Runway (company)

Based on Wikipedia: Runway (company)

In the climactic fight scene of Everything Everywhere All at Once, Michelle Yeoh battles interdimensional threats while the screen fractures into a kaleidoscope of visual possibilities. That Oscar-winning film used tools from a company that most viewers had never heard of. The same company helps edit The Late Show with Stephen Colbert. It counts Google, Nvidia, and Salesforce among its investors. And it was founded by three graduate students who met in an art program.

Welcome to Runway, a company betting that the future of filmmaking will be written in text prompts.

From Art School to Artificial Intelligence

The origin story begins at New York University's Tisch School of the Arts, specifically in a program called ITP. That stands for Interactive Telecommunications Program, though the name undersells what happens there. ITP is essentially a graduate program for people who want to make weird things with technology. It attracts artists who code and engineers who paint.

In 2018, three students crossed paths there. Cristóbal Valenzuela and Alejandro Matamala came from Chile. Anastasis Germanidis came from Greece. They shared a fascination with what happens when you give artists access to machine learning tools. Not machine learning as a dry technical discipline, but machine learning as a creative medium.

They founded Runway that same year with a straightforward pitch: what if creators could use sophisticated artificial intelligence models as easily as they use Photoshop? The company raised two million dollars to build a platform that would deploy machine learning models inside multimedia applications. In practical terms, this meant making neural networks accessible to people who had never written a line of code.

At the time, this seemed almost quaint. Machine learning models were powerful but clunky. Using them required technical expertise, expensive hardware, and patience. Runway wanted to change that by building an interface layer between cutting-edge research and everyday creative work.

The Funding Escalator

What happened next tells you something about how quickly the artificial intelligence landscape shifted.

In December 2020, Runway raised eight and a half million dollars. A year later, thirty-five million. By December 2022, fifty million more. Then in June 2023, everything changed. The company closed a one hundred forty-one million dollar extension that valued it at one and a half billion dollars.

The investors in that round read like a who's who of technology giants with AI ambitions: Google, Nvidia, and Salesforce. These aren't companies that write checks casually. They saw something in Runway that suggested the company might become essential infrastructure for the next generation of content creation.

By April 2025, Runway raised another three hundred eight million dollars led by General Atlantic, pushing its valuation past three billion. In seven years, the company went from a two million dollar seed round to unicorn status and beyond. The art school project had become a serious bet on the future of visual media.

The Stable Diffusion Moment

To understand Runway's significance, you need to understand what happened in August 2022.

That month, the company co-released something called Stable Diffusion. If you've spent any time on the internet in the past few years, you've probably seen images created with it. Stable Diffusion is what's known as a latent diffusion model. That's a technical term for a particular approach to generating images from text descriptions.

Here's how it works in simplified terms. Imagine teaching a computer to recognize patterns in millions of images, then running that process in reverse. Instead of looking at a picture and describing it, the model takes a description and generates a picture that matches. The "latent" part refers to a mathematical trick that makes this computationally feasible by working in a compressed representation of images rather than pixel by pixel.

Stable Diffusion was released as open-source software, which meant anyone could download it and run it. This was unusual. Most powerful AI models at the time were locked behind corporate walls. The decision to make it freely available democratized access to image generation in a way that caught the technology world off guard.

Runway developed this in collaboration with the CompVis Group at Ludwig Maximilian University of Munich. Stability AI provided the computing resources needed to train the model. The partnership produced something that would reshape creative industries and spark intense debates about art, authorship, and automation.

From Images to Motion

Text-to-image generation was just the beginning. In February 2023, Runway released Gen-1 and Gen-2, turning its attention to video.

Gen-1 approached video generation through transformation rather than creation from scratch. You give it a source video and instructions about style or composition, and it synthesizes a new video that applies those instructions to the original structure. Think of it like a filter, but one that understands content and context rather than just adjusting colors.

Gen-2 went further. It could generate entirely new videos from text descriptions, images, or video clips. This made it one of the first commercially available text-to-video models. You could type "a golden retriever running through a wheat field at sunset" and the model would create video footage matching that description. The footage wouldn't exist anywhere else. It would be synthesized from patterns the model learned during training.

The quality wasn't perfect. Early text-to-video models produced results that often looked dreamlike or slightly wrong, like memories filtered through fever. Hands appeared with too many fingers. Physics behaved strangely. But the technology improved with each iteration.

Gen-3 and the Training Data Controversy

Gen-3 Alpha arrived with claims of improved fidelity, consistency, and motion. Runway built it on new infrastructure designed for large-scale multimodal training. The company described it as a step toward building what they called General World Models, a reference to AI systems that understand how the physical world works well enough to simulate it convincingly.

But Gen-3 also brought controversy.

According to reporting by 404 Media, the training data for Gen-3 came from sources that raised ethical questions. A former Runway employee alleged that the company organized a company-wide effort to compile videos into spreadsheets, then downloaded them from YouTube using a tool called youtube-dl routed through proxy servers to avoid detection.

The implications were significant. Training AI models requires enormous amounts of data. Where that data comes from, and whether the original creators consented to its use, remains one of the most contentious issues in artificial intelligence development. YouTube's terms of service prohibit downloading videos without authorization. And the allegation that the training set included "potentially pirated films" suggested even murkier territory.

In tests, 404 Media found that typing the names of specific YouTube creators into Gen-3 would generate videos in their recognizable styles. This suggested the model had learned enough from individual creators' work to reproduce their aesthetic choices. Whether this constituted theft, transformation, or something new entirely depends on who you ask.

The Latest Generations

Gen-4 arrived in March 2025, marketed as Runway's most advanced model yet. The company emphasized its ability to maintain consistency across scenes, generating characters, objects, and environments that stayed coherent even as the video continued. This addressed one of the persistent weaknesses of earlier models, which often produced characters whose appearance drifted unpredictably.

A month later came Gen-4 Turbo, a faster and cheaper version. The economics of AI video generation matter enormously for adoption. If creating a minute of video costs fifty dollars in computing resources, the technology remains a curiosity. If it costs fifty cents, it becomes a tool that changes how video gets made.

Runway uses a credit-based pricing system, where users pay for the computational resources they consume. Making the Turbo model more efficient meant the same creative output could be achieved with fewer credits. This kind of optimization often matters more than raw capability improvements for driving adoption.

Performance Capture Without the Capture

One of Runway's more intriguing tools appeared in October 2024. Act-One lets users upload a video of themselves performing, then transfers that performance onto a generated or animated character.

To understand why this matters, consider how animated films traditionally work. When Pixar wants to capture an actor's performance for an animated character, they often use motion capture technology. The actor wears a suit covered in reflective markers. Dozens of cameras track those markers as the actor moves. Sophisticated software interprets that data to drive the digital character.

It's expensive. It requires specialized equipment and expertise. Most creators can't access it.

Act-One aims to democratize something similar using ordinary video. Record yourself delivering a line on your phone. Upload it. The model transfers your performance, including subtle elements like eye movements and micro-expressions, onto a digital character of your choosing. No motion capture suit. No special camera setup. No technical team.

Act-Two expanded this to include body movement and gestures, adding environmental motion as well. The goal is comprehensive performance transfer, letting anyone animate characters with the nuance that previously required Hollywood budgets.

Hollywood Takes Notice

The entertainment industry's relationship with AI-generated content has been complicated, to put it mildly. The 2023 Hollywood strikes included significant provisions about artificial intelligence, with writers and actors seeking protections against being replaced or replicated by AI systems.

Yet the same industry has also been experimenting with these tools. Runway's involvement in Everything Everywhere All at Once and The Late Show with Stephen Colbert demonstrated that AI tools could find places in professional production workflows.

In September 2024, Runway announced a partnership with Lionsgate Entertainment that suggested deeper integration ahead. Lionsgate, the studio behind franchises like John Wick and The Hunger Games, agreed to let Runway train a custom AI model on its proprietary catalog of over twenty thousand film and television titles.

This custom model would be exclusive to Lionsgate. Other Runway users couldn't access it. The arrangement gave the studio an AI tool that understood its specific visual language, its cinematographic patterns, its aesthetic sensibility. Lionsgate filmmakers could use this for previsualization, planning shots before committing resources to actual production, and for post-production work like generating alternative takes or extending scenes.

The deal represented a template that other studios might follow. Instead of training on scraped internet data of uncertain provenance, studios could build AI models on content they unambiguously owned.

The AMC Partnership

Cable television followed Hollywood's lead. In June 2025, AMC Networks became the first cable company to formally partner with Runway. The network behind shows like The Walking Dead, Mad Men, and Breaking Bad planned to use Runway's technology for marketing images and pre-visualization.

AMC's executive Stephanie Mitchko framed the partnership around enhancement rather than replacement. The goal was to give creative partners better tools to realize their visions, not to generate content that would displace human creativity.

This distinction matters enormously in how the technology gets discussed and adopted. The same tool can be framed as threatening or empowering depending on context. A paintbrush in the hands of an artist is creative expression. A factory that produces paintings without artists is something else entirely. Where AI video generation falls on that spectrum remains contested.

IMAX and the AI Film Festival

Runway has been actively working to establish AI-generated content as a legitimate creative medium rather than a technological curiosity. Central to this effort is the AI Film Festival, which the company has hosted annually since 2023.

The festival's growth tells its own story. In its first year, it received about three hundred submissions and screened films in small New York City theaters. By 2025, submissions exceeded six thousand. The culminating screening that year sold out Alice Tully Hall at Lincoln Center, one of the most prestigious venues in the city.

In August 2025, Runway partnered with IMAX to screen AI Film Festival winners in ten major American cities. IMAX's Chief Content Officer Jonathan Fischer noted that the IMAX experience "has typically been reserved for the world's most accomplished and visionary filmmakers." Opening those screens to AI-generated content suggested the company saw something more than a passing trend.

The partnership with the Tribeca Film Festival, running since 2024, reinforced this positioning. Tribeca CEO Jane Rosenthal emphasized the importance of engaging with AI companies directly rather than "avoiding or fighting" the technology. A collaboration called "Human Powered" showcased short films and music videos created with AI tools, followed by conversations with their creators.

The 48-Hour Challenge

Runway also runs a competition called Gen:48, a filmmaking challenge with strict constraints. Participants get forty-eight hours and a supply of AI generation credits to create short films between one and four minutes long.

The format deliberately pushes experimentation. Forty-eight hours isn't enough time to overthink. It's barely enough time to sleep. Participants must work quickly, making decisions based on what the tools can actually do rather than what they wish they could do. The constraints force a particular kind of creativity, where technical limitations become creative boundaries to work within rather than obstacles to overcome.

Winners receive cash prizes, additional Runway credits, and the opportunity to screen their films at the AI Film Festival. The competition creates a pipeline of creative work that demonstrates the technology's possibilities while building a community of artists who understand its capabilities and limitations.

Game Worlds and New Directions

In 2025, Runway launched something called Game Worlds, a tool for creating and playing text-based adventures accompanied by AI-generated images. This represented a different application of the underlying technology: interactive storytelling rather than fixed video content.

Text adventures have a long history in computing, stretching back to games like Zork in the late 1970s. The player types commands like "go north" or "open door" and receives text descriptions of what happens. Adding AI-generated images to this format creates something hybrid, preserving the imaginative latitude of text while providing visual grounding.

Whether Game Worlds becomes a significant product or remains an experiment remains to be seen. But it demonstrates Runway's willingness to explore applications beyond straightforward video generation.

The Education Pipeline

One indicator of a technology's staying power is whether it gets taught in schools. Runway has been incorporated into design and filmmaking curricula at major universities, including NYU's Tisch School of the Arts, where the founders met.

This creates an interesting loop. Students learn to use Runway's tools as part of their creative education. Some become filmmakers who incorporate those tools into their professional practice. Others become teachers who pass the knowledge on. The technology becomes embedded in how a generation of creators thinks about making visual media.

For Runway, this represents both validation and strategic positioning. If the next generation of filmmakers grows up using AI generation tools as naturally as they use editing software, the company that provided those tools during their formative years has a significant advantage.

What Runway Represents

Time magazine named Runway one of the 100 Most Influential Companies in the world in June 2023. That recognition captured something about the company's significance that goes beyond its specific products.

Runway represents a particular vision of how artificial intelligence and creative work might coexist. In this vision, AI tools augment human creativity rather than replacing it. They lower barriers to entry, letting more people tell visual stories. They accelerate production, letting ideas become images and images become motion with unprecedented speed. They democratize capabilities that previously required expensive equipment and specialized skills.

The opposing vision sees the same technology as a threat. If a computer can generate footage indistinguishable from human creation, what happens to the humans who previously created that footage? If training data comes from the work of artists who never consented to its use, is the resulting model built on theft? If studios can preview films before hiring crews, will they decide they don't need crews at all?

Both visions contain truth. The technology is simultaneously empowering and threatening, depending on where you stand and what you do. Runway has positioned itself as a company that serves creators, but the same tools could serve those who would displace them.

The Road Ahead

With billions in valuation and partnerships with major studios and festivals, Runway has moved well beyond its art school origins. The company now operates at the intersection of artificial intelligence research and commercial entertainment, a position that carries both opportunity and responsibility.

The training data controversies won't disappear. As AI models become more capable, questions about where their capabilities came from become more urgent. The artists whose work trained these systems have legitimate grievances, regardless of the legal gray areas involved.

The quality gap between AI-generated and traditionally produced content continues to narrow. Each new model generation brings improvements in fidelity, consistency, and control. At some point, the distinction may become invisible to viewers. What that means for the industry remains genuinely uncertain.

For now, Runway occupies a strange position. It's a company founded by artists that raises questions about what art means. It's a startup valued in the billions that began with three graduate students who wanted to make weird things with technology. It's a tool that empowers creators while potentially training on their work without permission.

The three founders who met at ITP couldn't have predicted all of this. They wanted to make machine learning accessible to artists. They succeeded, and in succeeding, they helped create a world where the relationship between human creativity and artificial intelligence would become one of the defining questions of our time.

How that question gets answered will shape not just Runway's future, but the future of visual storytelling itself.