In 2018, Amazon Web Services (AWS) was given a proposal. A visual effects (VFX) studio wanted to launch, and it wanted to do it entirely based on the cloud, a feat that, up until then, had never been done before.

Untold Studios was founded by Rochelle Palmer, Darren O’Kelly, and Neil Davies with the key philosophy of hiring the most “exceptional creative people” regardless of where they were located, and that meant not being tied to any particular physical location – or data center.

“Initially, when we spoke to AWS about it, they said it’s never been done before to do that kind of high fidelity imagery in the cloud, and it just wasn’t possible,” Palmer says. “We said, ‘we want to do it,’ so they built a lab for us.”

“We spent around three months testing at the beginning, and then we opened. There were lots of people who didn’t believe we were fully cloud, they would come and look for secret rooms in the back,” she adds, laughing.

Today, Untold is an established studio with an impressive array of Hollywood projects under its belt, with VFX work for blockbuster movies including Mission: Impossible – Dead Reckoning and Sonic the Hedgehog 3 showcased on its website, alongside work for many other clients.

Amaan Akram, CTO of Untold Studios, tells DCD that part of the draw of the cloud is that the company doesn’t want to be bogged down by its technology. While it has physical spaces and “studios” for staff in London, LA, and Mumbai, those facilities focus on providing a “physical space for employees” because “people need to see people, and we want to make sure that our creatives are sitting together and working together as much as possible.”

VFX and the cloud

The concept of rendering and creating VFX in the cloud is growing in popularity. DCD has reported on the topic in the years since Untold launched – with Wētā FX using AWS to render the compute-intense movie Avatar: The Way of Water, after an entire data center was not enough to handle the workload.

Cloud providers have added media and VFX-targeted offerings. AWS has its “Deadline Cloud,” and in April 2024, Akamai launched a cloud-based offering specifically for video processing.

That solution was based on Nvidia RTX 4000 Ada Generation GPUs, but John Bradshaw, Akamai’s field CTO, explains that while the GPUs are good for AI workloads, the company also offers a “VPU,” video processing unit, which is much better for the job.

“These are custom-built pieces of silicon that are designed to process video at massive rates, whether you are transcoding or changing the format, whatever you are doing – you can put a lot more through these cards than you can a GPU, and for a fraction of the price,” Bradshaw says. “When you get to a large scale, that starts to have an impact because if you are a VFX studio, you are burning through an awful lot of cycles.”

Akram says Untold Studios mostly relies on “many, many CPUs” on AWS, though GPUs are used when the company turns to AI.

“There are instances where we use both, and in the future, I think it will probably become a more equitable basis for CPU vs GPU,” he adds.

It uses AMD and Intel chips – as the company’s software isn’t built for one specific CPU configuration – accessed via AWS G4 and G6 instances, among others. According to AWS, the G instances are ideal for “graphics-intensive applications such as remote graphics workstations, game streaming, and graphics rendering,” and can also work well for AI workloads.

Notably, these instances do not feature Nvidia all-conquering AI chips. Akamai’s Bradshaw says this is probably because “AI-driven processors have ginormous back planes, so their ability to move gigabytes, or terabytes, of data in and out in fractions of a second is really important, because if you have an AI workload, you have to load the vector database in, and that contains all the relationships and the calculations you would run upon it. That is not very useful for video processing.

“By and large, GPUs are great to bring data in and out very quickly, and have huge capacity, but you’re not doing the same kind of maths in these cards as you do when you’re doing graphics or visualization.”

The company uses Nvidia GPUs for some of its VFX work, and the H200s for some AI work – adding that in the future it expects to use AWS’ own-developed chips for “internal AI needs.”

Beyond the G instances, Untold will tap into “spot instances” that can include anything from the C5, C6, M5, M6, R5, R6, or Z1 based on the needs of the project.

AI-generated content

The use of AI in the world of VFX and “high-end entertainment media” is becoming increasingly prevalent, though not without its challenges.

Stability.AI is known for its “Stable Diffusion” text-to-image tool, but has since progressed to AI-created video. Notably, in September 2024, James Cameron – the director behind VFX-fueled movies including Avatar and its sequel, among many others – joined the board of directors at Stability.AI.

At the time, Cameron said: “I’ve spent my career seeking out emerging technologies that push the very boundaries of what’s possible, all in the service of telling incredible stories. I was at the forefront of CGI over three decades ago, and I’ve stayed on the cutting edge since. Now, the intersection of generative AI and CGI image creation is the next wave. The convergence of these two totally different engines of creation will unlock new ways for artists to tell stories in ways we could have never imagined.”

That same month, Lionsgate – the studio behind the likes of John Wick and The Hunger Games – entered into a partnership with Runway, giving the AI company access to Lionsgate’s archive to develop an AI model exclusively for the Hollywood studio. Runway has previously partnered with AWS to train its models.

Another AI startup targeting video generation is Moonvalley, which was co-founded by former DeepMind researchers who worked on Google’s own video-generation model, and signed a multi-year contract with CoreWeave to use Nvidia GB200 NVL72 systems to train its models earlier this year.

But the use of AI-generated content comes with important considerations.

Issues remain around the ownership of the content. With AI models trained on existing creative works – that are human-made – copyright laws become a complex topic to navigate, and certainly one that Hollywood does not appear to want to deal with.

AI startups offering these services often make guarantees that all content used for generation is fully licensed, but this is not always enough for media executives.

“People create the artwork that goes into these models,” Untold’s Akram says. “On top of that, clients do not want us using data that comes from foundations they are not sure about. We’re very sensitive to their requirements as well, and we want to make sure that we don’t upload work to any of these many generative AI companies, as we don’t know how it will be used, and it might end up on some other customer’s generated content.”

As a result, Untold has its own models developed from its own content. This is another area where Untold benefits from its close relationship with AWS. Both Palmer and Akram describe themselves as “creatives” first and foremost, and that training new models is the domain of AWS. “We need the real experts working on that process,” Akram says.

Untold’s use of AI seems to be more so in the minutiae of VFX, rather than replacing the actual artists. It is currently working on a “de-aging model” to make actors appear younger.

“The people we are working with want absolute control over image quality; they will obsess over pixels,” says Akram.

Palmer adds: “We’re talking about milliliters of fluid in an animal’s eye or something. It’s that sort of detail, and we aren’t in a space where you get given an image and it can’t be changed.”

This level of detail makes creating convincing VFX very difficult. Palmer says one of the most common complaints from artists is about virtual water. “It’s very difficult to CGI and not have it look like jelly,” she says.

She recalls a particularly challenging project making adverts for Virgin Media: “They wanted creatures doing things they wouldn’t ordinarily do,” Palmer says. “The first thing we did was a highland cow on a motorcycle, and then a goat on a hang glider. Virgin asked; ‘What would be the most complicated thing to do?’ And we said probably a walrus – with the blubber and water and how it all moves – and we did that on a speed boat.”

Storage and scalability

Compute complexity is also challenging for VFX studios. Untold worked on a series of adverts for the Super Bowl, an effort that took a massive amount of processing power.

“It was five huge commercials with effects that all needed a lot of complex computing power, and had to be delivered on the same day,” Palmer says.

Untold usually works on many projects simultaneously, but these are typically at different points of development and do not need the same amount of compute. This was not the case for the Super Bowl. The studio calculated that to render all five of the commercials on one artist workstation would have taken 138.4 years. Instead, via AWS, Untold was able to leverage as many as 1,470 EC2 nodes to get the work done in time.

This scalability, often touted as a key benefit of cloud computing, is particularly useful in VFX, and applies not only to the compute needed for rendering and other workloads, but also for storage.

Untold has “petabytes and petabytes,” and while Akram says the company is careful with what needs to be kept “live,” it is still a huge trove of content to manage.

Even within the cloud, physical location is an important consideration. Untold operates in multiple cloud regions that are logically selected for proximity to its studios.

Akamai’s Bradshaw says location is also important from a provider perspective. The firm’s origins as a CDN vendor have given it a disparate footprint of 4,200 locations, and it would be wildly impractical to deploy its VFX and media cloud offering at each location, not least because of how expensive the required hardware is.

“We focused on our core data centers and cloud locations,” Bradshaw says. “I think there are five or six in Europe, and we chose the locations as they also contain our object storage offering, which makes it much easier to pull the file out and do whatever you need to do.”

While Akamai caters for a variety of media needs, including broadcasting and streaming, Bradshaw says location is particularly important for VFX.

“You get these rushes and you need to process things very quickly so that a producer or director can make their judgment call,” he says. “These are ginormous files – they could be 4K, 8K, or more video, and you need to be able to access them quickly. If that needs to travel halfway around the world before it can be edited, it’s going to be an uncomfortable experience.”

While Untold has embraced multiple locations, other studios take a different approach.

Shadow Magic Studios is a cloud-based production company with an integrated digital lab services offering.

The studio isn’t handling VFX workloads, but Jordan Maltby, its founder, says it faces similar challenges. Maltby says the company’s decision to hit the cloud was motivated by a desire to bring people and talent from any location in the world.

“The first step was ‘How do we get to the cloud?’,” he says. “Then, you start realizing that the cloud is in data centers in specific regions. Even with top-tier cloud providers, if it’s stored in New York and you do a pick-up shoot in California, the upload speed is halved. And it’s very, very important that we keep up with those uploads, because they have a very specific turnaround time.”

To get around this, Shadow Magic Studios uses a storage solution from Storj, which keeps the data “hot” and in the cloud in a way that means people can work from anywhere – “Even a coffee shop in Paris,” Maltby says.

The Storj platforms breaks files into 64 megabyte chunks that are erasure-coded across the vendor’s network. The data is distributed across large numbers of drives across Storj’s network – or, as Malby says, “scattered across data centers all around the world” – and can then be reconstituted by a node nearby to whoever needs to access it.

This apparently speeds things up considerably, particularly as, Maltby says, a lot of data needs to be kept hot “to some extent” throughout a project.

“Projects generally have a life cycle,” he says. “There’s a time when the project is being built, when you are working on it, when it needs to be distributed, and when it more or less goes to the graveyard to die.

“For the first two stages, you need immediate access to data, so hot storage. It needs to be on-demand to download almost to the millisecond. In the distribution phase, the core assets are not being used as much, but you are sending them to theaters for distribution.

“During this stage it’s common that small things need to be changed – a shot, or an error in the visual effect, or a region doesn’t allow certain content, so it needs to be modified,” Maltby explains.

If a project has been converted to cold storage, it suddenly becomes very expensive to access, and can take several hours. “If you’re in a high-stakes environment and need that media right away, that’s a big problem,” Maltby adds.

An industry scale shift

What seems to be universally shared, however, is the view that VFX studios’ use of cloud computing will only continue to grow.

Like many industries, media and film production has evolved since the Covid-19 pandemic, while other factors such as the California wildfires and the writers’ strikes have caused many studios to review the centralized nature of their work and conclude that it can be done away from Hollywood in other parts of the world. This is where the cloud comes into its own.

Untold, naturally, is a cloud believer, noting that other studios have followed in its footsteps and migrated to virtual infrastructure. Palmer notes that there are considerable financial benefits, as the cloud enables studios to transition from a capex model to an opex one, meaning they do not have to invest in expensive technology assets that depreciate in value.

While there is some debate about the financial longevity of this – many organizations report spiralling cloud bills – the cliché that time is money holds true, particularly in the fast-paced environment of film production.

For Untold’s Akram, the equation is not simply financial, but productivity-based.

He says: “I come from an artistic and technology background, and with on-prem, I never had the compute available for the project I was working on – it was always under somebody else’s desk. Brownies could be exchanged, and I’d get it for the day, but that’s not the best use of time. The productivity argument for the cloud is extremely strong.”