300 Titles and a 30% to 40% Saving: Netflix's AI Push Reaches Crews

Netflix has used AI in some form across about 300 titles, a production executive said at Busan, bringing the technology's effect on film work into an immediate industrial debate. ^[1]

Sung Q Lee, a Netflix production executive for Asia-Pacific, gave that figure at an October 10 panel at the Busan International Film Festival. Director Kang Yoon-sung separately put a striking number on the financial promise: feature-film workflows could save 30% to 40%, with his fundraising calculation for The Gulf of Aden suggesting a reduction of around 40%. ^[1]

Netflix and the Motion Picture Association co-hosted the discussion as part of the festival's Creative Asia program. The participants included Lee, Kang and Kimball Thurston of visual-effects company Wētā FX. This was an industry conversation about tools already entering production, alongside a director's case for financing a film with them. ^[1][2]

The labour risk deserves attention now. A tool that helps get an unaffordable film made can also strengthen a producer's argument for fewer shooting days, fewer effects tasks or fewer paid hours.

Those possibilities follow from the savings Kang described. They are not yet a count of jobs lost.

The figures have firm limits. The account gives no title-by-title breakdown of Netflix's use, staffing comparisons or evidence of displacement.

Kang's estimate comes without a budget, cost breakdown or independent check. Netflix did not announce a 40% saving or instruct filmmakers to cut budgets by that amount. ^[1]

Lee said Netflix tests new tools under strict internal guidelines and is not "wildly obsessed" with adopting them quickly. ^[1] That reassurance concerns how the company introduces technology. Workers also need to know what happens after it arrives.

What 300 titles actually tells us

Lee identified previsualisation, production design, advance set-dressing concepts, editorial and visual effects as areas where Netflix had used AI. ^[1] These cover several stages of making a film, from planning what the camera might see to shaping the finished image.

Previsualisation lets filmmakers sketch out a sequence before shooting it. Set-dressing concepts help establish what a location or set should contain. Editorial shapes footage into a finished story. Visual effects create or alter elements of the image. AI can enter any of those processes without being responsible for the whole production.

The 300-title figure therefore demonstrates reach. It cannot serve as a measure of how much filmmaking has been automated.

That is still a substantial disclosure. The debate has an operational starting point: Netflix says the tools have entered hundreds of titles, across multiple kinds of work. Producers and workers can now ask about specific production decisions rather than argue only about hypothetical capabilities. ^[1]

Lee showed a reel about Eyeline's hybrid approach, which combines live-action photography, visual effects and generative tools. He also presented large crowd scenes created with AI and VFX for Netflix's Indian title Glory. According to Lee, those scenes would otherwise have been dropped because of tight budgets and schedules. ^[1]

That example gives the strongest case for adoption. The alternative Lee described was a film without the crowd scenes. A cheaper method preserved something the production wanted to put on screen.

It also identifies the decision point that matters. When a tool makes a sequence affordable, who does the work that remains, how much are they paid, and how much time do they get? A producer's account of a rescued scene is a useful beginning. A labour account would follow the money and hours through its production.

Forty per cent is a financing argument

Kang's estimate for The Gulf of Aden reaches beyond replacing computer-generated imagery and visual-effects work. He said the potential savings also came from simplifying the shoot and shortening the overall production period to some extent. ^[1]

That is why the number matters to workers outside an effects department. A shorter shoot can affect the amount of work available to the people who build, light, dress and run it. A shorter production period can change how long specialists are needed. These are possible consequences of the proposed workflow, not reported outcomes for this film.

On Kang's estimate, a project costing 100 units under its comparison budget would cost about 60. The arithmetic is simple. Delivering the same film for that sum is the proposition that needs testing.

Kang said AI had brought projects that had fallen over at the budget stage back into consideration. He argued that it could make more genre and science-fiction films possible amid difficulties in the Korean industry. ^[1]

That argument has weight. An unmade film offers no production work at all. If a cheaper workflow unlocks financing, it could create work that would otherwise never happen. Workers have a stake in productions being viable, just as filmmakers do.

But that benefit does not settle how the savings should be distributed. Producers can use a lower cost to finance another scene, support another film, improve working conditions or demand a smaller budget. The technology supplies options. The contract decides who benefits.

The threat to below-the-line workers - the crews and specialists who carry out a production - is therefore a bargaining threat before it is a documented redundancy total. Once a director publicly puts a 30% to 40% saving into the financing conversation, workers have reason to challenge the assumptions behind it. ^[1]

A percentage can travel farther than its evidence. Kang's estimate belongs to his proposed workflow and project. Treating it as a standard discount that every filmmaker should deliver would turn a fundraising calculation into a budget mandate.

A pixel repair is not a generated creature

The panel also showed why "AI use" needs a better description than a single label. Thurston discussed a deep-learning denoiser used by Wētā, an assistive tool that cleans up pixels without changing creative intent. He said the company had used machine learning for at least a decade, or 20 to 25 years if early forms of its Massive system were included. ^[1]

A denoiser can improve an image while leaving the creative choices intact. A system that generates a creature or a scene enters the work differently. Both can be described as AI, but that description alone says little about authorship, staffing or the amount of human effort involved.

Kang supplied an example of the latter. He co-directed Run to the West with AI filmmaker Kwon Han-seul and said AI substituted for CG, mainly in compositing and creatures. Compositing joins separate visual elements into one image. Kang described the tools at that stage as rudimentary. ^[1]

He said their output had improved enough in roughly a year to become commercially usable. His test footage for The Gulf of Aden was generated entirely with Kling, an AI video-generation tool, and rendered at reduced quality. He said current tools could produce 4K images at 10-bit colour or higher, describing a higher-resolution output with a broader range of colour information. ^[1]

For production finance, commercially usable is the crucial phrase. A tool need not master every task to change a budget. It needs to perform selected tasks well enough that someone is willing to build a schedule around it.

Lee's caution about the pace of adoption is welcome, but it should lead to specific disclosure about which tasks Netflix considers suitable and how it evaluates their effects on production work. ^[1]

The useful dividing lines are practical: assistance or generation, time saved or work removed, a scene added or a crew engagement shortened. Those distinctions would make the 300-title number far more informative.

The cost of removing the conversation

Kang's warning about creative work was more searching than a defence of image quality. He said AI could remove the discussion, effort and experience of many people involved in filmmaking, leaving images shaped by one person's ideas. He also said technically strong AI footage could be weaker at conveying emotion and moving audiences. ^[1]

That is a criticism of how a film gets made. Crew members bring judgment as well as labour. Their contribution includes noticing a problem, proposing a solution and pushing back on a choice. Removing work can also remove those opportunities to improve the film.

Lee said some tasks now take a single click, while agreeing that human involvement remains essential to conveying emotion. Kang expected short-form production to put greater emphasis on productivity, with film and drama placing more weight on credibility. He also saw AI's role around actors as more likely to involve stunt and stand-in work than replacing performers. ^[1]

Those views suggest selective adoption rather than a complete automated production. Selective adoption can still have large consequences for the workers whose tasks are selected.

The case for cheaper filmmaking is credible enough to take seriously. The case for making workers absorb its savings has not been made. A production that becomes viable through new tools can be a gain for filmmakers, crews and audiences, but only if viability is not used as a permanent excuse to weaken the people making it.

The next reporting should move from percentages to production records. Netflix needs to define what qualifies a title for its 300-title total and identify changes in crew size, paid hours, schedules and budgets.

Kang's team needs to show the baseline and assumptions behind The Gulf of Aden estimate. Those are the numbers producers and unions need before a possible saving becomes an expected cut.