AI VFX Studios: 8 Verified Users in Production (2026)

AI-Aided VFX: Eight studios have verified, named-production track records using AI in visual effects work today: Metaphysic (real-time de-aging on Here), DNEG through its Brahma AI division, Digital Domain (the Charlatan facial tool), Industrial Light & Magic, Netflix’s Eyeline Studios (the first disclosed generative-AI VFX shot on Netflix, in El Eternauta), MPC (AI-driven crowd simulation), Weta FX (the machine-learning facial pipeline behind Avatar: The Way of Water), and MARZ (the Vanity AI de-aging tool used on 27+ productions). Two AI tool vendors, Wonder Dynamics and Runway, are also directly relevant to a vendor evaluation, though neither is a production studio itself.

Most coverage of this topic doesn’t get that specific. A search for “AI VFX studios” surfaces dozens of listicles making the same broad claim about an undifferentiated trend, most without naming a single company, tool, or verifiable production credit, exactly the gap in the earlier version of this article that this rewrite corrects. That’s a costly gap for a producer or studio executive acting on a vague claim about “AI-powered VFX” without knowing which specific technique, de-aging, generative rendering, or crowd simulation, a given vendor actually offers.

How Is AI Actually Changing VFX Production?

Conflating these three categories is where most coverage of this topic goes wrong, and it’s exactly what the previous version of this article did: describing “AI in VFX” in the abstract without naming a single studio, tool, or project. Treating “AI VFX” as one undifferentiated trend also makes it harder to evaluate a vendor, since a studio strong in real-time facial de-aging isn’t necessarily equipped for generative full-shot work, and vice versa.

What’s driving adoption isn’t novelty for its own sake. It’s cost and schedule. Netflix has said publicly that certain 2026 sequences using generative AI VFX “just wouldn’t have been feasible” at the show’s budget using conventional methods. That’s the actual business case studio executives and producers should be evaluating against, not a vague sense that AI makes effects “better.” The same economic pressure is visible across the broader VFX and CG production pipeline for high-volume content, where studios face growing episode counts on flat or shrinking per-episode budgets.

Precision matters here because “AI” covers genuinely different technology across these three categories. De-aging and digital-human work typically uses generative adversarial networks or diffusion-based facial synthesis trained on a specific performer’s likeness. Generative full-shot rendering, as in Netflix’s Eyeline case below, uses broader image and video generation models to construct an entire environment or effect from a text or reference-image prompt. Crowd and behavior simulation, as in MPC’s decades-old ALICE system, is a different, older category, rule-based agent AI, not the deep-learning approaches usually implied when the term “AI VFX” gets used today. A production team that doesn’t distinguish these when evaluating a vendor risks assuming a capability the vendor doesn’t actually have.

Which Studios Lead AI-Powered De-Aging and Digital Human Work?

Metaphysic’s Real-Time De-Aging on Here

Metaphysic, a London-founded company established in 2021, built “Metaphysic Live,” a real-time generative AI system that de-ages or ages a performer’s face on a monitor during filming, rather than in post-production months later. It was used on Here (2024, directed by Robert Zemeckis) to shift Tom Hanks and Robin Wright’s ages across decades within the same continuous take (Hollywood Reporter, 2023).

Metaphysic’s approach differs from older de-aging methods in one specific way: it renders the altered face in real time, viewable on a monitor during the actual take, rather than requiring the director and performers to imagine the final result and wait months for a VFX team to deliver it in post-production. For productions weighing whether to build this into their content development and production planning from the outset, that changes the on-set workflow meaningfully, since a director can adjust a performance based on what the aged or de-aged version actually looks like, not a guess.

DNEG’s Brahma Division and the Metaphysic Acquisition

DNEG, the London-headquartered VFX giant behind work on Dune: Part Two and Godzilla x Kong: The New Empire, uses Ziva VFX and Ziva RealTime, machine-learning-powered muscle, fat, and skin simulation tools that won a 2025 Academy Sci-Tech Award. In February 2025, DNEG’s dedicated AI division, Brahma, acquired Metaphysic itself for a reported $1.43 billion post-money valuation (Deadline, 2025), consolidating one of the largest traditional VFX houses with one of the highest-profile AI-native de-aging companies under one roof.

The acquisition is a useful signal for anyone tracking where deal activity in AI-driven media companies is actually concentrated: that reported post-money valuation for a four-year-old company (Deadline, 2025) shows established VFX houses are willing to pay a premium for AI-native technology rather than building equivalent capability in-house from scratch, a build-versus-buy calculation that’s becoming a standard part of studio M&A strategy across the sector.

Digital Domain’s Charlatan Facial Tool

Digital Domain, based in Playa Vista, California, built Charlatan, a proprietary neural-network facial capture and face-swap tool that ages or de-ages a performer without requiring a 3D scan. Charlatan aged David Beckham for the “Malaria Must Die” campaign in 2020, blending current footage with an older stand-in performer, and was separately used to create a virtual recreation of singer Teresa Teng (Befores & Afters, 2020).

ILM’s Deepfake-Assisted Young Luke Skywalker

ILM combined deepfake-informed machine learning with traditional CG and a performer double to create young Luke Skywalker across roughly 80 shots in The Book of Boba Fett Episode 6 (2022). The studio refined the pipeline after hiring deepfake artist “Shamook,” who had gone viral for an unofficial fan-made fix to ILM’s earlier Luke Skywalker work, as a senior facial capture artist (Hollywood Reporter).

ILM’s decision to hire deepfake artist “Shamook,” after his unofficial fan-made fix to the studio’s earlier young-Luke-Skywalker de-aging work went viral, and to fold elements of his technique into the Book of Boba Fett pipeline, is an unusually direct admission from a major studio. An independent fan working alone with consumer-grade deepfake tools had produced a result ILM’s own team judged good enough to hire him and adopt his approach. That’s a different signal than simply licensing a vendor’s proprietary tool: it shows how far accessible, non-proprietary machine learning had already closed the gap with big-budget in-house VFX pipelines by 2022, well before generative AI became a mainstream production topic. It also previews a pattern across nearly every studio in this article. The technology usually arrives first through a smaller, faster-moving team, then gets absorbed into larger studios’ pipelines afterward through a hire, an acquisition, or a licensing deal, and rarely gets invented from scratch inside the incumbents.

Which Studios Use AI for Full-Shot Generation and Crowd Simulation?

Eyeline Studios, Netflix’s internal VFX unit formed by merging with Scanline VFX in October 2025, used generative AI to render a building-collapse sequence in Buenos Aires for El Eternauta (2025), the first disclosed instance of a fully AI-generated VFX shot in Netflix final broadcast content. Netflix has said the sequence was roughly 10 times faster and cheaper than a traditional VFX approach would have allowed (Variety, 2025).

MPC’s crowd and behavior-simulation system, ALICE, is agent-based AI, not modern generative machine learning, a distinction that gets conflated often enough to flag directly. It was used for the zombie swarm sequences in World War Z (2013) and ragdoll physics blended with motion capture in The Lone Ranger (MPC Film). It’s an older category of “AI” than the generative tools above, but it remains in active production use.

Weta FX, the Wellington, New Zealand studio behind Avatar: The Way of Water (2022), built a neural-network-based facial animation pipeline using muscle-fibre curves and machine-learning autoencoders to keep character expressions “on-model” across a production with thousands of complex shots, alongside a separate real-time machine-learning depth-compositing system for blending on-set footage with CG (fxguide).

The scale problem Weta FX was solving is specific and instructive for any producer weighing an AI-assisted approach: a feature with thousands of character shots across a multi-year shoot risks drift, small inconsistencies in how a digital character’s face reads from shot to shot, that a purely manual pipeline struggles to catch at scale. The autoencoder-based system was built specifically to hold every shot to the same underlying facial model, a quality-control problem as much as a creative one. It’s the kind of infrastructure investment that only pays off across a production of Avatar’s scope, which is also why smaller production and post-production vendors further down the entertainment supply chain have generally adopted lighter-weight, licensed tools rather than building equivalent systems in-house.

Which Studios Have Built AI Tools for Rotoscoping and VFX Cleanup?

MARZ (Monsters Aliens Robots Zombies) positions Vanity AI as an end-to-end AI de-aging and touch-up application rather than a single-effect tool, cutting shots that used to take days down to minutes in the company’s own account (BusinessWire, 2023). Separately, the studio’s creature and puppet AI work appeared on the “Thing” character in Wednesday (CBC News).

The Framestore and Untold Studios cases illustrate a pattern worth watching, not dismissing: a studio building or testing AI tooling internally, ahead of a public, named production credit, is a leading indicator of where a capability is headed, even if it can’t yet be verified against a finished release. For a producer or production team evaluating vendor software and tooling more broadly, the honest way to treat this kind of claim is as “in development, unconfirmed in finished work,” not as equivalent to Metaphysic’s or MARZ’s verified production history, and not as fabricated or dismissed either.

Are AI VFX Tool Vendors Like Wonder Dynamics and Runway Worth Watching?

This distinction matters for a B2B buyer. A studio is a vendor you commission finished shots from. A tool vendor sells software that a studio’s own artists, or an independent VFX team, uses inside their own pipeline. Wonder Dynamics, a Los Angeles company founded in 2017, built Wonder Studio, a cloud-based tool that automates animating, lighting, and compositing 3D characters into live-action footage. Autodesk acquired it in May 2024 (Autodesk investor relations, 2024), folding it directly into one of the industry’s dominant production-software providers.

Runway, a New York AI video company founded in 2018, doesn’t produce VFX itself, but its generative rotoscoping and background-removal tools were used directly by VFX editor Evan Halleck on the “rock universe” sequences in Everything Everywhere All at Once (2022), completed by an 8-person team instead of the dozens a conventional approach would have required (Variety, 2023).

That 8-person team detail is the actual headline, not the visual effects themselves. It’s a small independent production achieving a result that would previously have required commissioning a mid-size studio, the kind of shift that matters most to producers working outside the tentpole-budget tier who are sourcing from regional studios in markets like Asia or the Middle East, where tool-assisted small teams can now compete on specific sequence types against larger, traditionally-staffed studios.

How Should a Production Team Evaluate an AI-Aided VFX Vendor?

The distinction between owning a proprietary tool, as DNEG, Digital Domain, MARZ, and Metaphysic do, and licensing or testing a third-party tool, as Untold Studios did with Slapshot, has real budget and IP implications. A studio with an owned, proprietary pipeline typically has tighter control over output consistency across a long production, while a studio integrating a third-party tool may move faster on adoption but carries less negotiating leverage on pricing or customization.

It’s also worth asking directly whether a vendor’s claimed AI work has ever shipped in a released, named production, given how much of the coverage in this space describes tools and pilots without a finished-work credit attached, exactly the gap this article flagged for Framestore’s Futon and Untold Studios’ Slapshot above. A vendor unwilling or unable to name a specific released project is a meaningfully different proposition than one that can point to Here, Avatar: The Way of Water, or El Eternauta.

This same diligence applies whether the vendor under consideration is one of the global names in this article or a regional house sourced through a directory of production houses worldwide. Company size and headline reputation aren’t a substitute for a specific, checkable production credit tied to the specific AI technique a project actually needs.

A practical due-diligence sequence looks like this: first, ask the vendor to name the specific released production where their claimed technique shipped, not a pilot or a demo reel. Second, confirm whether the tool is proprietary and owned by the vendor, or licensed from a third party, since that affects both pricing leverage and long-term support risk if the underlying vendor is later acquired, as happened with Metaphysic and Wonder Dynamics. Third, ask directly how the proposed workflow affects headcount and the performer consent process, since both carry contractual and labor implications that are easier to negotiate before a deal closes than after. None of these three questions require specialized AI expertise to ask, they just require actually asking them instead of accepting a vendor’s marketing description at face value.

What Are the Real Risks of AI-Generated VFX?

These risks sit alongside, not instead of, the genuine benefits documented throughout this article. Naming them isn’t an argument against adopting AI-assisted VFX, it’s the missing half of a topic that’s mostly been covered in purely promotional terms. A producer who understands both the documented capability and the documented risk is in a materially stronger negotiating position than one working from marketing copy alone.

Metaphysic and Digital Domain’s tools both operate directly on a performer’s face and voice. SAG-AFTRA’s 2023 negotiations with major studios specifically addressed digital replica and AI consent terms, following disputes over how, and under what compensation, an actor’s digitally altered likeness could be used. Any production team evaluating a de-aging or face-swap vendor should treat performer consent and compensation terms as a contractual requirement, not an afterthought handled after the shots are already cut.

The second risk is simpler but easy to underweight: a single high-profile case study, like Netflix’s one disclosed generative-AI shot in El Eternauta, doesn’t establish that a technique is production-ready across an entire slate. Several of the vendors named in this article, notably the tool vendors in the previous section, have one strong documented use case and a thinner track record beyond it. Treating one successful pilot as proof of general readiness is the same overclaiming problem this rewrite was written to correct.

A third, less-discussed risk is workflow fragility. A pipeline built around a proprietary AI tool, whether Metaphysic Live, Charlatan, or Ziva RealTime, creates a dependency on that specific vendor’s continued operation and support. When DNEG’s Brahma division acquired Metaphysic, it consolidated a previously independent AI vendor into a larger company’s internal stack, which is a reasonable outcome for DNEG but a relevant data point for any other studio that had been relying on Metaphysic as a standalone, independently-negotiated vendor. Vendor consolidation in a fast-moving technology category is a normal part of the market maturing, but plan around it as a genuine operational risk, not just a footnote.

Union agreements are a fourth layer, separate from the consent question above. Beyond SAG-AFTRA’s negotiated terms on digital replicas, the VFX workforce itself, artists doing rotoscoping, cleanup, and compositing work, has raised concerns about AI tools displacing entry-level and mid-level positions that traditionally served as the industry’s training pipeline. MARZ’s claim that Vanity AI runs roughly 300 times faster than manual work is a genuine productivity gain from the studio’s perspective, but it also represents fewer billable hours of the kind that used to be distributed across a larger team. Producers negotiating with a VFX vendor should ask directly how a proposed AI-assisted workflow affects headcount and staffing on their specific project, rather than treating labor impact as someone else’s problem.

There’s a fifth consideration that gets less attention than it deserves: verification of the finished work itself. As de-aging and generative techniques improve, distinguishing a genuinely AI-assisted shot from marketing language describing conventional CG work as “AI-powered” becomes harder for a non-specialist to do by eye. That’s part of why this rewrite leaned so heavily on named, dated, third-party-reported sources for every claim rather than studio marketing copy alone, and it’s a reasonable standard to hold any vendor’s own pitch materials to as well. A studio or vendor that can point to independent trade press coverage of a specific technique on a specific release, the way Metaphysic, DNEG, Digital Domain, ILM, Weta FX, MARZ, and Eyeline Studios all can, has cleared a meaningfully higher bar than one whose evidence is limited to its own website.

What Does AI VFX Actually Change About Production Budgets and Timelines?

Neither of those figures is an independently audited benchmark. State that plainly instead of repeating the numbers as settled fact. They’re the vendors’ and platform’s own public statements (Variety, 2025; BusinessWire, 2023), each drawn from a single disclosed case study rather than an average across a measured sample of productions. That doesn’t make the figures false, but a producer building a budget should treat either claim as a starting estimate to validate against their own project’s specific shot complexity, not a fixed multiplier that applies automatically.

The realistic budget conversation is usually narrower than the headline figures suggest. AI-assisted de-aging and cleanup tools most reliably reduce cost on high-volume, repetitive shot categories, extended dialogue scenes needing consistent facial correction, large numbers of similar rotoscoping tasks, background cleanup across many frames, rather than on a production’s hero visual-effects sequences, which still typically require bespoke artist work regardless of which AI tools are layered underneath. A producer expecting AI tools to meaningfully cut costs on a film’s signature effects shots is likely to be disappointed; a producer expecting cost reduction on the unglamorous bulk of a shot list, which is most of any production’s actual VFX budget line, has a much better-supported case.

Timeline compression follows a similar pattern. Real-time tools like Metaphysic Live remove an entire round-trip between production and post-production for specific shots, which shortens overall schedule more than it reduces line-item cost. That distinction matters when a studio is negotiating delivery dates rather than just a total budget number, since a faster iteration loop on set can be worth more to a tight shooting schedule than an equivalent dollar saving realized months later in post.

Put the pieces of this article together and a clear picture emerges: AI VFX in 2026 is not one trend but eight distinct, verifiable capabilities spread across eight different companies, each solving a different, specific production problem. Metaphysic and Digital Domain solve the de-aging problem. Eyeline Studios and MPC solve full-shot generation and crowd scale. Weta FX solves cross-production facial consistency. MARZ solves high-volume cleanup speed. Framestore and Untold Studios are still proving out their approaches in public. None of the eight is a general-purpose “AI VFX solution.” Any vendor pitch that frames itself that way needs a second, more specific round of questions before signing anything.