At the end of February, the first animated series created with generative artificial intelligence premiered in China. There have been other audiovisual creations that have used AI in part or throughout the production, but in the case of the Chinese series Qianqiu Shisong, it involves several episodes that follow a certain theme and aesthetic.
In total, Qianqiu Shisong consists of 26 episodes, each about 7 minutes long, in which different traditional stories from Chinese culture are animated. This anime made with generative AI joins other audiovisual creations that have already used this technology in one or more parts of the creative process, both for video production and in 3D creation with AI.
For the creators of the series, combining narration with animation is a more dynamic and vivid way of introducing classic poems, which children enjoy more, and therefore, it is a better way to promote traditional Chinese culture.
How the AI-animated series Qianqiu Shisong was made
The AI-animated series Qianqiu Shisong is being broadcast on China Central Television (CCT), which is part of the public entity China Media Group (CMG). Together with the Shanghai Artificial Intelligence Laboratory (SAIL), CMG has developed its own GPT, CMG Media GPT, capable of generating video from text (text-to-video), in line with other models like Sora.
CMG Media GPT has been trained with large amounts of poems and other popular literary works from the Chinese tradition, as well as videos and other materials from the immense archive of the public network.
In the presentation of Qianqiu Shisong, SAIL researcher Wang Yanfeng, assistant director of the series, said that CMG wanted to take advantage of various AI tools to create new audiovisual productions that meet the country’s needs. Moreover, CMG Media GPT is part of the Chinese government’s plans to promote the use of AI tools in public management bodies. According to Wang, the project began 6 months earlier to finally present a series of anime developed from CMG’s archive and the use of text-to-video AI technology, meaning it was generated with prompts after training a machine learning model.
The premiere of Qianqiu Shisong coincided with CMG’s introduction of its new AI audiovisual production company. According to the newspaper South China Morning Post, this new studio of the public broadcasting corporation aims to boost research and development of more AI-based television programs.
Other examples of audiovisual creations with generative AI
China’s commitment to generating audiovisual content with AI is not unique. In fact, although it might be considered the first series, Qianqiu Shisong is not the first attempt to create an animated story with AI in a major production.
Last year, Netflix Japan released the animated short Dog & Boy, although its reception did not garner the same enthusiasm as the Chinese AI-made series. The problem probably stemmed from the company claiming that AI was used in this animation project due to a “labor shortage” of specialists in Japan. Understandably, this excuse was not well received by professionals in the Japanese animation industry, and Netflix had to clarify that the short film was an “experiment.”
Another project with an experimental feel, though it is likely to become common sooner rather than later, is the AI capable of creating new episodes of popular animated series like South Park, in which, for example, we could be one of the protagonists. A San Francisco-based company called Fable Studio presented a project in which they explained their attempt to generate new high-quality content for intellectual property using large language models (LLMs), custom diffusion models of the latest generation, and their own multi-agent simulation for contextualization, narrative progression, and behavior control.
In the essay on this project, titled “To Infinity and Beyond: SHOW-1 and Showrunner Agents in Multi-Agent Simulations,” it is noted that the LLMs used, like GPT-4, have been trained with large amounts of data from TV series, so developers believe that, when properly guided by users, these models could rewrite entire seasons of TV shows. They cite Greg Brockman, co-founder and president of OpenAI, speaking about the controversial ending of Game of Thrones: “This is how the world of entertainment will be […]. Imagine you could ask an AI to create a new ending that ends differently and even puts you in it as the protagonist.”
Although the potential for AI-generated audiovisual content seems limitless, there are, for now, certain limitations. That same article also points out the current creative limitations of generative AI systems. Specifically, it discusses the introduction of an element of unpredictability and lack of creative control in this type of content, and the “slot machine effect,” which refers to a situation where AI-generated content feels more like a game of chance than a deliberate creative process. The results, like in these games, provide an immediate “high,” but can be detrimental to achieving a specific creative goal in the long term.
If you want to achieve professional AI-generated animation content that can be integrated into a commercial, educational, support, or any other type of strategy, it is necessary to have a vision, knowledge, and experience in the creative arts sector to use generative AI models for what they are: tools to help achieve a specific goal more quickly and effectively.
How models are trained, how they are used, or for what purpose are questions that industry professionals know how to answer. At Many·Worlds, we know this because we have been using the latest technological tools for over 15 years to develop the specific project of each of our clients. For us to advise you and help you develop yours, all you need to do is get in touch with us so we can start talking and finding solutions.
