About 300 Netflix works have used AI, saving an estimated 40% of the budget in The Gulf of Aden.

 AI has already entered the production process of approximately 300 Netflix productions.Lee Sung-gyun, senior director of Asia-Pacific production at Netflix, disclosed this figure in a conversation at the Busan International Film Festival; Jiang Yuncheng, director of The Outlaws on the same stage, said that based on his own project calculations, the AI ​​workflow can reduce the production budget of a feature film by 30“ to 40”.

This “Film Production and AI: New Tools, New Possibilities” dialogue was co-organized by Netflix and the American Film Institute and is a Creative Asia event for emerging talents in Asia.Wētā FX technical director Kimball Thurston also participated in the discussion, which was moderated by film critic Yun Sung-eun.

Lee said that AI is currently being used for previews, production design, exploring set concepts in advance, and also in editing and visual effects.He played a demo video from Eyeline, a special effects studio owned by Netflix, showing hybrid production that combines real shooting, special effects and generative tools.Another clip is from an Indian work - Glory - (tentative translation), where AI and visual effects work together to complete large-scale crowd scenes.When budgets and schedules are tight, scenes like these are often abandoned.

Jiang Yuncheng's experience starts from the last movie."Run to the West" (tentative translation), which he co-directed with AI filmmaker Kwon Han-seul, was released in South Korea in October last year and is positioned as South Korea's first feature film to use AI in the entire production process.At that time, he mainly used AI as an alternative tool to CG for synthesizing and creating monsters, but it was difficult for the generated creatures to naturally integrate into real-life scenes.

A year later, he believes the quality of the technology has improved to the point where it can be applied commercially.

At the beginning of this year, Jiang Yun became the production test film for the maritime military action movie - The Gulf of Aden (tentative translation), all produced using Chinese software Keling.The test film used a lower output quality. He said that existing tools can already output images with 4K, 10-bit color depth or higher specifications, which can meet the needs of theaters.

When asked about the cost savings, he gave an estimate of the financing stage: We calculated how much production costs could be saved in the Gulf of Aden, and the actual reduction was about 40“.” budget changes involve CG and visual effects, as well as the simplification of the shooting process and the shortening of the overall production cycle.This 40“ figure comes from project calculations in the Gulf of Aden.

For Jiang Yuncheng, some plans that were originally shelved due to lack of budget can finally be discussed again, especially genre films and science fiction films.But he also feels uneasy: in the past, film production relied on the discussion, labor and experience of many people. After AI takes over a large number of links, the picture may only be the idea of ​​one person.As the image quality gets higher and higher, the emotion that impresses the audience may become weaker.

Lee agrees with this assessment.He said that some results are now available with a click, but when it comes time to convey emotions, people still need to process them again.He regarded human intervention as a necessary link, and also pointed out that new tools help solve the problem of how to make a work well, and the reason for telling this story is still determined by the creator.

The example Thurston gave came from assistive AI”.The deep learning noise reduction tool that won him his second Academy Award for Science and Technology can correct pixels but cannot change the creative intention.Wētā has been using machine learning for at least a decade; that history goes back twenty to twenty-five years, if you include early forms of the Massive crowd simulation system developed for The Lord of the Rings.

There are still gaps in color accuracy.He made an analogy: A model trained using only video images may not be able to restore the accurate red color of a Coca-Cola can. What you get is the red color of Netflix, but not the red color of Coca-Cola.

Tools are updated quickly, and both companies have their own adoption procedures.Lee said Netflix follows strict internal guidelines, tests new tools first, and doesn't go crazy for speed.Thurston revealed that Wētā is developing open source tools through the Academy Software Foundation to track the use of AI and the origin of materials, and is preparing to connect with mechanisms such as C2PA.

Sustainability was also talked about on site.Thurston said Wētā's data center in Wellington, New Zealand, uses closed-loop cooling and currently uses zero water per year.Lee said that Netflix promotes sustainability goals at the company-wide level, including the 2030 carbon reduction plan, and does not study the impact of AI separately.

Jiang Yuncheng predicts that short content will pay more attention to production efficiency. He mentioned that China already has short content projects that use AI from planning to distribution; movies and dramas will pay more attention to credible performance.He is developing an animated feature film with photo-realistic backgrounds and wants animated characters to move like real actors.

As for actors, he believes that AI is suitable for stand-ins and stunt work, and it is still difficult to replace performances.Even if the imitation is very accurate, the mechanical feel may block the transmission of emotion.Thurston also emphasizes that audiovisual storytelling requires the study of the human condition.At the end, Lee brought the topic back to creation: “ is still written by people in the end, and we hope to see more stories told by people.”