What Real Studios and Brands Built With Runway

Chapter 1

Runway: The Ad That Cost Four Thousand Dollars Instead of Five Million

A marketing director at a national financial services company is looking at two numbers on the same slide. The first is what a broadcast commercial used to cost the company to produce: north of five million dollars, the kind of budget that gets a campaign killed before a single frame is shot. The second is what the company's first AI-generated campaign cost, using Runway, before it aired on NFL Sundays: three to four thousand dollars.

That gap is not a rounding error and it is not a marketing claim written by the company doing the selling. It is one line inside a longer study Runway ran across hundreds of its own enterprise customers, and it is the reason this book exists. A leader deciding whether to bring AI video into their own organization does not need another feature tour. They need to know what already happened when someone else made the same call, and whether the result held up once it left the pitch deck.

The pattern repeats at a scale that stops looking like an outlier. A home goods retailer now replicates an eight-hundred-thousand-dollar visual project for under ten thousand. A global consulting firm rebuilt a three-hundred-to-six-hundred-thousand-dollar client campaign for roughly three thousand dollars inside a two-day window it would never have made otherwise. A US cable network turned a ten-to-fifteen-thousand-dollar-per-season end-credits process into a run that costs about ten dollars.

None of these figures are Runway's own pricing. They are what customers reported spending on the same work the year before, measured against what the same work cost after they switched. That distinction matters for the rest of this book: everything here is a documented outcome, not a projected saving, and the difference between the two is exactly what a leader needs before asking their own finance team to sign off on a pilot.

The question this book is built to answer is not "can AI make video." That was settled some time ago. The question is: who has already staked something real on it, what did they actually get, and where did it still fall short. The rest of the chapters follow that evidence in the order a decision-maker would actually need it — starting with the hardest test of all, which is not cost. It is whether the output is good enough that an audience stops noticing it was made by a machine at all.

Chapter 2

Whether The Work Actually Clears The Bar

Cost savings mean nothing if the finished video looks wrong. This is the test every skeptical stakeholder raises first, and it is worth taking seriously rather than dismissing, because for most of the history of AI video the skeptics were right.

Runway itself names the failure mode it spent years working against: the uncanny valley, the point where synthetic media gets close enough to real that its flaws become distracting rather than invisible. Strange eyes. Drifting faces. A mouth that almost syncs to speech but not quite. Individually small, collectively enough to keep an audience watching for the mistake instead of watching the story.

In a research effort the company calls Project Luxo, three fully AI-generated short films and a spec ad were shown to a wide range of people across the media industry — producers, actors, guild members, studio staff, press — and asked to rate them on whether the story held their attention and their emotional investment. The reported result, in Runway's own account, was consistent: every reviewer said the films worked as films, not as demonstrations of a tool.

A useful way to hold this claim at arm's length: it is Runway's own study of Runway's own films, and a decision-maker should read it exactly that way — evidence worth weighing, not a verdict to accept on faith. What makes it worth citing here is the standard it sets, because the same standard shows up independently from a completely different source. Electronic Arts' Chief Strategy Officer, discussing AI inside games rather than film, draws an almost identical line: "It's not enough for things to look real. That's what you would call a dream. It has to be real, which is what you'd call a simulation." In a game, a car's turning physics has to be felt by the player at sixty frames a second, not merely look correct on screen.

The two statements come from unrelated contexts and land on the same test: does the audience stop noticing the machine. Runway's own reporting says its recent film work has aired on national television with an on-screen AI credit, run during a major sporting event, and appeared in a trailer for a major franchise. Those are checkable claims a leader can ask their own team to verify against public record before treating them as settled — but they are a materially different kind of evidence than a demo reel, because a demo reel is built to succeed and a broadcast slot is not forgiving of failure.

Chapter 3

When A Studio Puts Its Own Name And Money On It

A vendor's case studies are one kind of evidence. A studio taking an equity stake in that vendor is a different kind, because equity is a bet the studio's own leadership has to defend to its board.

Lionsgate did exactly that in 2026, expanding a partnership that began in 2024 and taking an equity interest in Runway alongside a joint program to develop new IP together, starting with a short-form series built from Lionsgate's existing catalog. Lionsgate was already the first Hollywood studio to hire a Chief AI Officer and to build internal infrastructure around an AI strategy, which is worth noting for a reason beyond novelty: a studio with that much institutional exposure to the technology's actual limits chose to go deeper rather than pull back.

📰 Recommended for You

Do Hair Dyes Cause Cancer and Kidney Problems? Protect from Lead Poisoning?

Read More →

TeamoRouter Developer Architecture Guide

Read More →

Netflix took a narrower but more concrete kind of bet: a specific creative problem it could not otherwise solve. The father-and-son filmmakers behind a period piece recreating the 1970 World Cup wanted one shot — tilt the camera up from the actor playing Pelé and reveal a packed, roaring stadium. That shot was not affordable to build practically. Netflix's effects team blended practical photography with generative tools to fill the stadium, and the result, in the studio's own account, was the difference between a shot making the final cut and a shot being cut for cost. Netflix's own framing of the decision is worth carrying forward: the question they now ask on every production is what used to end up on the cutting room floor that AI finally lets them keep — not what AI replaces.

Electronic Arts' interest is different again, and arguably more skeptical than either of the others. Its Chief Strategy Officer explicitly resists what he calls "neural everything" — the instinct to hand every system over to a generative model regardless of whether the task calls for determinism. A basketball's physics should not be probabilistic. What EA is watching for instead is where generative tools expand what is possible without threatening the reliability a live, multiplayer game depends on, using its own long-running franchise The Sims as the clearest example of scale a generative layer could unlock.

The lesson for a decision-maker evaluating their own adoption is not "if Lionsgate did it, we should too." It is narrower and more useful: each of these organizations attached its bet to a specific, named problem it already had — a shot it couldn't afford, a franchise it wanted to scale, an IP pipeline it wanted to accelerate. None of them adopted the technology in the abstract.

Chapter 4

Getting Legal To Say Yes Before The Question Is Asked

A production or marketing leader can be fully convinced and still get stopped at the door by their own legal department. This chapter is for that conversation specifically, because two general counsels who have already had it — at Block and at Spotify — describe an approach that is unusually practical rather than defensive.

Block's Chief Legal Officer states the reframe directly: "If you are saying no, that is a higher risk than saying yes, because you fall behind so fast." Her counterpart at Spotify runs the same logic structurally, embedding legal review early enough in a product's development that the guardrails exist before anyone has to ask for permission — his own phrase for it is giving "the yes with caveats before the question even comes up."

Neither treats an AI feature as legally novel. Both describe running it through the exact same review any other product launch would get — copyright, privacy, employment law, the standard set — rather than inventing a new specialty called "AI law." That is a useful thing to bring into an internal conversation: a legal team that is worried about AI specifically, as its own new category of risk, may be over-scoping the problem in a way that slows the review without making it more rigorous.

Both also describe writing internal policy ahead of external regulation rather than waiting for it. Spotify's own account: regulators tend to react to a small, predictable set of concerns — privacy, copyright, transparency — so a company that addresses those concerns on its own terms, in its own timeline, is less exposed than one waiting to be told what the rule will say. Spotify's disclosure requirement for any AI-generated voice impersonating a real person is the example given as a policy the company wrote before being forced to.

The practical instruction for a reader trying to move this internally: don't bring your legal team a demo. Bring them the specific, narrow problem you want AI to solve, and ask them to run it through the review process they already use for everything else. Both executives describe that as the difference between legal becoming the department that blocks a project and the department that gets there first.

Chapter 5

What Happens After The Pilot Ends

The assumption behind most pilot programs is that enthusiasm fades once the free trial or the first quarter ends. Runway's own reporting on its enterprise customers describes the opposite pattern happening repeatedly, and it is worth taking seriously as a planning input rather than a sales claim, because it changes what a first-year budget for a pilot should actually look like.

A global consulting firm grew from twenty seats to one hundred within a year and now reportedly consumes half a million credits every two weeks. A Nordic national broadcaster expanded from twenty seats to fifty. A global production group is reported to have reached ninety-six percent seat activation at renewal across more than seventy production companies inside its group. None of these are numbers a vendor invents to look good in a slide — they are the kind of internal usage metric that only a customer, not Runway, would actually have reason to track and report.

Adoption also shows up in hiring decisions, which is a stronger signal than usage statistics because it reflects a company changing what it expects from every new employee, not just what its existing team happens to try. One large online education platform, described as one of the largest US advertisers by creative asset volume, reportedly made familiarity with Runway a mandatory line item in every creative job description it posts.

In the most mature accounts Runway describes, AI-generated work stops being a supplement and becomes the primary production mode: one home goods retailer generating three-quarters of all its visual media this way, one game publisher producing half of all marketing for its flagship title through the platform, one independent film studio building full-film animatics that are eighty percent generative video before a single frame of live action is shot.

The planning consequence for a reader running their own pilot: budget for the pilot to grow rather than conclude. The pattern across every account Runway describes is expansion after the first real test, not a return to the old process once the trial period ends. A pilot sized to prove the concept and then stop is, on this evidence, sized wrong.

Chapter 6

The Program Built For Teams That Haven't Committed Yet

Not every organization evaluating this is a Fortune 50 brand with a five-million-dollar production budget to compare against. A startup building a customer-facing product on top of generative media has a different problem: proving the idea works before spending real money finding out.

Runway Builders is the answer to that specific problem, and it does not sit anywhere in the site's main navigation — it lives at a standalone product page most visitors never reach unless they are already looking for it. It is open to companies from seed stage through Series C, and it offers up to five hundred thousand free API credits, the platform's highest API usage tier for throughput, direct access to Runway's real-time Characters API, and a private community with training sessions and direct support.

The program launched alongside Runway Characters — persistent, real-time conversational avatars built from a single photo, capable of appearing in customer support, onboarding flows, training simulations and interactive entertainment. The specific use cases Runway names as active interest from applicants: AI customer support agents, interactive brand characters, personalized onboarding, real-time sales assistants and synthetic media tools built into someone else's product.

Why this matters for a reader who is not a startup founder: it is the cheapest, lowest-commitment way to run a genuine test of what this technology can do inside a real product, before recommending a paid enterprise contract to anyone above you. Applying costs nothing but a form — company name, funding stage, a description of the intended use case, and which models are currently in use — and the credits are free, not a discounted trial that reverts to billing later.

The catch, stated plainly rather than left implicit: this is scoped to companies actively building a product, not to a marketing department wanting to test a campaign concept. An enterprise team evaluating adoption for internal production work will not qualify for Builders and should expect to run its first real test through a standard paid plan instead — but a startup considering the same platform for a product feature has a genuinely free way to find out before committing.

Chapter 7

What Nobody Has Proven Yet

Every case study in this book is real, and every one of them is also incomplete in the same specific way: none of it answers what happens when something goes wrong in front of an audience, rather than in a controlled study.

The researchers who built Runway's real-time Characters system are candid about where the open questions sit, and their own account is the strongest source for this chapter precisely because it is the team defending the work, not a critic attacking it. Asked directly about safety at a research conference, one team member described the two questions they hear most often: whether guardrails exist for experiences built for children, and how a character built for one function — reading a bedtime story, for instance — stays on task if a conversation drifts somewhere it shouldn't. Their own description of where this stands: "Moderation and the certainty of moderation came up repeatedly, and it's something we are actively prioritizing" — present tense, not a solved problem being reported after the fact.

A second open question sits inside the roadmap rather than the safety review. The same researchers describe navigable environments — a character existing in a world a person can move through, rather than a fixed conversational frame — as work they are "working toward," not work that ships today. A leader evaluating this for a customer-facing product should treat anything beyond the current conversational format as a stated intention, not a documented capability, and should ask directly what is shipping versus what is on a roadmap before committing a launch date to it.

There is also a plainer, more mundane gap worth naming: this book was written from what Runway itself publishes about its own case studies, partnerships and research. It is not an independent audit, and none of the dollar figures or seat-growth numbers here were verified against the customers named. Treat every number in this book the way you would treat a reference a job candidate supplies themselves — worth calling, not worth assuming.

None of this is a reason to dismiss the evidence gathered in the earlier chapters. It is a reason to run a small, real pilot inside your own organization before making a large claim to your own leadership based on someone else's case study. The organizations in this book that expanded after their pilot did so because the pilot itself produced evidence they trusted — not because Runway's report told them to.

Chapter 8

The Test Worth Running Before The Budget Meeting

If your goal is proving a concept cheaply before asking for real budget, Runway Builders or a standard paid plan run against one narrow, well-defined task is the right first move — not a full production pilot across your whole team.

If your goal is convincing a skeptical creative director, the Project Luxo standard is the one to borrow: don't show them a demo built to impress. Show them a finished piece and ask whether they noticed it was AI-made before you told them. That is the test Runway's own research used, and it is a fair one to apply to your own output.

If your goal is getting legal to move fast rather than slow you down, bring them a specific problem and route it through the review process they already run for everything else — not a special AI review that treats the technology as a new category of risk it usually isn't.

If your organization has already run a small pilot and it worked, the evidence in this book suggests budgeting for expansion rather than a return to the old process — that is the pattern across every account Runway describes, not the exception.

And if none of the case studies here match your situation — a documentary team, a small in-house department, a regulated industry with its own review process — the honest answer is that this book cannot tell you what happens in your specific case, because nobody has published that case study yet. The only way to get that evidence is to become the organization that runs it.

Questions readers actually ask

Contact / More useful information from RamthaMedia

Official source links:
Runway