Runway ML Review

Runway makes AI video, and it has been at the front of that field for long enough to have shaped how everyone else talks about it. Text to video, image to video, and video to video all run here.

The framing has widened considerably. Runway now describes itself as building general world models — systems meant to understand and simulate the physical world — with creative video as one application among several, alongside developer tools and robotics.

This review sticks to what a video team can actually do with it today: which models handle which job, what the workflow features are for, and the constraints that still define the medium.

Runway Platform Facts

Three Products Under One Name

Runway is no longer a single app, and knowing which line you are looking at prevents a lot of confusion when reading its material.

• Runway Creative — the video generation and editing platform filmmakers and marketers use
• Runway Dev — API access and developer tooling for building on the models
• Runway Robotics — the models applied to physical systems and synthetic data

The Models And What Each Is For

Runway ships several models rather than one general video generator, and matching the model to the task is most of the skill.

Gen-4.5

Gen-4.5 is the mainline generation model — the one behind most text-to-video and image-to-video work. It is where you start for creating a shot that does not exist yet.

Its notable strength is holding a subject or style consistent across separate generations, which is what makes assembling a sequence rather than a single clip realistic.

Aleph 2.0

Aleph handles transformation of existing footage rather than creation from nothing — restyling, altering, and editing video you already have.

For teams with real footage, this is often the more useful half of the platform. Modifying a shot you filmed avoids the control problems that come with generating one from a description.

Act-Two

Act-Two is performance capture: driving a character with a recorded performance so expression and timing come from an actor rather than from a prompt.

This addresses the weakest part of generated video. Prompt-driven acting reads as uncanny; a real performance mapped onto a generated character does not.

GWM-1

GWM-1 is the general world model underneath the newer work, aimed at simulating how the world behaves rather than producing pretty frames. It is the foundation for the robotics and synthetic data work.

Workflows, Agent, And Characters

The tooling around the models is what separates Runway from a generation endpoint. Workflows chain steps together so a repeatable process can run without rebuilding it each time.

Characters keeps a figure consistent across shots, which is the single hardest problem in generated video and the reason most AI films look like a collection of unrelated clips.

Agent and MCP support point at automation — Runway being driven by other systems rather than only by a person in the browser. Media Router handles moving assets between steps.

Who Is Actually Using It

Runway's user base spans professional film and advertising work, marketing teams producing volume, and independent creators. The vendor cites more than 60 million creatives across the platform.

In professional pipelines it is most often used for specific shots rather than whole productions — previsualisation, establishing shots, effects that would be expensive to film, and inserts.

For marketing teams the calculation is different: it produces usable video without a shoot, which changes what is possible on an ordinary budget rather than improving what a large one can already do.

Runway Against Other AI Video Tools

Compared with the newer general-purpose video models, Runway is less about producing one impressive clip and more about giving you the pieces to finish something. Transformation, performance capture, and character consistency are production concerns, not demo concerns.

That focus cuts both ways. If you want a single striking shot from a prompt, a simpler tool will get you there with less to learn. If you want six shots that look like they belong in the same film, Runway is built for that and most alternatives are not.

The world-model direction also signals where the company is going. Runway is increasingly betting on simulation rather than on being the prettiest video generator, which is worth knowing if you are choosing a platform to build on for years.

The Constraints That Still Define It

Clip length remains short. Generated video is measured in seconds, so anything longer is assembled from many generations, and holding continuity across that assembly is real work.

Control is coarse compared with filming. You can describe and influence a shot, but you cannot direct it the way you would direct a camera, and getting a specific result often takes many attempts.

Hands, text, and complex physical interaction remain the reliable failure cases. Generated footage still betrays itself in the details more than in the overall impression.

Cost accumulates through iteration, not through the final render. The generations you discard are the expense, and that is difficult to budget before you start.

Working With It Effectively

Start from footage or an image where you can. Image-to-video and Aleph's transformation of existing material give far more control than describing a shot from scratch.

Use Characters and consistent references from the first shot rather than trying to match later. Continuity is much easier to establish than to repair.

Plan for iteration explicitly. Budget generations the way you would budget takes, and accept that a usable shot may be the tenth attempt rather than the first.

Is Runway The Right Video Tool?

Runway is the most complete AI video platform for people doing real production work, and Aleph and Act-Two address exactly the problems that make purely generative video hard to use professionally.

Choose it if you are producing video seriously — for film, advertising, or high-volume marketing — and want transformation and performance capture alongside generation. Choose something simpler if you need occasional short clips and do not want to learn a platform.

Note that runwayml.com now redirects to runway.com, so older links and listings point at the previous domain. Check runway.com for current plans and credit limits before committing to a project.