When everyone talks about AI disrupting the game industry, front-line practitioners use Deepseek this way.

Article source: Game Grapes

Image source: Generated by AI

I wonder how big an impact Deepseek has on the gaming industry.

You see, this tool was so popular that it was called a “national sport-level scientific and technological achievement” by the producers of “Black Myth: Wukong”.

For a period of time, Deepseek itself became a traffic password and a source of wealth. From concept stocks to skyrocketing, to big companies in various fields to connect to Deepseek, and from small to zero basic use Deepseek to make games… Along this trend, maybe the “99% of developers” Cai Haoyu said will really be forced to change careers ahead of schedule.

“Cold Against the Water” and “Peace Elite” access Deepseek

But is this really the case? After chatting with many game practitioners in different positions, Grape King came to a different conclusion.

The following are the self-accounts of these practitioners:

01

Planning: cerebral violence participants,

Collaborative efficiency improvement tools, search engines

– -Level planning of a Shanghai Four Little Dragons game

When I have checkpoint colleagues around me doing demos, they will use their own requirements to first ask Deepseek to write a code, and then demonstrate the effect to technical colleagues. This operation method will allow technical students to better understand the planning requirements.

Use Deepseek to directly generate corresponding level slices through the required content

The source of the picture material is Grape King Simulation, for demonstration only

The other is to use Deepseek as a search engine, such as asking it which games have been used with a certain gameplay mechanism.

Recently, I have also been thinking about how to expand Deepseek’s application, because this tool has the special advantage of giving references and reasoning processes, so that it doesn’t really matter whether the results it gives are accurate or not-I can see it myself and conduct a secondary search based on these intermediate results.

This feature is particularly suitable for personal cross-bank learning and development:

Another example is to help me sort out the logic of the planning document and check for leaks and fill in gaps. This is very meaningful for my personal development.

I feel that Deepseek may be similar to “random” in the past. No one imagined that “random” could produce “Roguelike”, and no one could imagine how much potential AI has for gameplay.

–Copywriter planning of a stand-alone game

The last time Deepseek was helpful to my work, it should have helped me find information.

Under the networking and in-depth thinking mode, let it give the names, information, selection reasons, packaging plans, and packaging logic explanations of historical figures in a certain era who are suitable for secondary packaging as heroes, and then manually screen and modify them twice. As materials for superiors to choose.

The source of the picture material is Grape King Simulation, for demonstration only

But I don’t think Deepseek can really join the core R & D workflow for the time being. It does provide some help in terms of inspiration and packaging, but the efficiency of finished product transformation is not as high as imagined. One reason is that you need to spend energy to cross-verify it because of the nonsense; the other is that its own model corpus has a distinctive style (such as science fiction flavor), which is difficult to break into the style you want. At most, it will give you some Tips, but this will lead to very average results.

In my opinion, it can only serve as a brain-explosion agent at most. I usually use it to predict more.

–An independent game producer

During the planning process, we will communicate with Deepseek about some problems we encounter. It will give some suggestions and directions, especially the process of reasoning problems, which can help us find gaps and fill in gaps, or provide some new thinking directions:

–Game play planning of a Shanghai Four Little Dragons

I also often use tools such as Deepseek as search engines or self-censorship tools.

For example, after writing the plan, you tell it,”Now you play as a test engineer to check and see what marginal conditions I didn’t think of, so come and question me.”

Then I began to trick it into “working”.

The picture material comes from Grape King simulation. The game is fictional and is for demonstration only.

I think AI is more beneficial to individuals than to projects, because this link will also be correct when you connect and collaborate with other colleagues. However, after using these AI tools, you will be more “careful” in the eyes of everyone. Be reliable.

In addition, I personally have two main aspects of AI tools and applications: making business tools and building gameplay prototypes.

For the former, although our company has a special technical center to help plan and develop various tools, it may cost a lot of costs from demand raising, scheduling, communication, execution to implementation. Research and development with AI tools such as Deepseek and GPT will save a lot of time and energy.

For example, last time I spent an hour writing a relatively complex business tool that can realize data visualization and verification. Before I didn’t have this tool, it might take me a day or two to review the tabular data exported by these editors, but now it only takes 5 minutes.

Data verification tools

The situation of building a gameplay prototype is similar. Because many manufacturers have bottom-up projects, we need to review the boss whether the things we do are good or not. In the past, doing these things may require a plan or a PPT, but even if you talk too much, there is no way to verify it, and everyone still doesn’t know whether it will be fun or not.

Therefore, our planning needs to be a simple prototype of the gameplay. It may not need to be complete or packaged, as long as it looks like something and achieves the demonstration effect.

At this time, if you follow the original model and development cycle, it may take a week or two and labor costs, and you need to worry about not going through review and pulling changes back and forth after making it; but with the iterative evolution of AI tools such as GPT, we only need to split up what functions this gameplay prototype requires, and then talk to, optimize, and combine with the AI tools one by one… In the end, it may only take a day to make a gameplay prototype.

When I was doing these things, I was also thinking: Is this a “salary theft”? Because people around me really rarely use AI tools consciously to improve effectiveness.

But when I think about it, even if I really type code by hand, there will actually be prediction input tools like Copilot that can improve the efficiency by 20%-30%. This should be another “technical dividend”.

Copilot makes predictions based on the annotations, and the source is Zhihu @Jayden

02

Procedure: Even if you make up nonsense

get things done

–Narrative planning of a large factory (technical direction)

I don’t use DeepSeek much in my current programming work because it is too divergent and likes to remind me of issues that I don’t need to pay attention to.

First of all, my job is divided into two parts, one is to find solutions, and the other is to implement solutions.

On the former, I will ask: What methods are there and what kind of function can be achieved? At this time, the answers I get are basically enlightening and are still useful; but when I say that I think one of the plans is good, it will always revolve around this plan and cannot get out.

For example, let me ask first: In the GAS of Unreal Engine, how do you usually determine whether a Gameplay Ability can be triggered (for example, it can be triggered when the blood level is below 90%)? It will give me some valuable reference;

The source of the picture material is Grape King Simulation, for demonstration only

My second question is: Assuming that there is a source object and a target object, and I need to rely on the status of both the source object and the target object as an Ability availability condition (for example, the blood of both parties needs to be less than 90%), can I use the Gameplay Ability System to make this judgment?

DeepSeek gave me a complete solution: a combination of attribute listening, conditional checking logic, and network verification.

The source of the picture material is Grape King Simulation, for demonstration only

But do you know what I need most for the answer to my second question?

It’s “can’t”. Ability had no way to get Target’s information before activating it. As long as I knew this, I could give up this direction and find another way.

That is to say, for a specific question, sometimes I ask “can it”, and it will think of a way to give me a “can” answer. But if it really can’t, the way it thinks of may turn into a rumor, or at least a waste of time. Therefore, DeepSeek can only be used in plan selection, and it cannot jump out after entering a plan.

However, in terms of learning, assisting understanding, and story creation, DeepSeek’s effect is not bad, and it may be used more on the creative end.

03

Art: Master of Prompt Optimization

and creative implementation tools

–Technical Art (TA) of a game art outsourcer

I mainly use Deepseek in three aspects:

The first is to generate picture prompts, which can accurately understand my creative needs and transform them into professional Tips. When combined with other AI tools such as Claude and Imagefx, a standard poster can be generated in just 5 minutes.

1. Claude generates poster suggestions
2. Deepseek conducts analysis and optimization and generates accurate Wensheng prompt words
3. IMAGEFX generates pictures through prompts
4. Manual screening and detail adjustment

Secondly, I will use Deepseek to do some creative extensions, such as asking it to generate some game worldviews and core gameplay architecture for reference.

And the scene design and combat details under this framework:

Based on this prompt word, the card concept map generated using Capture:

The third is to satisfy its own horizontal development and creative implementation.

I am a full liberal arts student. However, with the help of AI, I have developed web pages, Mini programs, and even apps myself, and made some gadgets to make my work more efficient. Interestingly, where I work now, I have my own IT department, and they are much more professional than me. But after knowing that I can program AI, they sometimes come to me to ask me for some professional knowledge.

For example, I once used three free days to successfully develop decompression Mini programs using Deepseek, Windsurf and Cursor. During the development process, through dialogue with AI, it can accurately achieve a series of functions that I expect. In the end, Mini programs have interesting functions such as recording the mood of migrant workers, giving encouragement and reminders to migrant workers, breathing meditation, and encouraging twisting.

Of course, I also know that the above three AI workflows may have existed in the GPT era. However, Deepseek’s advantage is that it can fully understand my context and needs in the Chinese context, reducing the cost of repeated communication and the threshold of use. Especially when dealing with some slightly complex tasks, its reasoning capabilities and networking skills can provide a lot of ideas to expand.

–Art design of a large Guangzhou factory

Our company’s planning colleagues have often used AI to generate renderings, and then asked art colleagues to lean in the direction of the renderings.

The source of the picture material is Grape King Simulation, for demonstration only

But I always feel like this is not right to work.

02

Audio: Competitors, efficiency improvement scripts

–Lu Xiaoxu, founder and CEO of Xiaoxu Music

Foreign audio music generation mainly uses two softwares, Suno and Udio. After the launch of Suno v3.5 last year and v4 this year, it has become very mature in AI music creation.

However, it still has two shortcomings: First, the sound quality of the generated audio is low, and it can be listened to on a mobile phone, but the obvious difference can be heard when used in a console game or placed on a high-fidelity stereo; Second, Suno cannot yet be accurately controlled and modified, and the music generated in one go is determined. If you say how many seconds the piano needs to be changed to a guitar, it cannot be changed.

Therefore, Suno has little impact on our games, audio, music and other businesses. The demands we receive are still 100% made manually. At most, AI may provide some direction reference.

Not only that, after AI entered, many types of work in the audio music business were no longer needed. For example, game dubbing requires a recording studio. We used to have 4 recording studios in Beijing with 4 sound engineers, but now that TTS (Text-to-Speech) matures, many manufacturers are no longer here to record; for example, the production and singing of many game songs have been replaced by AI, and the workload of singers and CVs will be reduced quickly, and the number of talents will also decrease accordingly.

Netease Fuxi’s AI dubbing has been implemented in the mobile games of “Cold Against the Water” and “Eternal Tribulation”

In this case, I predict that our traditional business may shrink by 80% within three years. If we wait until AI creation is truly mature and can be commercialized, it will definitely not fall into the professional field or some experts. -People’s game companies can use AI to generate any music, so why should they spend money to find others to do it? AI replaces not labor costs, but a complete set of business processes.

So we are also changing our needs, such as making some content that combines AI audio and video to reduce substitutability; another example is trying some content that includes live performances and AI interactive performances; at the same time, I am also asking for the team’s comprehensive capabilities, such as composers The ability to play musical instruments, after all, live performances may be more valuable than music production in the future.

Short Video produced by Xiaoxu Music combined with multiple AI content are popular

As for the issue of anxiety you mentioned, maybe I would have been a little anxious last year, but this year will be relatively calm. After all, the general trend of AI is the anxiety of all mankind, not our family’s issue alone.

–Shanghai Dachang Audio Planning

I think there will be some prospects for connecting AI tools such as Deepseek to the creation tools.

I once used Deepseek to write an efficiency improvement script for the audio editing software Reaper. Simply put, it integrates some plug-ins commonly used in the software into an additional interface, and the corresponding plug-ins can be loaded with the click of a button, saving the time cost of reviewing the plug-ins one by one in the plug-ins list.

98% of the code in this enhancement script was generated directly by me using Deepseek and debugged and modified by talking to AI:

There were also some problems with this. After talking to Deepseek for a certain number of rounds, it began to make things up, and the output code became completely unusable. Fortunately, it had already provided a relatively complete and basically usable code. Based on this code, I made manual fine-tuning based on the Deepseek reasoning content, and finally realized the implementation of this small script.

In the later period of Deepseek chaos, replicate again

I can’t give it the full code

05

Operations: Better assistance than ChatGPT

–Northern Lights Game Operation

When I am looking for content ideas such as promotional copywriting and text messages, I use Deepseek, and I feel that it is much easier to use than ChatGPT at the time. However, when things like operating announcements are directly aimed at players, once they are released to millions of people, they must definitely start them in person, and there cannot be any mistakes at all.

Deepseek directly generates Valentine’s Day promotion text messages (game information is fictional and is for demonstration only)

–Tianmei Game Operation

We have many AI tools internally. But I don’t use it much, and I will use it to check information, but I just take a look and provide assistance. I used it to tell fortune some time ago… It is difficult to meet demand by working properly.

06

conclusion

Judging from the practical operations of the above-mentioned practitioners, Deepseek and other AI tools have provided some help to personal work, far from bringing some revolutionary changes to the traditional game development process.

This conclusion is also within expectations. As Lu Xiaoxu said, AI is an affirmative action tool. Its original intention may not be for professional people to use it, but for laypeople to do professional things without barriers.

A game art outsourcing boss

So, the actual impact of AI on the gaming industry is not that great?

That doesn’t seem right. Chen Chao, CTO of Qingpao Network, believes that he feels that everyone’s understanding of Deepseek is similar to when other AI tools first came out: it overestimates its short-term effect and underestimates its long-term effect-the impact of AI workflow on the game industry is not so much a revolution as a penetration.

These “infiltrations” are like practical explorations by the above-mentioned practitioners. They may not seem serious, but they may become the early accumulation of AI’s transformation of the game industry. Cao Xiaowen, founder of Qingpao Network, also said that many manufacturers have used AIGC to replace manual production of art assets early on. They may not have achieved the right ROI yet, but through early accumulation, they have raised the level of AI in the game industry.

“If the human structure of the game industry is a pyramid, then AI is a primary neural network. It will be submerged from the bottom up like seawater. The first thing to be replaced may be just some scraps, but with the continuous development of AI, and under the influence of the characteristics of the pyramid structure, the seawater will be submerged faster and faster, even beyond our imagination; this kind of submerged is not just a one-to-one replacement, but it is likely to change the working structure, model, and culture together.”

“When the sea water of AI does not pass the middle reaches of the industry, a greater turning point may come.”