AI In VFX Industry | AI Tools For Filmmaking | Future Of VFX With AI

Everyone Said AI Couldn’t Replace VFX. Then This Happened

A look at the industry before and after AI.

Image Edited by Author

Read here for FREE

The seven-person VFX team behind Everything Everywhere All at Once used Runway AI to rotoscope a scene that, on a traditional production, would have taken a far larger crew weeks to finish by hand. The film had a $25 million budget. It won Best Picture. And almost no one watching noticed where the AI stepped in.

That is a pretty good summary of where things stand right now.

For Anyone New to This

Visual effects, or VFX, is the work that happens after filming. It is everything you see on screen that was not physically there on set.

When you watch a character fall from a skyscraper, a spaceship drift through an asteroid field, or even a crowd scene with thousands of people, you are watching VFX. Most of the time, the work is invisible. That is the point.

Before AI entered the picture, creating that invisibility was intensely manual. A film like House of the Dragon could list over 1,000 VFX workers in its credits. Thousands of people doing frame-by-frame work, tracing edges, fixing lighting, rendering environments, removing rigs and cables from shots.

One of the most common tasks was rotoscoping. It sounds technical but the concept is simple: you trace around a person or object in every single frame of footage so the background can be replaced or altered. A two-minute scene can be thousands of individual frames. It took time, it took people, and it cost money. A lot of it.

Then the Tools Started Arriving

Around 2022 and 2023, AI tools started showing up that could handle parts of this work automatically.

Adobe’s Roto Brush in After Effects began generating base masks around subjects without manual frame-by-frame tracing. Runway ML arrived with tools that let editors remove objects, clean up backgrounds, and generate new frames. Mocha Pro added AI-assisted masking. None of these replaced artists overnight, but they changed how long certain tasks took.

The tools available right now, as of mid-2026, include:

• Runway Gen-4 — currently one of the strongest all-around AI video tools for professional use, with camera control and character consistency
• Kling 2.6 — developed by Kuaishou, generates up to 3-minute videos, strong cost efficiency
• Google Veo 3 — enterprise-grade, strong photorealism, long-form clips
• Pika Labs 2.0 — fast, built for short social clips
• Adobe Roto Brush 3 and Mocha Pro (Mask ML) — used directly inside editing pipelines

The Things AI Has Actually Done On Screen

This is not theoretical. Some of it is already in films you have seen.

Eminem’s Houdini music video Metaphysic’s AI de-aged Eminem to his late-90s Slim Shady look. The entire effect was completed in under three weeks. It won Best Visual Effects at the VMAs.

Robert Zemeckis’s Here Director Zemeckis used Metaphysic’s AI tool, Metaphysic Live, to de-age Tom Hanks and Robin Wright — in real time, on camera while shooting. Previously, this kind of work required weeks in post-production. The same technology was used on Furiosa and Alien: Romulus.

Val Kilmer’s posthumous role An upcoming film called As Deep as the Grave will feature an AI-generated likeness of Val Kilmer, who died in 2025. It is the most significant case yet of AI being used to place a deceased performer on screen.

These are not small indie experiments. They are mainstream productions from major studios using AI in ways that were not practically possible three or four years ago.

What Is Actually Changing for Workers

Here is the part that is harder to talk around.

In February 2025, Technicolor, one of the largest VFX companies in the world and a major vendor for Disney, Paramount, and Netflix, shut down its India operations. About 3,000 workers in Bengaluru and Mumbai were let go with no pay, no notice, and no severance.

Los Angeles County has lost 41,000 film and TV jobs in three years. That is roughly a quarter of the county’s entire entertainment workforce.

Computer graphic artist roles dropped 33% in 2025. VFX-adjacent writing and photography roles fell by similar margins.

Netflix acquired Ben Affleck’s AI production company InterPositive and opened a new facility called Eyeline Studios in Hyderabad shortly after. The message was not subtle. Disney had committed $1 billion to OpenAI and licensed over 200 characters from Marvel, Pixar, and Star Wars before the deal collapsed when Sora shut down.

Experts quoted in industry reporting have said that once AI can automate enough tasks reliably, VFX crews on major productions could be cut by 80% or more. That projection has not happened yet. But the direction is clear.

The jobs most at risk are the ones that were always the hardest to talk about: entry-level, repetitive, technical. The work that juniors did to build experience. The pipeline that let people grow into senior roles.

But Here Is What the Hype Leaves Out

AI in VFX is real. The job losses are real. And so are the limitations.

One VFX company owner, quoted in a recent California labor report, described using AI on a major TV production this way: they had 15 artists sitting at workstations fixing AI output. When you multiply the hourly rate of those 15 artists by the hours spent correcting the AI, the savings disappeared entirely.

AI currently handles parts of the pipeline well:

• Rotoscoping and masking
• Wire and rig removal
• Background generation and extension
• De-aging and face replacement
• Rendering denoising (making renders faster and cleaner)

AI still struggles with:

• Consistent character faces across multiple shots
• Complex physics and fluid dynamics at production quality
• Making creative decisions about what a scene should feel like
• Adapting when a director changes their mind mid-project

VFX director Jay Harwood described it plainly: AI speeds up the first 80 percent of the work, but artists still do the final 20 percent that actually matters for quality.

That last 20 percent is also the part that takes the most judgment, experience, and creative sense. It is the part studios do not want to get wrong.

The Honest Picture

The VFX industry is not being replaced. It is being restructured, and the restructuring is uneven.

The artists who spent years doing rotoscoping, tracking, and wire removal are finding that work automated away. The senior artists making creative decisions about how a digital character should move or how an explosion should feel are still employed.

What seems to be emerging is a two-tier situation:

• A smaller, more senior workforce doing higher-level creative and supervisory work
• AI handling the volume work underneath them

Whether that is a fair trade depends on where you sit. For studios, the savings are real. For the junior artist who planned to spend the first three years of their career doing rotoscoping and working up, the entry point is shrinking.

The industry was already under pressure before AI. The WGA and SAG strikes highlighted how unsustainable Hollywood’s cost model had become. VFX companies had long been squeezed by studios demanding work at prices that left no margin. AI is accelerating a restructuring that was already happening, not creating it from scratch.

Where Things Are Heading

Right now, in mid-2026, no AI tool produces final-quality VFX on its own. Everything still goes through human artists, supervisors, and directors. But the number of human hours required for a given project keeps falling.

Runway has partnered with Lionsgate. DNEG, the house behind Dune, acquired Metaphysic. The Academy of Motion Picture Arts and Sciences adjusted Oscar eligibility rules, blocking films with AI-generated actor performances from competing starting in 2027. Unions are negotiating AI safeguards into contracts.

The tools are getting better every few months. The resolution limits that were a genuine barrier in early 2025 are already less of an issue. Character consistency, one of the biggest unsolved problems, is improving.

There is also a deeper architectural shift happening underneath all of this. Meta’s JEPA (Joint Embedding Predictive Architecture) works differently from tools like Runway or Kling — instead of predicting pixels, it tries to model how the world actually behaves. V-JEPA 2, trained on over a million hours of video, already shows stronger understanding of motion and physical cause-and-effect. If that line of research matures, the physics and consistency problems that still make AI footage feel slightly off could get solved faster than expected.

The one thing that is probably true across all of this:

The industry will not look the same in five years. What it will look like instead, nobody has figured out yet.

The VFX industry employs more than two million people globally. Most of the debate around AI happens at the level of major Hollywood productions, but the effects reach every VFX house, every freelance artist, and every student currently deciding whether to build a career in this field.