The design meeting for a new production has a familiar shape. The director describes something she can see perfectly clearly in her head. The set designer nods. The lighting designer nods. Six weeks later, in the room, it turns out all three people were imagining different things, and there is no money left to fix it.

This is not a failure of talent. It is a failure of medium. Theatre and live music are time-based art forms being planned with static tools — sketches, mood boards, ground plans. None of them can show how something moves, or how long a moment breathes. Directors have always had to hold the temporal dimension in their heads and hope everyone else was holding the same one.

A number of companies and touring artists have started using AI video generation to close that gap, and the interesting thing is that most are not using it to make finished work at all.

Pre-visualisation for the rehearsal room

The dominant use is concept development. Before the build, before the load-in, a designer generates short moving sequences of a proposed visual idea — the way a projection might bloom across a scrim, how a colour shift lands against a set, what a transition feels like at the intended tempo.

These are not presentation pieces. They are arguments made visible. A twenty-second clip settles a debate that three meetings could not, because everyone is finally looking at the same thing.

The technical requirement here is duration. Four-second fragments cannot show pacing, and pacing is the entire point. Tools that generate up to 30 seconds of continuous video in a single pass — Seedance 2.5 among them — are long enough to hold a complete theatrical beat, which is the minimum unit anyone can meaningfully respond to.

Reference capacity matters just as much. Feeding in the actual set model, the designer’s renderings, the costume palette, and a clip of the movement quality being aimed for produces something recognisably related to the production. Feeding in a text prompt alone produces generic theatre, which helps nobody.

Teasers and the touring problem

The second use is promotional, and it solves a scheduling problem that has frustrated marketing teams for decades.

Tickets go on sale months before anything exists to photograph. The set is unbuilt, the cast is not in costume, the show has not been lit. Marketing needs assets anyway, so it falls back on a poster and a text card — the least compelling material the production will ever release, deployed at the moment when it needs to be most persuasive.

Generated teaser content offers a way to communicate atmosphere before the physical production exists. A visual mood, a fragment of the world, a suggestion of tone. For touring musicians the equivalent is a visual identity for a single or a tour that arrives with the announcement rather than three weeks after it.

Native vertical output at 9:16 handles the platforms where these announcements actually spread, while 16:9 versions serve venue websites and press. Generating at the right ratio rather than cropping preserves composition that a crop would wreck.

Concert visuals and the economics of ambition

Touring acts below arena level have long faced a hard limit: screen content costs the same whether you are playing to 800 people or 8,000, but only one of those rooms pays for it. The result is that most mid-level tours run either no visuals or the same static loop for the whole set.

Generated content changes the arithmetic. An artist can develop distinct visual worlds for eight songs across a few weeks, iterating between rehearsals, at a cost that does not require a label advance. The material sits behind the performance rather than competing with it, which is exactly the register generated video is currently best at.

The honest limits

It would be a mistake to oversell this to a readership that knows what live performance actually requires.

Generated video cannot act. It produces surfaces convincingly and interiority not at all, and theatre is an interiority business. Nothing here substitutes for a performer making a choice in front of an audience.

It cannot be relied on for technical accuracy. A generated sequence showing a set piece moving is a suggestion, not a plan. Somebody with a tape measure still has to confirm the thing fits through the door.

And there is a real conversation to be had about whose work this displaces. Projection designers, video artists, and motion designers are colleagues, not overhead. The companies handling this well are using generation to expand what a designer can explore, and paying the designer to do the exploring — not removing the role and hoping a producer with a prompt window can replace it. Disclosure matters too: audiences and funders are entitled to know what was generated, particularly in a form built on live human presence.

Where it is heading

The most likely settled position is that AI video becomes standard in the development phase and stays rare in the finished work. Pre-visualisation, pitch material, teaser content, and reference for departments — all places where speed serves the art. The stage itself remains a room where real people do real things in real time, which is the one thing no model will ever generate.

For a company deciding whether to try it, the low-risk entry point is the next design meeting. Generate three versions of one contested visual idea, put them on a screen, and see whether the conversation gets shorter. Tools like this AI video platform earn their place in a production workflow by saving argument time, not by replacing anyone’s craft.