Netflix generative AI: how studios use it and what changes
VFX, post-production and one model per production: how Netflix and other studios use generative AI in 2026, and what changes in production contracts.
Netflix uses generative AI mainly in visual effects and post-production, not to replace actors. Its first declared case was El Eternauta, where in 2025 a building-collapse sequence was finished ten times faster; in March 2026 it bought InterPositive, Ben Affleck’s company that trains a model on the footage of a single production.
For anyone producing for a platform, the way of working changes: data, permissions and approvals are settled before the shoot. Here is what Netflix and other studios are doing as of November 2026, where AI sits in the pipeline and what to ask for in the contract.
How Netflix uses generative AI
El Eternauta, the first case
The Argentinian series, released in April 2025, needed a building to collapse in Buenos Aires when budget and schedule were almost exhausted. According to TheWrap, Eyeline’s Kevin Baillie estimated that traditional VFX would have cost about ten times more, with delivery due in ten days. On the July 2025 earnings call, Sarandos said the sequence was completed ten times faster.
His message, reported by TheWrap, was explicit: “AI represents an incredible opportunity to help creators make films and series better, not just cheaper”. In August 2025 Netflix also published guidelines for its production partners on using generative AI.
Eyeline, the in-house unit
In October 2025 Netflix folded Scanline VFX into the Eyeline Studios brand, with three divisions: visual effects; Studios, covering virtual production, volumetric capture and generative AI; and Labs, for machine learning and computer vision research. Eyeline had already used volumetric capture on the second season of Wednesday.
At that point Eyeline did not build its own models: it used commercially available video models under enterprise licences. Netflix named no partners, but Bloomberg reported that it had tested Runway’s tools.
InterPositive, one model per production
On 5 March 2026 Netflix announced the acquisition of InterPositive, founded by Ben Affleck. Its model trains only on the dailies of a specific production and is used to replace missing shots, reframe, correct lighting, and replace or enhance backgrounds. The team joins Netflix and Affleck becomes a senior adviser.
According to Bloomberg, as reported by TheWrap, the price could reach $600 million, partly tied to performance targets. Elizabeth Stone, Netflix’s chief technology and product officer, stressed that this is not text-to-video prompting and that adoption will be led by creators, not imposed by Netflix. The tool is still at an early stage and is being integrated into Eyeline, as a TheWrap analysis explains.
Other studios: Disney, Lionsgate and the independents
Netflix is not alone. On the 5 August 2026 earnings call, Disney CEO Josh D’Amaro said AI helps above all in pre- and post-production, improves personalisation on Disney+ and is being tested in ad creative. On 9 November 2026 Disney announced a multi-year alliance with Adobe for models customised on its properties, with projects already under way at StudioLAB, part of Walt Disney Studios. The full story is in our piece on Disney, OpenAI and Sora.
Lionsgate handed its library to Runway in 2024 for a private model. By September 2025, people familiar with the project told TheWrap that a single library was not enough; the studio replied that the deal was not exclusive and that it plans to use Runway’s and other companies' tools in pre- and post-production.
Outside the platforms there is Doug Liman’s Bitcoin: Killing Satoshi. According to TheWrap, it was shot on a grey-box stage in London, with sets and lighting generated in post around the actors' real, unaltered performances. The producers put the budget at $70 million, against more than $300 million for a traditional shoot.
| Who | What AI does | Stage | When |
|---|---|---|---|
| Netflix | Building collapse in El Eternauta | VFX | 2025 |
| Netflix | Model trained on dailies (InterPositive) | Set and post | March 2026 |
| Disney | Custom models with Adobe | Pre and post | November 2026 |
| Lionsgate | Runway and other tools | Pre and post | Since 2024 |
| Doug Liman | Generated sets and lighting | Post | 2026 |
Where AI enters the pipeline, and where it does not
Line the cases up and the picture is consistent. Generative AI works where footage would be expensive or impossible to shoot, and leaves performances alone.
- Pre-production. Sets start from human drawings: on Liman’s film they were designed traditionally and then fed to the AI as references. We cover this in our piece on AI storyboards and previz.
- On set. Virtual production and volumetric capture, as on Wednesday, sit alongside footage shot to be completed in post, like Liman’s grey box.
- Post-production. Missing shots, reframing, lighting, backgrounds, destruction and set extensions: almost every case above lands here.
- Language versions. Dubbing and lip-sync are a chapter of their own, covered in our guide to AI dubbing.
- Performances. They stay human. InterPositive does not generate scenes or actors from a prompt, and Liman’s film leaves the performances untouched.
On synthetic performers the US actors' union is clear. In September 2025 SAG-AFTRA reminded signatory producers that its contracts require notice and bargaining whenever a synthetic performer is used, as TheWrap reported.
Netflix defends its titles
The company buying AI technology also guards its characters firmly. On 17 February 2026 Netflix wrote to ByteDance calling Seedance 2.0 a high-speed piracy engine and threatening litigation, according to TheWrap. The letter cited content inspired by Stranger Things, KPop Demon Hunters, Squid Game and Bridgerton; Disney, Warner Bros. and Paramount had also sent cease-and-desist letters.
For a producer the consequence is practical: no third-party characters, worlds or costumes in prompts, and tools whose training data you can account for. InterPositive, for instance, says it used no public or copyrighted material to train its model.
What changes for producers
- Read the commissioner’s rules. Netflix published guidelines for partners in August 2025: read them, and agree every use of generative AI in writing before you start.
- Use enterprise-licensed tools. Eyeline worked with commercial models under enterprise licences, not consumer apps.
- Decide who owns the model. If a model trains on your dailies, the contract should say who holds the weights, who may reuse it and when it is deleted.
- Keep faces and voices out without consent. Performances are the most sensitive point, legally and with audiences.
- Plan time and versions. A saving like El Eternauta’s should go back into revisions and quality control, not shortcuts.
- Communicate clearly. In an NRG study with TheWrap published in March 2026, 56% of respondents said AI actors would never be as good as human ones.
Human direction, new tools
As of November 2026, the story of Netflix and generative AI is one of post-production rather than replacement: models trained on a production’s own footage, faster effects, untouched performances and intellectual property defended hard.
It is the same approach we take at Kult Studios: the director decides, AI executes, and the client approves every step from pre-production to the edit, as our process sets out. For a brand film or a series, the starting point is always the brief.
Frequently asked questions
Does Netflix use AI to create actors?
Not in any publicly known case. Netflix uses generative AI for visual effects and post-production fixes, and SAG-AFTRA contracts require notice and bargaining whenever a synthetic performer is used.
What is InterPositive and why did Netflix buy it?
It is the AI company founded by Ben Affleck and acquired by Netflix in March 2026. It builds a model from a single production’s dailies, so lighting, backgrounds and framing can be fixed without going back to set.
Does generative AI really cut production costs?
In some stages, yes: Eyeline estimates the El Eternauta collapse would have cost about ten times as much with traditional VFX. The savings apply to specific steps, mostly in post-production, not to direction or performances.
Sources
- AI Is Still a Dirty Word in Hollywood. That’s the Problem · TheWrap, 2025-08-25
- What Happened to Lionsgate’s Splashy Plan to Make AI Movies With Runway? It’s Complicated · TheWrap, 2025-09-22
- SAG-AFTRA Refutes the Idea of 'AI Actors' Like Tilly Norwood, Reminds Producers of Contractual Obligations · TheWrap, 2025-09-30
- Inside Netflix’s Bid to Supercharge Faster, More Efficient VFX Through Virtual Production and AI · TheWrap, 2025-10-15
- Netflix Slams ByteDance’s Seedance 2.0 as a 'High-Speed Piracy Engine,' Threatens 'Immediate Litigation' · TheWrap, 2026-02-17
- Netflix Acquires Ben Affleck’s AI Production Startup · TheWrap, 2026-03-05
- Can Ben Affleck Lead Hollywood’s Embrace of AI? | Analysis · TheWrap, 2026-03-09
- Netflix to Pay Up to $600 Million for Ben Affleck’s AI Company · TheWrap (citing Bloomberg), 2026-03-11
- AI Actors Are Coming, Even if Audiences Aren’t Buying It (NRG study) · TheWrap, 2026-03-16
- Inside Doug Liman’s $70 Million AI-Made Movie Starring Casey Affleck and Gal Gadot · TheWrap, 2026-04-15
- Q3 FY26 Earnings Conference Call transcript · The Walt Disney Company, 2026-08-05
- Adobe and The Walt Disney Company Announce Multi-Year Strategic Alliance to Support Disney Innovation with New AI-Powered Tools · Adobe, 2026-11-09
Keep reading.
AI Christmas ads in 2026: who uses AI, who refuses it and why
Prada draws by hand, Levi’s goes back to claymation, Dior faces AI accusations: Christmas 2026 shows what audiences punish and accept from AI.
Disney OpenAI Sora deal: why it collapsed and what it teaches
A billion dollars and 200 characters in Sora: the Disney-OpenAI deal collapsed before it began. What happened, and what it means for IP in generative tools.
AI slop: what it is, why brands should care, how to avoid it
What AI slop is, why dictionaries and linguists made it the word of 2025, and how to tell a directed AI video from content churned out in bulk.