AI is showing up in Hollywood for two reasons: people think it could unlock new creative options, and studios also see a chance to speed up production. In Paramount Skydance's latest earnings call, CEO David Ellison argued that AI could end up mattering more to filmmaking than CGI ever did. He called it a potential "creative unlock" and said the company is developing tools to move work faster across its business.
That comparison is useful - but it doesn't answer the hard questions. CGI expanded what filmmakers could show on screen, but it didn't automatically improve stories, cut budgets across the board, or eliminate the need for skilled artists. AI's impact will depend on where studios actually use it, how rights and consent are handled, and whether any cost savings make their way into creative decision-making.
This article separates the promise from the day-to-day reality. We'll look at Ellison's claims, where AI could shift production economics, what it could mean for artists and creative quality, and how studios can test these tools without confusing "faster output" with "better filmmaking."
What did David Ellison actually claim?
On Paramount Skydance's earnings call, Ellison described AI as a production technology that could broaden what filmmakers can make. He compared AI's potential impact to CGI and digital compositing arriving in the early 1990s, and he pointed to films like Jurassic Park and Terminator 2 as examples of stories those tools helped make possible.
His pitch comes in two parts. First, AI could make production more efficient - making content easier to produce, projects faster to move, and cross-company workflows different. Second, lower costs could make it more realistic to take creative risks on original stories. He used Terminator to illustrate the scale: he said a film that cost $4 million to make would cost $60 million to $70 million to recreate shot for shot today. His takeaway was that AI tools could make projects at that level more financially manageable.
But that's still a claim, not a demonstrated result. The call didn't lay out which AI systems Paramount has built, what production tasks they cover, or how much money or time they've saved. And comparing past and present budgets doesn't tell you how much cheaper a new production could become with AI - remakes operate in different conditions, with different expectations, workflows, talent needs, distribution realities, and technical requirements.
Ellison also made a business-facing point: he suggested premium intellectual property and live events might become more valuable as AI-generated material becomes widespread. He framed AI as something for artists - not a replacement - and argued that audiences would still value handcrafted storytelling amid what he called "AI slop."
There's a clear tension here. A studio can see automation as a way to reduce costs, while artists may worry that the same tools reduce paid work or dilute control over their contributions. Both outcomes can be true at once. The useful question isn't whether AI is good or bad for film. It's which parts of production change, who benefits, and what quality controls stay in place.
How could AI change film production economics?
A film budget isn't one number you can neatly squeeze. It's labor, locations, sets, equipment, post-production, insurance, schedules, reshoots, and financing risk. AI may help with certain tasks, but a cheaper step doesn't guarantee a cheaper whole film. Sometimes new workflows add review time, legal checks, or extra revision rounds - so the savings that look obvious at first can shrink quickly.
The strongest near-term case often isn't "generate a finished scene." It's reducing friction around work that already happens: testing visual directions, organizing production information, creating temporary assets, translating materials, or helping teams search and analyze large amounts of footage. Those are realistic places to start testing, and they're not the same thing as claims about what Paramount's current tools can already do.
Where efficiency might show up
A studio can look for AI benefits in stages where the work is repeatable and a human can verify the output. For example:
- Development: organize research, compare script drafts, or build rough visual references for discussion. These can help teams explore options, but they shouldn't replace editorial judgment or factual checking.
- Pre-production: draft provisional storyboards or help with planning and document searches. The value is faster iteration - assuming the director and relevant departments keep the final approval power.
- Post-production: support tasks like transcription, rough localization, or asset organization. This might reduce administrative effort, but it still needs to be tested in the real workflow and language pairs involved.
- Marketing and distribution: adapt approved materials for different platforms or markets. That can save time, but each version still needs brand and rights review, plus factual checks where they apply.
These use cases don't carry the same risk. Indexing internal documents is different from generating a performer's voice or likeness. Studios shouldn't treat "AI" as one procurement bucket or assume every application has the same access to data and the same approval process.
Why lower costs do not guarantee more original films
Ellison's argument is that manageable budgets could make a wider range of stories financeable. That's plausible, but it hinges on what studios actually do with any savings. They might fund more projects, protect production quality, or reduce the risk profile of ideas they'd otherwise avoid. Or they might simply improve margins without changing what gets made.
And the same math spreads across the industry. If AI makes a task cheaper for everyone, the advantage may fade as competitors adopt similar tools. Meanwhile, audience attention and distribution slots remain limited. If AI also increases content volume, it could intensify competition rather than create a durable edge.
A meaningful test has to track more than "how often did we use the tool?" Compare total cost, turnaround time, rework rates, quality outcomes, rights exposure, and how many projects actually make it into production. If a workflow cuts a small expense but adds legal risk - or creates output that artists can't trust and have to fix - then it might not be an efficiency gain at all.
What does AI mean for artists and creative quality?
Calling AI a tool for artists doesn't settle how it should be used. Tools can expand what people can do, but they can also shift work away from them, change bargaining power, or make it harder to say who actually created something. The real impact depends on the task, the terms of use, and how studios handle staffing and credit.
In some workflows, AI can help teams see rough ideas earlier. A director can test visual approaches before committing to expensive production. An editor can automate searching through a big footage library. A writer can organize research or compare drafts using an approved system. In those cases, AI can assist while people stay responsible for selection, interpretation, and final decisions.
The boundary gets messier when a system generates material based on a person's distinctive work or likeness. Actors, writers, visual effects artists, designers, and other contributors all have different stakes, but the common issues are permission, compensation, attribution, job displacement, and control over future reuse. A studio's internal goal of saving time doesn't answer those questions. They require clear agreements and real consultation with the people whose work is involved.
Human review is a production requirement
"Human in the loop" should be built into the workflow, not treated like a final checkbox after a system has already shaped the work. Reviewers need to know what the model produced, what source material it relied on (and where that info is available), and what decisions were already made based on the output. Otherwise, the human review step can become something like "approve it quickly" without enough context to do a serious assessment.
At minimum, a practical policy should cover:
- Which tools and data sources are approved for each task.
- Whether employees or contractors can submit scripts, footage, or confidential production material to a third-party service.
- When a performer's voice, likeness, or style can be generated or altered - and what consent is required.
- Who checks generated work for accuracy, quality, rights, and cultural context.
- How AI-assisted contributions are recorded and how disputes get handled.
Legal requirements vary by jurisdiction and contract. In the US, rights issues can involve copyright, publicity rights, and labor agreements. In the UK and elsewhere, the rules may differ around data protection, copyright, and performers' rights. Studios operating across markets need specific legal advice, not one generic template.
Craft is more than avoiding automation
Ellison's contrast between "AI slop" and handcrafted storytelling taps into a real audience concern: volume doesn't equal value. But filmmaking has always used software and technical tools, from editing systems to digital effects. The question isn't whether the tool exists - it's whether it serves a clear creative purpose, whether the final work feels coherent, and whether the people responsible retain meaningful control.
AI can help turn a weak idea into something faster. It can also help a skilled team test an idea that would otherwise be too expensive to explore. The tool doesn't decide which outcome happens. Creative leadership, production constraints, review standards, and incentives do.
What should studios do next?
The sensible response to a big technology claim isn't to adopt it everywhere right away, and it isn't to reject it out of hand. Studios should start with specific production problems and test whether AI solves them better than existing methods. That may sound modest, but it avoids a common failure: buying a capability first and then looking for tasks that can be reshuffled to match it.
Start by mapping the workflow. Pick a task that consumes real time or money, has a clear definition of acceptable output, and doesn't force the AI to make an unreviewable creative decision. Then run a small pilot with a team that actually understands the current process. If the existing task is poorly defined, AI probably won't fix it - it may just make the confusion harder to diagnose.
Run a production pilot with a baseline
Before introducing any tool, document how the work happens today. Track elapsed time, labor hours, revision counts, error rates, and the cost of downstream corrections. Then run the same type of work using the proposed AI-assisted process. Make sure the comparison includes setup, training, review, legal checks, and rework. If you count only the time spent generating output, you'll overstate the savings.
Use a pilot scorecard that answers five questions:
- Did the task get faster or cheaper overall? Include the time spent reviewing and correcting the result.
- Was the output good enough for its intended use? Internal references have different standards than a final shot or a public-facing asset.
- Can the team explain how the result was created? Record the tool, version, data permissions, and human approvals.
- Were workers' rights and expectations respected? Check contracts, consent, credit, and any relevant collective agreements.
- Would the process still hold at ten times the volume? Consider support, storage, governance, vendor dependence, and quality control.
The last question matters because one department's demo can hide operational costs. A tool might work well in one team but struggle when multiple teams need access, when project data is sensitive, or when output has to stay consistent across a long production.
Set boundaries before scaling
Group use cases by risk. Internal search or administrative support may need standard controls for security and accuracy. A generated likeness, voice, script contribution, or final visual asset needs stronger review around permission, rights, provenance, and disclosure. Even the same model can be low risk in one workflow and unacceptable in another.
Keep a record of consequential AI-assisted work. You should be able to trace which tool was used, what material it processed, who approved the result, and where it ended up. That record helps teams investigate quality issues and answer questions from artists, partners, regulators, or distributors. It also makes it easier to stop a workflow that doesn't meet your standards.
Finally, decide up front how savings will be used. If the pitch is that AI will enable more original storytelling, define what that means in practice - more development slots, more low-budget productions, more time for creative iteration, or a different risk tolerance. Otherwise, "creative benefit" stays vague while only the cost-cutting story has measurable proof.
Conclusion
Ellison's CGI comparison captures why people think AI could change what production teams can test - and how quickly they can work. But the analogy isn't a business case on its own. The real question is whether a specific tool improves a real production workflow without creating hidden costs, weakening creative control, or shifting risk onto artists.
Studios should pilot narrowly, measure the full process, and set rights and review rules before scaling up. If AI really does make original storytelling more affordable, the evidence will show up in what gets funded and made - not in how quickly a system can generate material.