng, He Desheng, Chen Yufeng, and Ma Yiqun.

“The special training base has just started, so it needs to progress step by step. They can’t train too many people at once, so having only ten people in the first batch is reasonable.”

“But… why these ten people and not others?”

“There are so many department heads at Tenda, there must be a corresponding evaluation standard, right?”

“Mr. Pei attaches great importance to this matter, and is willing to temporarily impact work to arrange for all core department heads to go, which is clearly for some special purpose. And this purpose is even more important than short-term work or even profits.”

“Is it just to support Bao Xu’s work? That’s definitely not possible. Mr. Pei is a person who distinguishes between public and private matters; what he does is ultimately beneficial to Tenda.”

“If it seems disadvantageous in the short term, then it must be beneficial in the long term.”

“What is Mr. Pei’s true intention…”

“Hmm, perhaps I should start with this list. If I can figure out the standard by which Mr. Pei filtered out this list, I should be able to roughly deduce Mr. Pei’s true intentions…”

Lin Wan looked at the computer screen again, deep in thought.

Although she was no longer a department head at Tenda, she still kept a close eye on internal developments. After all, every movement within Tenda represented Mr. Pei’s will, and Mr. Pei’s actions often implied deeper meaning. Learning more would definitely be beneficial.