Who is Kizuna Ai

Brief Introduction:

Kizuna AI (キズナアイ) is a virtual YouTube on youtube, claiming to be the first virtual YouTube in the world, and the YouTube channel is called a.i.channel. The character image is designed by moricangyuan, and the 3D model is made by TDA.

Development to success:

At the end of 2016, a virtual idol called Kizuna AI went online on youtube, a foreign video website. With its lovely image and sweet voice, it instantly attracted countless fans. How is this "artificial mental retardation" that can interact with the audience in real time behind the dumping of countless otaku men?

In order to find out this problem, we consulted a large number of relevant materials and deduced the production process of the virtual image of love sauce. Although it can not be guaranteed to be completely consistent with the real production process, the general routine will never be wrong.

We can analyze it from the following points:

  1. Profile setting
  2. Environmental setting
  3. Avatar character
  4. Topic selection and operation

1.Profile setting

From this picture, we can see that the virtual image of love sauce is portrayed very vividly: changeable expression, cute, stupid and mentally retarded! Compared with other peers, Kizuna AI's virtual modeling is more obvious, and the overall VI effect has been seriously considered in the preliminary design. The features of modeling and color matching are simple and obvious, high recognition and easy to remember.

It is very convenient for enthusiasts to make map modification meme for network communication, it is very convenient for physical peripheral planning, and it is very convenient to transform the corresponding theme clothes during special festivals (for example, this Christmas clothes still maintain the basic style of clothes and retain the character characteristics after replacing the art elements).

Then the character appearance. The modeling of Ryokusenen(もりくら えん)'s models design + TDA fully reflects the temperament of health, liveliness and youth.

2.Environmental setting

When it comes to the background layout of kizunaai… Of course, the invariable "infinite white space" has only one more drawing board in the past year. It can't be seen from the ceiling all the time. On the one hand, this open background can be used as the self abuse stem of the anchor, but the success in operation is that this seemingly simple design completely cuts off the invalid information.

The biggest difference between the pictures of the two-dimensional works and the three-dimensional works is that all the appearance contents are controllable by the author, and there is no meaningless noise information. Try to add a large number of meaningless scenes, hoping to reflect. On the contrary, when the audience psychologically predicts that they are watching the works of the two-dimensional system, The content presented in all pictures will be taken as information intake - on this premise, the design of "white space" can achieve the purpose of completely focusing the audience's attention on Kizuna AI.

3.Avatar character

It is not "Idiot's words and deeds" that can make complaints about hot topics, but "worth talking about." In fact, the highlight of kizunaai's fool design is that the performer himself occasionally shows a little cheap character. Cheap and cute (commendatory) - it's good for everyone to have their own ideas

In the virtual anchor, there is a bright night moon who seems to have drunk fake wine. In terms of other peers, akari will be closer to this field in the future. Fuji Kui is too airwaves. In fact, it's a little difficult to get a smile. Xiaobai and Yue can't let go. In contrast, it's really common sense. You can't scold but you can't force it. Viewers who like girls of sister or local flavor should think it's great.

Compared with the aura of the anchors, I feel that Kizuna will be "a girl who hopes to fall in love with her", Huiyueye is "a close friend with a bad character", Akari will be "a vigorous neighborhood Idol", Yue is "an uneasy sister", Xiaobai is "a sister like a small animal", so…) together, one phase and one session… But wait a minute, Kizuna AI, do you use the wrong idiom again

4.Topic selection and operation

The scope of early selection is very wide, and the content in various directions has been tried quickly and iteratively( This principle is similar to comic serials. For example, in the pioneering period, also tried various games in the form of unit plays. The reverse of cards is particularly good, so we can find the direction.) This mode requires frequent data review and various duplicate disks. In fact, the implementation is quite annoying( The bitter tears of operation, and the update frequency has gradually changed from half a week to one Saturday, which requires high intensity of creation.

It is commendable that the topic selection of the program is not limited to funny play and game live, but also focuses on app recommendation, web tool introduction, VR trial and demonstration, so as to get more extensive cross-border cooperation( For example, cooperate with Microsoft in Japan to promote MR equipment in physical stores. From this point of view, BD members of kizunaai team are also great)

In the first half of 2017, I was very worried that kizunaai's smile showed poverty. I always received some strange mobile game and page game advertisements. Was it not enlightened by the operation, could it make normal profits, and would such fun programs go bankrupt… But at the current time node, I was still too depressed, Originally, Kizuna's operation team has been gradually implementing the flow realization plan at a stable pace. I'm afraid the paid promotion of mobile game cooperation is just an addition… It's better to say that game cooperation is a means to bring traffic to the channel. Indeed, the real profit is still through the live rockets and audience monthly cards that have been experienced by countless domestic peers… ( ̄▽  ̄)

At present, the team's response to crisis management is also good. One of the cases is to maintain fans immediately after ban, and then to tout herself after the event. The two case is to operate in Sina micro-blog semicolon, and make complaints about "celebrate Japanese holidays" at the time of the "77 incident" in China. Make complaints about feed in the comment area, and control the evil evaluation within a very small range (after learning the lesson, only a calendar time is posted on the "918" day). This localization operation is worth affirming.

On the day when the tubing fans broke millions, Xiao AI was so excited that she replied to the fans' comments one by one. She still wrote back until 3:00 in the middle of the night. This enthusiasm was quite moving w

Virtual live broadcast:

From the above points, we can see that there is no big difference in the development path of many IP's. The biggest difference is that her virtual live identity can interact in real time, and there is no waiting for update. Real time communication is more like a real person, which makes people more willing to communicate. However, there are still disputes about whether love sauce is real artificial intelligence.

Although love sauce always claims to be "super AI", in fact, this is only the official "setting" of the virtual image of Kizuna AI. In fact, the production process of love sauce mainly includes the following three elements:

① Voice actor doing action + dubbing + voice recognition lip type

② Rich facial dynamic expression package controlled by special personnel

③ "Motion capture system kila" with motion capture as the core

Before explaining these three points, we should first make it clear that the video or live broadcast of love sauce is actually just a "3D model role".

A process of the birth of "AI mentally retarded":

Characters and actions:

First of all, the voice actress who dubbed the virtual image of "Love sauce" needs to wear an inertial motion capture device called "Perception Neuron" (provided by the manufacturer Noiton);
Then, after a few steps of movement calibration, Ai-chan can appear on the screen vividly according to the movements made by the seiyuu lady;
Finally, in the Unity environment, set up a virtual camera for the "love sauce" that has completed the motion capture calibration, and you can observe the legendary "artificial mental retardation" from different angles!
And capture this image, or edit it into a video, or directly push the stream and start the live broadcast, you can see the "love sauce" as usual!
Since Aichan's voice and movements are all "performed" by the voice actress behind her, how did Aichan make all kinds of mentally retarded facial expressions?


At first, we guessed that Ai Jiang’s expression was demonstrated in real time by FaceRig or similar image recognition software by capturing the facial expressions of the voice actors, for example:

But in fact, none of Kizuna AI 's videos and live broadcasts use similar technology!

In KiLA’s official demo video, we found that Ai Jiang’s rich expressions are actually a lot of "emoji packs" prepared by the government in advance. When the voice actress is dubbing and making certain actions, Ai Jiang’s eyes 'S expression needs to be controlled by another staff member:

Control Aichan's eyes, gaze and other eye expressions through the handle!

Kizuna AI 's mouth shape is automatically matched based on speech recognition technology (the technology is currently very mature).
Therefore, our favorite AI sauce for artificial mental retardation is actually completed by two people working together! This also explains why Kizuna AI 's expression is so "mentally retarded" in many cases (the feeling is not a person at all)!
So here, I can tell you very clearly: Kizuna AI is indeed "artificial", but she is not "intelligent"! However, this does not prevent everyone from loving the virtual image of Kizuna AI !
However, the operation of such a virtual image is not only technically difficult, but also requires professional training of relevant personnel. The difficulty and threshold are very high. If you are not careful, various errors will occur, or even impossible to solve. It makes it difficult for small teams to stick to it.

So is there any better solution? For example, such as facerig.

One person can realize the virtual live broadcast APP:

Regarding this, we have now many other manufacturers have started to do it, and some have done very well. For example, the facerig we mentioned above, of course, is a European and American style. If you like Kizuna AI ’s "Japanese comic" style more, you can check out Live3d's related products, which is the closest virtual live broadcast software to Kizuna AI .

The real-time 3d surface capture function allows the virtual character to perform simultaneously with the real person.