Dialogue with Jurilu AI Siyu: The era when one person finishes one drama series per day has gone | NEXTA Innovation Night Talk

The production speed of AI short dramas has already surpassed the editing speed.

An 80-episode short drama with a budget of 2000 yuan can be started for production before off-duty, and the AI-generated episode content will be available the next day.

This is a production scenario described by Si Yu, co-founder of Juri AI, a leading AI short drama creation platform, at the NEXTA Innovation Night Talk.

Under the "Script In, Drama Out" model that Juri AI is advancing, users only need to input the script and confirm the budget, and the system will generate a production proposal showing character settings, picture styles and partial shot effects. After confirming the plan, you can start batch generating all the complete episodes.

Just a few months ago, the "One Person, One Drama a Day" concept proposed by Juri AI already surprised many traditional film and television practitioners. Now, they are starting to hand over more production links to AI.

Changes are coming faster than many people expected.

In March this year, Si Yu shared the "One Person, One Drama a Day" concept at the TV Festival and Short Film Festival events in Shenzhen. Facing the traditional film and television directors in the audience, he was a little worried that the statement was too ahead of its time. By the Shanghai TV Festival in June, the industry's acceptance of AI had increased significantly.

The speed of tool updates is also changing the employment logic of the film and television industry. AI directors who were once fought for by headhunters last year can now manage more projects with the help of workflow; once sought-after producers are also facing job changes. The production team of some AI short drama projects can even be reduced to only two people.

Juri AI is a platform focused on the large-scale production of AI short dramas. According to Si Yu, the platform's clients have accumulated a large number of hit dramas with over 100 million views, covering short drama production companies of all types and scales.

However, the continuous decline in production costs has also brought new competition to the industry. Si Yu found that some content companies may have achieved good results in half a year, only to be overtaken by teams adopting new processes a few months later. After the production threshold is lowered, high-quality scripts, excellent content judgment and distribution channels still directly determine whether a drama can make a profit.

When the production of short dramas becomes increasingly automated, how will the organizational structure of film and television companies change? With model capabilities strong enough, what else can AI application companies rely on to build competitiveness? What opportunities are there for individual creators?

In this issue of NEXTA Innovation Night Talk, we invited Si Yu, Founder and CEO of Juri AI, to talk about the fast-changing AI short drama business.

The following is the transcript of the conversation, sorted and edited ——

"One Person, One Drama a Day" Is Already a Thing of the Past

NEXTA Innovation Night Talk: What is the difference between Juri AI and other AI video tools on the market?

Si Yu: From the very beginning, we wanted to build a platform for mass-producing AI short dramas.

The success of short drama content has considerable randomness. For the same theme, if different teams produce it, or release it on a different platform or at a different time, the final data may be completely different. So we have always paid close attention to automated and industrialized production. Since there are many factors affecting the success of content, it is very important to improve efficiency and reduce trial-and-error costs.

We previously developed products related to novel text promotion, and accumulated hundreds of thousands of users. From mobile phone novels and novel commentaries, to graphic text promotion and PPT commentary comics, content has always been developing in a more intuitive and visually expressive direction. Back then, we were waiting for the video model to reach a critical point, so that stories could be directly turned into videos.

NEXTA Innovation Night Talk: What problem in the short drama industry did you think AI could solve first back then?

Si Yu: The most direct one is efficiency and cost. In the past, no matter how fast live-action short dramas were shot, it would take 5 to 7 days to finish one drama. Actors, venues, costumes, props and makeup all need to be coordinated.

Just imagine, if a company is filming 50 dramas at the same time, how busy the production team will be? Some are produced by its own team, and some are outsourced to other teams. Coordinating all these resources alone will take a lot of energy.

After AI is introduced, many links that originally required coordination are eliminated. So when our technical capabilities realized efficiency iteration under sufficient effect, we took the lead in putting forward the concept of "One Person, One Drama a Day" in the industry. In March this year, I participated in an industry event in Shenzhen, where many traditional film and television teachers and directors sat in the audience. When people first heard this concept, it must be hard to imagine. By the Shanghai TV Festival in June, I clearly felt the difference. The traditional film and television industry's acceptance of AI has increased a lot. But now we have actually taken a step further.

NEXTA Innovation Night Talk: You are no longer satisfied with producing one drama a day?

Si Yu: Yes. We are now working on a process called "Script In, Drama Out".

For example, you give me a script with 80 episodes in total, a budget of 2000 yuan, and plan to release it on an overseas platform, which is a volume-focused drama. The system will first give you a proposal based on the budget, select some clips for trial production, so that you can see what the characters look like, what the lighting is like, and whether the camera movement meets expectations. If you think the camera movement is not rich enough, or the character setting is not quite right, you can continue to put forward suggestions, and it will give you another version. When you think it's almost done, click to start production.

You can even confirm the plan before off-duty, and come back to get the drama the next day. The whole 80-episode content can be run by the system in the background. In the past, under the "One Person, One Drama a Day" mode, you still had to check how each episode was produced. Now most of the production links can be handed over to AI.

NEXTA Innovation Night Talk: Why didn't you adopt the canvas mode used by many AI video tools?

Si Yu: Because our goal is industrial mass production, built for teams and mass creation. Canvas is of course very good, creators can freely adjust at each node. For content that requires strong personal expression, there is no denying that canvas is a very good choice.

But the production logic of short dramas is different. Many production features have a very short "flowering period". You work hard to develop a feature, but two months later, the capability of the new model comes out and directly replaces it. So we prefer to focus on automation. Our idea is a bit like self-driving: most of the time the work can be handed over to the system, but if you are not satisfied, you can take over and modify it at any time.

Production Tools Bring Efficiency Leap, Creative Teams Need Continuous Evolution

NEXTA Innovation Night Talk: As AI becomes more and more automated, how will it change the team structure of film and television companies?

Si Yu: In fact, it is already changing, and very fast. Now when we give suggestions to clients, we have to adjust them every two or three months. Because after the production speed increases, the original staffing may no longer be suitable. For example, if the speed of AI content generation has exceeded the editing speed, the proportion of team members must be readjusted.

In the past, when creating character assets, there might be a dedicated team for character generation. If the generated result is not satisfactory, you have to use Photoshop to retouch the image. Now we directly generate six versions at a time, continue to generate if you are not satisfied, and the unselected ones will be deposited into the "Enterprise Asset Public Pool". Moreover, every choice you make will be recorded, and the system will gradually learn your aesthetic preference. A lot of experience that was accumulated manually in the past can now be gradually deposited into the workflow.

NEXTA Innovation Night Talk: If you want to form a minimum team that can deliver stably, how many people do you roughly need?

Si Yu: I think the minimum unit only needs two people. One is responsible for art and character assets. This is very important. We often tell clients that you can't make a 100-point work with 80-point assets. Whether the character looks good, matches the style of the whole drama, and can impress the audience, all require aesthetic judgment.

The other one is a senior editor, who is better to have screenwriting and directing thinking, understand both scripts and how to direct the scenes, and complete the delivery in one integrated process during editing.

NEXTA Innovation Night Talk: Why do people still need to do the final editing? Can't AI finish it now?

Si Yu: Of course AI can do editing. The problem is, is the commercial effect of AI-edited content really better than that of human-edited content?

For example, if AI edits 500 pieces and humans edit 500 pieces, you can't judge which one is more profitable just by sitting there and watching. You have to actually launch all these 1000 pieces for distribution, which may cost 1 million yuan, and finally you can see which one has higher recycling efficiency. But how many bosses are willing to spend this money to specifically help technology companies verify the effect of automatic editing? So there is not yet sufficient data to prove that AI editing is definitely better than manual editing.

NEXTA Innovation Night Talk: AI directors and character generators that were very sought-after last year, are they still scarce now?

Si Yu: When AI dramas were at their most popular last year, good directors were really very expensive. Everyone wanted to poach AI directors with hit works, and even needed to find them through headhunters, and job hopping was very frequent.

But the situation has changed now. In the past, one director might lead two teams, but now one director can lead as many as ten teams. Some companies don't even have dedicated directors anymore. Because the experience of directing scenes and splitting storyboards can be gradually deposited into the intelligent agents. As for character generators whose aesthetic level is below the average, I think they are not that important now. Creators need to realize this problem.

This industry is very exciting. A company may be doing extremely well in the first half of the year, but after three or six months, a new process comes out, and it is suddenly overtaken by others.

Models Are Already Over-configured, What Do AI Short Drama Companies Rely on to Compete?

NEXTA Innovation Night Talk: What are the remaining capability shortcomings of current video models?

Si Yu: For the short drama scenario, I think the current models are already a little over-configured. For example, you drive to and from work every day, a car worth 200,000 yuan is already very easy to use. Now if you are asked to drive a car worth 500,000 or 1 million yuan, it is of course better, but do you really need such high configuration?

The same is true for short dramas. The capabilities of many current models are already redundant. After all, short dramas are not theater movies, and there is no need to polish every single shot. So production teams should pay more attention to content capability, art capability, organizational capability, and understanding of scripts in the future.

NEXTA Innovation Night Talk: Since the models are already sufficient, what is the most important breakthrough Juri AI wants to make next?

Si Yu: Model orchestration itself is what we are very good at. Juri AI has strong engineering capabilities, and currently we are focusing on polishing model routing. There are so many models on the market, each of which has its own strengths. For example, for a simple dialogue shot, there is no need for me to use the most expensive model, and adaptation is more important.

But if it is a complex fight scene, I can assign this shot to a model that is better at fighting scenes. This shot may cost a little more, but the total cost of the whole drama is even lower. Our ideal is that in the future, users don't need to worry about these things. You tell me the budget and the effect you want, and I will deliver the satisfactory drama for you.

NEXTA Innovation Night Talk: The underlying models are constantly upgraded, and some features originally developed by application companies may be replaced soon. How do you deal with this?

Si Yu: We have always maintained sufficient sensitivity to new models and new AI-native capabilities.

As soon as a new model comes out, the team will test it immediately. Our operations staff will pay for advanced memberships themselves first. Because you never know which update hides a new business opportunity or process revolution. When I interview candidates, I also like to ask, what AI products have you paid attention to recently? What have you subscribed to with your own money? Excluding those paid by the company, I want to know what you are willing to pay for yourself.

As for how to compete with big manufacturers, I think it depends on how deep your understanding of the industry is. The models that everyone can use are almost the same, just like buying vegetables, you all get cabbage, radishes and potatoes, but what dishes you can finally serve depends on your own engineering capabilities, understanding of the industry and users.

Production Is Getting Cheaper and Cheaper, Who Can Make Money?

NEXTA Innovation Night Talk: Is AI short drama already a profitable business now?

Si Yu: The business model has always been feasible, but many people can't make it through the whole process. For short dramas, I think the three most important things are: IP, production and distribution. First of all, you need a good script. Some people have bad scripts, no matter how beautifully they shoot, they may not get good data.

Secondly, production. For the same script, different workflows and different models will lead to different final effects and different costs. I often tell clients not to always focus on the difference of a few cents in computing power cost. Continuously upgrading and optimizing the production process, reducing the number of personnel and saving working hours is much more valuable than saving those few cents.

NEXTA Innovation Night Talk: To what extent has the cost been reduced?

Si Yu: Some of our low-cost tiers under the "Script In, Drama Out" mode can already achieve two to three cents per second, which is equivalent to more than ten yuan per minute. The production cost in the era of manual character generation last year was dozens of times of the current level. There is still room for further reduction in the future, but it also depends on how the upstream costs such as GPU and chips change.

I also hope that more and more open-source models will emerge. If they are deployed on our own computing power and combined with model routing, the cost may be further reduced. All in all, cost reduction should rely on engineering capabilities, to use the right model in the right place.

NEXTA Innovation Night Talk: What is Juri AI's own business model? Why not do IP and distribution by yourself?

Si Yu: Our charging model is relatively simple, calculated based on points consumption. You recharge as much as you use, and recharge again after using up, which is a bit like topping up phone credits. As for why we don't do distribution ourselves, I think every company has its own strengths and team genes. We are a team with great advantages in product development. We still hope to make good tools, empower our clients, and build a good ecosystem. Our team's gene is R&D, and we will set very high standards for our products, to make them faster, cheaper and easier to use. Compared with building a distribution platform with annual consumption of tens of billions, I think we are more willing to focus on technology.

NEXTA Innovation Night Talk: As production costs get lower and lower, will short dramas become more and more homogeneous?

Si Yu: I don't really agree with the idea of homogeneity. Because every theme has its own cycle. Who could have imagined that themes like business management and farming would be so popular in a certain period of this year? There are also stories about mathematical geniuses, international competitions and so on, which might not have attracted much attention before.

Now with AI, the production cost is low, so we can try more themes that have never been done before. I even think that very niche AI short drama apps will appear soon. For example, an app full of dramas suitable for the elderly, with 3000 dramas is enough.

NEXTA Innovation Night Talk: Why not just put it in a channel of the existing platform?

Si Yu: Because some hobbies are really very niche. But these people are particularly immersed in this theme. Look at ACG and various subcultures, they are all very segmented. I think there is still a lot of content that has not been fully developed.

NEXTA Innovation Night Talk: Is there a bubble in the AI short drama industry now?

Si Yu: I don't think so. At least from our experience, everyone's production demand is still very strong, and the audience is willing to watch.

And the