Runway introduces Aleph video editing model

Background

Generative video models are usually described in terms of what they make from nothing: a text prompt goes in, a clip comes out. Editing existing footage is a different problem. The model has to leave most of a scene untouched while changing one specific thing — an object, the lighting, the position of the camera — and it has to keep that change consistent from frame to frame, because a video is a long sequence of images that a viewer sees in motion. Any flicker, drift, or shape that shifts between frames is immediately visible. Systems that take a video as their input and operate on it directly are often called in-context video models, because the footage itself is the context the model works from.

What happened

On 25 July 2025, Runway introduced Aleph, an in-context video model ^[1]. The company's announcement page for Aleph carries the same date ^[1]. Rather than generating a clip from a written description alone, Aleph starts from footage that already exists and performs edits on that input video ^[1].

What it does

Aleph can add, remove and transform objects in existing footage ^[1]. Those three operations cover a lot of ordinary post-production work: putting something into a shot that was not there, taking something out and filling the gap with plausible background, or changing how an object looks while keeping it in place. The model can also generate new camera angles for a shot, and change its style or lighting ^[1]. A new camera angle is the more demanding of these, since the original footage only ever recorded one viewpoint; producing another one means the model has to infer what the rest of the scene would look like from somewhere else and keep that inference stable across the whole clip.

Runway describes the range of capabilities broadly, saying that Aleph can carry out a wide range of edits on an input video, among many other tasks ^[1]. In other words, the pitch is not one specialised tool for one kind of fix, but a single model asked to handle many different editing jobs on the same piece of footage.

Availability

Aleph is available on all paid plans ^[1]. It is presented as part of the product that subscribers can use, not as a separate demonstration.

Why it matters

Runway describes Aleph as setting a new frontier for multi-task visual generation and as a novel step forward for manipulation ^[1]. Both halves of that description are worth separating. "Multi-task" refers to one model covering many kinds of edit instead of a separate system for each. "Manipulation" refers to controlled change to existing imagery — as distinct from generating something new — treated as a research direction in its own right ^[1].

For anyone making video, the practical appeal is straightforward. Reshooting a scene to change a detail, remove a distracting object, or alter the light is expensive and sometimes impossible once a set has been struck or a location is gone. A model that edits the footage directly moves that work into post-production. It also changes what a first shoot has to get right: if lighting and style can be adjusted afterwards, some decisions can be deferred.

Open questions

The value of a tool like this depends on details that are easy to state and hard to achieve. Does an edit hold for the whole shot, or drift after a few seconds? Does a removed object leave a believable background behind it? Does a new camera angle stay geometrically consistent with the original? Those are the questions viewers ask with their eyes, whether or not they know anything about the model behind the clip. Runway's announcement frames Aleph as a step in a research direction; how far that step carries in everyday production is something users of the paid plans will establish as they put it to work on real footage ^[1].