Doug Liman's AI film 'Bitcoin' cut post-production from 40 years to six months
At Google I/O 2026, Doug Liman and 30 Ninjas detailed how AI compressed a feature film's post-production from 40 years to six months using Veo and Gemini Live.
AI
Doug Liman stood on the Dialogues stage at Google I/O 2026 on May 22, 2026, and described a production timeline that would have been unthinkable two years ago. The director behind "The Bourne Identity," "Mr. and Mrs. Smith," and "Edge of Tomorrow" has completed what he calls the first fully AI studio feature film - a cryptocurrency thriller starring Casey Affleck, Gal Gadot, and Pete Davidson. When the project began, his honest estimate for post-production was 40 years. The finished film is now six months from delivery.
The session, titled "Directing the future: craft and creativity in the age of AI," was published by Google for Developers on May 22, 2026, as part of the Dialogues at Google I/O 2026 series. Liman was joined by Jed Weintrob and Julina Tatlock, co-founders of 30 Ninjas - the AI filmmaking studio behind the production - and by Mira Lane from Google, who moderated the conversation.
From 40 years to six months
The 40-year figure was not rhetorical. According to Liman, when he and producer Ryan Kavanaugh committed to the project roughly two and a half years ago, the state of generative AI meant that a film produced entirely through AI-rendered environments would require an implausible post-production span. "When we started this, my estimation, with no exaggeration, was that post-production was going to be 40 years, 4-0 years," Liman said. The plan was to keep moving and count on technological acceleration to close the gap.
It worked - faster than expected. As recently as two months before the Google I/O session, the post-production estimate still stood at three years. That dropped to six months, which, according to Liman, falls within a standard commercial film schedule. The compression came from advances in Google's models, specifically Gemini Omni and related tools, which 30 Ninjas integrated into their pipeline while the film was already in progress.
The production architecture of 'Bitcoin'
"Bitcoin" tells the story of Craig Wright, played by Affleck, who may or may not have created Bitcoin, and the forces that assemble to destroy him. Davidson plays a billionaire magnate making what Liman describes as the largest gamble in history. Gadot plays a reporter attempting to separate fact from fiction. The film is set against the world of cryptocurrency - a subject Liman described as already framing a tension between human judgment and machine verification.
Every scene was captured on a single stage. No matter whether the script placed characters inside a swimming pool, on a boat, on a beach, or in Las Vegas, all footage was shot in one location. That single set then served as the source material for AI-generated environments. The practical implication was a complete dismantling of the usual trade-off between a director's creative ambition and a producer's budget constraints. Liman described a standard film negotiation: request a scene in Las Vegas, get offered a conference room in a London hotel basement, compromise. "Actually, I don't have to choose," he said. "I can do Vegas and I can go to Antigua."
He ended up using 150 cities around the world across the film's runtime. A traditional feature at comparable scale might use 40 locations. "Bitcoin" deployed 125 to 140 actors - Liman's line producer pushed to cut that to 70, pointing out that "Bourne Identity" used 50. He refused. The expanded location count required more characters to populate those environments, and those were human performances, not generated ones. "I wasn't going to do any computer-generated performances," Liman said.
Capture, animatics, and the iterative pipeline
The technical sequence of "Bitcoin" began with six months of full-prep research and development, followed by four and a half weeks of principal photography on the single stage. Post-production then runs through the AI rendering pipeline before delivering a final cut.
On set, Tatlock served as AI supervisor. Simultaneously, a team of AI artists worked within the editing room alongside the art department and the production designer, generating environmental stills and frames. Those images fed into an animatic - a working rough cut - so Liman could evaluate how well each scene's AI rendering was functioning before committing to a final approach. Some scenes did not work. Those were rewritten and reshot. That iterative loop is unusual in traditional film production, where the line between pre-production, production, and post tends to be firm.
The stage also ran 360-degree capture to record actor likenesses in full, which gave the post-production team the spatial data needed to place performances into AI-rendered environments with consistent lighting and perspective. Liman drew a direct lesson from his earlier "ASTEROID" project - a 180-degree stereoscopic XR film made for the Samsung Galaxy XR headset - where all-performance-capture footage left the editing room holding only data. "We have to generate shots from that data first, even being able to figure out how to start to edit," he said. That experience led to a key structural decision for "Bitcoin": composite shots needed to be composed on set, not deferred entirely to post. Witness cameras provided the initial reference. In post, angles could be changed, but something concrete had to exist to edit from.
Weintrob described the studio's workflow philosophy as agile by design - borrowed from software development rather than from traditional film production. VFX studios operate with written-in-stone pipelines. 30 Ninjas does not. "We're not doing it right if we're not changing that every couple of weeks," he said. The pipeline adapts to whatever model capability arrives. That approach also meant working directly with Google's research teams and with DeepMind to identify where the technology needed to go, rather than simply using what was already publicly available.
'ASTEROID' and the agentic character extension
Alongside "Bitcoin," the session covered a second 30 Ninjas project that demonstrates a different application of AI in storytelling: interactive character extension. "ASTEROID" was directed by Liman for the Samsung Galaxy XR headset, shot in 180-degree stereoscopic format, and stars Hailee Steinfeld and Ron Perlman. The story involves a crew on an asteroid-mining expedition that ends badly for most of the characters.
One of those characters is DK Metcalf, the NFL wide receiver, playing a version of himself who gets stranded and left for dead. After the linear film ends, the viewer - assuming the role of a NASA operator on Earth - can attempt to rescue Metcalf by talking to him in real time. The rescue interaction is powered by Gemini Live native audio, which enables natural-language conversation rather than the pre-scripted branching dialogue used in earlier interactive media. According to Tatlock, Gemini Live native audio arrived approximately four months before the project's premiere, and it changed the character interaction model fundamentally. "You move past the NPC and you can still stay within the story world," she said.
The technical stack for the interactive layer is a Unity application. The viewer is positioned as a ground-based NASA contact, not inside the asteroid environment, which means the visual representation is photorealistic but abstracted. The DK character that viewers interact with is not simply the public figure: he is the specific character that Liman wrote, trained on the story world he inhabits. He knows he is stranded on an asteroid. He knows the stakes. He will talk about football because he plays himself, but he operates within the fictional context of the film.
Building that character required, according to Tatlock and Weintrob, roughly 1,000 pages of writing for a short interactive experience. The model needed training on every situation the character might encounter, not just a list of possible dialogue lines. Writers had to define what the character would and would not say - including guardrails against attempts to make him trash-talk other NFL players. Those guardrails were carefully designed to redirect rather than break. When pushed to disparage other players, the DK character responds by noting that everyone who reaches the NFL is a super talented athlete. Metcalf reviewed the character and approved it.
The job question
The Dialogues session repeatedly addressed the question of whether AI filmmaking displaces workers. Liman's answer was grounded in his own company's hiring. Because 30 Ninjas is a startup, it is creating jobs rather than replacing them - including jobs for film professionals who are not working in traditional Hollywood. "The business is broken," Liman said, describing a production landscape where a quarter of the number of large-budget films that used to be made are now being made. 30 Ninjas has hired top film editors currently unemployed because of that contraction, and screenwriters who are generating shots for "Bitcoin." "They're bringing all of that expertise as storytellers in the industry," he said.
Weintrob was careful to distinguish between what 30 Ninjas does and what sometimes gets associated with AI production. "We're creating the jobs. We're not generating jobs in AI," he said. The script is a script written by a writer. The screenwriters working on "Bitcoin" are doing new kinds of work - writing for prompts, generating visual sequences - but they are writers, not algorithm outputs.
The session connected to a broader context at Google I/O 2026, where the company's AI tooling extends well beyond filmmaking. The search changes announced on May 19 included persistent background agents, a rebuilt search box, and AI Mode upgrades, while Veo has been moving progressively into consumer and advertising tools throughout 2025 and 2026. Google's Flow filmmaking platform launched in May 2025 and added speech generation in July 2025, running on the same Veo 3 model that underpins the rendering work at 30 Ninjas.
What this means for the marketing community
The "Bitcoin" pipeline offers a concrete illustration of what AI-assisted production looks like at feature-film scale when it is driven by experienced creative professionals rather than automated systems. For the marketing and advertising industry, where Veo has been integrated into Google's Asset Studio since September 2025 and is now available to all personal account holders through Google Vids, the Liman-30 Ninjas approach to production design carries practical lessons.
The 150-location production was not a cost-reduction exercise. It was a creative expansion made possible by the removal of constraints. In traditional advertising production, location, cast size, and set complexity are the primary cost drivers. AI-rendered environments do not eliminate those decisions, but they do change their economics. The iterative animatic process - generating environments, reviewing them in the edit, and reshooting what does not work - is structurally similar to the iterative creative testing that Google introduced at Marketing Live 2026 through 1-Click Creative Testing in Asset Studio.
Liman also noted that the removal of location constraints changed how the script was written. "Those limitations are no longer there," he said. "You can imagine a playwright has to write with this idea of, how do I have everything take place in this living room, because all I have is that set? And filmmakers, you wouldn't think they have limitations, but we really do. And how do we fit it into that box? And I'm like, well, actually, there is no box anymore." That structural shift - from constrained to unconstrained creative scope - is one that brands and agencies operating with AI video tools are beginning to encounter at smaller scales.
The "ASTEROID" interactive character, powered by Gemini Live, points toward a separate development in the marketing space. Agentic characters trained on specific brand worlds, capable of natural-language conversation without breaking character, are a near-term application of the same technology. The 1,000 pages of writing required to build DK Metcalf's interactive character represent the content engineering cost behind a convincing AI persona - a cost that does not disappear simply because the interaction layer is generative.
A 10-minute preview of "Bitcoin" was shown at the Cannes Film Festival last week, according to Liman, who said the response was strong. The film is not yet in distribution.
Timeline
- 2015 - 30 Ninjas produces "Invisible," the first-ever long-form virtual reality series, in partnership with Google Cardboard and Samsung, earning a Guinness World Record.
- ~Early 2024 - Doug Liman and producer Ryan Kavanaugh begin the "Bitcoin" project after a Google dinner introduces Liman to AI filmmaking possibilities; initial post-production estimate is 40 years.
- May 2025 - Google launches the Flow AI filmmaking platform at Google I/O 2025, built on Veo 2 and Veo 3. PPC Land coverage
- July 10, 2025 - Google expands Flow with speech generation capabilities powered by Veo 3, extending availability to 76 additional countries. PPC Land coverage
- July 31, 2025 - Google launches Veo 3 Fast for developers via the Gemini API at $0.40 per second. PPC Land coverage
- September 10, 2025 - Google expands Asset Studio with Imagen 4 and Veo integration inside the Google Ads interface. PPC Land coverage
- ~Late 2024 to early 2025 - Six months of prep R&D for "Bitcoin" production design, including single-stage workflow development.
- ~Early 2025 - Four and a half weeks of principal photography on "Bitcoin" completed on one stage.
- ~Early 2026 - Post-production estimate for "Bitcoin" reduced from 40 years to three years.
- April 2, 2026 - Google Vids receives free Veo 3.1 generation for all personal accounts. PPC Land coverage
- ~Three months before Google I/O 2026 - Gemini Live native audio integrates into "ASTEROID" interactive character extension, approximately four months before its premiere.
- ~Three months before Google I/O 2026 - Post-production estimate for "Bitcoin" reduced further to three years, then to six months.
- May 19, 2026 - Google I/O 2026 keynote; Google announces sweeping Search changes, Gemini 3.5 Flash in AI Mode, and background agents. PPC Land coverage
- Week of May 19, 2026 - Doug Liman shows a 10-minute preview of "Bitcoin" at the Cannes Film Festival.
- May 22, 2026 - "Directing the future: craft and creativity in the age of AI" session published on YouTube as part of the Dialogues at Google I/O 2026 series.
Summary
Who: Director Doug Liman, 30 Ninjas co-founders Jed Weintrob and Julina Tatlock, and Google's Mira Lane.
What: A detailed account of the production of "Bitcoin," described as the first fully AI studio feature film, starring Casey Affleck, Gal Gadot, and Pete Davidson. The session also covered "ASTEROID," an XR film with an interactive agentic character extension powered by Gemini Live. Post-production for "Bitcoin" was compressed from an initial estimate of 40 years to six months through the use of Gemini Omni and Veo-based rendering tools.
When: The session was published on May 22, 2026, as part of the Dialogues at Google I/O 2026 series. The underlying "Bitcoin" production began approximately two and a half years prior. A 10-minute preview screened at Cannes the week of May 19, 2026.
Where: The Dialogues stage at Google I/O 2026. "Bitcoin" was shot on a single stage, with AI-rendered environments representing 150 cities worldwide. "ASTEROID" was produced for the Samsung Galaxy XR headset.
Why: The session illustrated how a Hollywood director is applying generative AI - specifically Google's Veo video model and Gemini Live - to feature-length narrative filmmaking, removing traditional constraints on location, scale, and post-production timelines, while retaining human creative direction, live-action performance capture, and professional writing staff.