Shulou(Shulou.com)11/24 Report--

This article comes from the official account of Wechat: ID:chuappgame, author: Zhu Siqi

People are still trying to find the right place.

In recent years, AI has aroused a lot of topics and debates. In September last year, Toule published its first article on AI painting. Since then, AI technology has made greater progress. AI can participate not only in drawing, but also in text, dubbing and even soundtrack. The iteration of open source software and the emergence of many powerful plug-ins further improve the ease of use of AI tools.

At the same time, the voices of controversy and anxiety around AI have never been far away. Not long ago, it was reported that the Rhea game fired the painter and completely replaced it with AI drawing. Reya quickly issued a statement denying it, while players still questioned the decline in the quality of the game card surface and other reasons. On the other hand, senior practitioners in the original painting industry released the results of a questionnaire that summarized the recruitment requirements of several manufacturers on job seekers'AI skills, as well as the extent to which painters at different levels improved their efficiency after using AI. The main purpose of this questionnaire is to show that art practitioners are still quite irreplaceable at this stage and alleviate some of the spreading panic.

The Rhea game, once known for its unique art style, is now in a state of controversy because of the suspicion of using AI instead of painters. When the emergence of AI is combined with the economic downturn and even the contraction of the industry as a whole, it is hard to tell how much AI factors account for the wave of unemployment, and whether the technology is as omnipotent as rumored. As a result, Toule once again contacted the game company's producers, visual executives, and even members of the AI team from the game industry to try to understand the current stage of AI application in the industry.

We finally found that even though AI looks so powerful, its applications vary greatly from vendor to developer, and not all of them are integrated into the workflow-depending on the size of the vendor, the type of game product, and even the developer's own understanding of AI technology and game development itself.

The status quo of AI applications "how to integrate AI into the workflow?" One game developer's answer was unexpected: "good question, we also want to know, so we are going to have a meeting to discuss it."

In a sense, this is the current situation of a considerable number of game manufacturers. It is true that the wave of AI has come, but many people are still testing the water.

Akihiko Yoshikawa is currently a member of an AI team, mainly providing AI drawing services and software technical training for enterprises. The team was founded in early 2023 and is very optimistic about the prospect of the rapid development of AI technology. At present, the people in the team have different backgrounds. Akira Yoshikawa is a game producer, so he has more contact with game manufacturers.

In the process, he found that the popularity of AI varies from manufacturer to vendor. Big companies like Tencent and NetEase and independent developers are very active in the application of AI technology. They are at the extremes of the spectrum and want to put similar technologies into use as soon as possible for completely different reasons. A considerable number of manufacturers in the middle are still in the exploratory period, looking for ways to really improve efficiency. In other words, AI technology has not been fully integrated into the workflow.

Examples of big companies can be obtained from their official case sharing posted online. Starting from this spring, teaching documents shared by leading manufacturers have appeared sporadically on the Internet. It includes a fairly detailed introduction to StableDiffusion, as well as some practical application cases.

In the summary of AIGC workflow announced by NetEase, we can see that manufacturers have tried to apply AI technology to CG business, event business, virtual roles and many other fields. This set of process is mainly carried out by people to carry out the initial requirements analysis and the final refinement, the middle of the search for reference, through the draft to generate the design, and even the local refinement stage can let AI participate.

In April this year, NetEase shared some of the workflows that AI participated in, covering not just game development. In terms of images, AI drawings are most commonly used in areas such as icons, live special effects and other areas that do not usually be carefully examined by users, and do not require particularly high precision, such as repeated treasure box icons and medal icons in the game. The second is to quickly output a variety of schemes in the early stage of scene design, clothing design and other applications, and select them for the demand side after the preliminary refinement. Finally, there are some miscellaneous, such as generating realistic-style virtual faces, simplified Q-version images, and even unifying e-sports contestants' make-up style.

In addition to drawing, ChatGPT is currently used to generate user questionnaire outlines and to analyze existing data texts.

The historian of the game producer provides an application case of the division of labor among medium-sized manufacturers.

Due to the advances in graphic technology and the emergence of the plug-in ControlNet, which allows binding to the bone adjustment screen, the process of drawing a draft for AI refinement is becoming more and more convenient. Selecting useful ones from a large number of refinement results of AI for manual refinement can shorten the original 10-day workload to three or four days. "in fact, it frees manual work from picking out details, leaving repetitive work and open-ended work to AI." Introduced by the historian.

In terms of program, historian programmers have basically formed the habit of "ask GPT", regard it as an advanced search engine, and ask GPT which algorithm and practice are not clear. This move to improve efficiency, with the evolution of GPT3.5 to GPT4 to achieve a leap. Even if the code given by GPT has Bug, you can throw it back and fix it yourself, which is tantamount to giving the programmer a hard-working assistant at almost no extra cost.

The planner may also use GPT in the plot, such as setting a NPC character for GPT's thread, and then throwing the lines that the NPC wants to say to GPT and letting it say it with the specified character. In this way, it can avoid the situation that the copywriting planner cannot reflect enough character differences when writing the dialogue.

Pengge, head of vision at a manufacturer, said that in addition to painting, AIGC is also doing well in the video field. AI dubbing and soundtrack can already be used in video advertising material.

As for smaller independent teams and even individual developers, they are likely to be more aggressive with AI applications. On the one hand, these teams may already be understaffed, cutting costs is also a rigid requirement, whether they can use AI or not, and sometimes even determine whether they can enable a feature. An independent developer told Toule that they could not afford to use the original painting, and that with AI, some images could be made to look like big companies. The evolution of ChatGPT gives individuals with no programming experience the opportunity to start trying to develop games.

In addition, independent developers may try experimental works according to their personal interests, regardless of trial and error. Now there are sporadic pictures, scripts, dubbing and other games generated by AI on Steam. They may not be very playful, or even formal works, but they are also a way to apply this kind of technology.

There are still many examples of AI available, but compared with some of the more exaggerated statements and effects on the Internet, it depends on many other variables to really increase the output of actual work. For example, the legendary "reducing costs and increasing efficiency", the actual situation is not quite the same as imagined.

According to Pengge's feedback, in terms of "efficiency", such as the drawing of characters and scene diagrams, the average efficiency is more than 50% in the case of AI, which does have a significant effect. However, in terms of "cost reduction", the company's cost savings are mainly on outsourcing, and internal costs are actually significantly higher because more manpower is needed to be responsible for the use and exploration of AI. "what AI generates now will still be different from what you describe and imagine, especially when it's impossible to be as good as human design in detail, so it will take more time to try and make mistakes and modify it over and over again."

Historians also feel that there is room for improvement in the accuracy of the feedback given by ChatGPT and others. "it is not yet possible to give particularly complete and systematic feedback. If you can hand-in-hand give the accurate code that can achieve a certain function, it will be more difficult for individual developers to develop."

"AI is really easy to use, so it will give you the illusion that it is easy to use at first." Brother Peng describes his experience of using AI in his work. "but if you want to apply it deeply and accurately to your work, there is still a higher threshold. It requires not only the professional ability of fine arts, but also good logical thinking."

This mainly means that the initial ideas and ideas are the most important for the operator to know what to let AI generate. "to put it bluntly, AI can help you do a good job from 1 to 100, but from 0 to 1 still depends on the person."

It is also necessary to explore the complete integration of AI into the workflow. "for example, if an original painting is drawn by a painter from scratch, there will be drafts, line drafts, and color manuscripts, with many layers in the middle. Some companies have explicit rules on this, so that people who take over the manuscripts before and after you can get started smoothly. If you want to modify it, you can clearly go back to some intermediate stage to modify it." "in addition, many characters have the need to do Live2D and need to split the artwork," Yoshikawa said. "however, AI is directly thrown to you with a single layer, and it is actually very troublesome to change it on this basis."

Integrating the AI tool into the workflow is equivalent to making it work well with manpower because when it comes to modification, AI does not improve the efficiency of painters at different levels. In the survey mentioned at the beginning of this article, downstream and upstream painters tend to think that the efficiency improvement of AI is not significant. Because even if the downstream painter sees the problem of AI drawing, because of the lack of modification ability, the efficiency improvement is still not great. The upstream painter thinks that there are too many problems and the effect of modification is too poor, so it is better to redraw. It is considered that the significant improvement in efficiency is dominated by middle-tier painters.

From the point of view of the team as a whole, in addition to the efficiency of the output itself, improving the efficiency of communication is also very important. "because in the past, maybe after brainstorming, it would take weeks to put this concept into practice visually. It's very fast now, maybe in a few hours you can come up with something very close to the finished product for the demand side or the boss to choose." Akihiko Yoshikawa said, "many people will ignore such communication costs, in fact, usually, such communication takes up a lot of people's time."

Pengge believes that the essential reason why he does not know how to integrate into the process is that AI has revolutionized the mode of production of game manufacturers. Just like when PS and digital painting first appeared, people should fundamentally consider changing the way of creation.

"for example, it turned out that we went to find an outsourced drawing, and there was a group in the outsourcing company, the lead writer was responsible for designing sketches, the assistant was responsible for painting, the painter who was good at face painting was specialized in face painting, and then there were special people who took apart pieces, completed the details, or tied bones to generate Live2D and so on. This is such an assembly line division of labor." Brother Peng said, "now that you have AI, AI' will directly give you a picture, and the process may be reversed, and people will be responsible for matting, layering, and completing the details of modification." Compared with some outsourcing organizations with varying standards, the quality of AI drawings may be more stable.

Although it looks like manual work, the latter currently takes less time than the former. "originally, manual direction finding and manual refinement would take a total of about 10 days, even if they were divided into layers at the beginning. Now it takes 1 day for AI to find direction, 1 day for AI refinement, 2 days for manual layering with AI semi-finished products, and 2 days for manual refinement, which is a total of 6 days, which is still shorter than before." The historian calculated the overall efficiency, "and when AI is in charge of work for two days, people are doing something else, or they just get off work and leave it to the graphics card at night, which does not affect each other."

Of course, AI is not a panacea. Several respondents mentioned a typical example: AI is still not much help to products with unique art style as the selling point.

One reason is caution. Manufacturers of a certain size rarely use AI-generated content directly in commercial games, nor do they promote it to players and users. "there are still some people who resent this, the proportion may not be high, but the absolute number is not small." Akihiro Yoshikawa said, "especially in two-dimensional games, it is easy to be 'inflamed' by players." After all, the value of such games often lies in art resources, whether it is a live painter "rollover" or the real use of AI, will be questioned.

Another reason is maneuverability. Last year, some developers suggested that they wanted to train AI locally to suit their own style. According to Akihiro Yoshikawa, this problem, as well as the compatibility of AI tools and other office software that has been mentioned sporadically, has almost been solved technically. Nowadays, not only common tools such as Photoshop have launched built-in AI plug-ins and ChatGPT has been connected to the search engine, open source software such as StableDiffusion has allowed manufacturers to split and specialize functions according to their own needs. In theory, each manufacturer can now have a set of AI toolsets tailored to their own projects and be able to train locally for their own product style.

AdobePhotoshop surprised designers by introducing the official built-in AI function, but this set of tools is not available to everyone. Brother Peng explained the invisible resources needed by AI tools from the perspective of medium-sized manufacturers.

"for example, the future development direction of StableDiffision should mainly rely on B-side companies with redevelopment capabilities to make customized optimization adjustments to cultivate more unique models. However, this thing needs to rely on two foundations: one is computing power, which means money; the other is data, and you need to have a lot of local data to feed AI."

Large companies must have more data than small and medium-sized ones. "just like advertising, large advertising companies (such as 4A advertising) do hundreds of similar projects around the world every year, while small advertising companies (such as creative hot stores) have only a few cases a year. Although these two may be particularly good, all of them are high-quality products, this amount of data is difficult to support the targeted training of AI." Brother Peng said. Without orientation training, AI drawing will be very homogeneous. "to put it bluntly, it is easy to train the most common and standard two-dimensional painting style circulated on the Internet, which is not competitive."

So in Pengge's view, the rapid development of AI technology does "devolve" productivity, allowing small teams that can't afford the original painting to have decent art, and independent developers who can't program to copy available code from ChatGPT, but the important thing is that big companies still have a leading advantage. "if you want to develop AIGC in the future, the basic data accumulated in the previous era may be wealth."

Looking for the right place from the feedback of these developers, AI has been able to play a considerable role in the development process, but it is not as omnipotent and powerful as some people think or boast.

Historians feel that if the use of AI in the team is not effective, it may be that the manager has not figured out exactly what to do with AI. After all, this is a matter of concept.

"some people want AI to achieve the output of finished products from scratch, which is obviously not possible at the moment. When they find that AI does not directly give the most satisfactory results and does not completely replace a human function, they think AI is useless." The historian said, "this one-step idea is wrong. AI is an acceleration tool for developers, and it's hard to use AI if you don't recognize it." In other words, AI is not even an invisible or intelligent employee, but an assistant to existing employees to improve their efficiency for the right people.

Pengge's idea is that after the popularity of open source tools like StableDiffusion, which allow manufacturers to customize, the future AI software should be like office software like flying books and nails, configured by large manufacturers or professional teams, and purchased and used by small and medium-sized manufacturers. Mr. Yoshikawa's team has begun to consider providing similar services in long-term projects.

It is true that we stand on the edge of a new era, but we are far from really stepping into it. A large part of the use of technology, whether it is radical or not, and whether it is accepted by the audience, is still waiting to be explored and verified.

In the final analysis, as "using AI" becomes the norm, people in the industry are trying to find the right place-- both as a development tool and as a user. It will be a long period of exploration. It may still take time to prove whether the last thing waiting for developers during this period is an opportunity or a more severe challenge.

(at the request of the interviewees, some of the characters in the article are aliases. )

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.