Can Generative Artists Sell the System?

I have spent years making systems that produce images, and I am still not entirely sure what the real object is. The finished image can be collected. The code contains far more of the work, along with all the possibilities that never appear in a single output. Then there is the experience of watching the system run and finding things I did not expect it to make.

A music platform launched this week that brought that question back to me.

A market for instruments

On 17 September, Lunacy Audio launched Nova, a collection of cloud-based instruments that work in a browser and inside a digital audio workstation. There are ten instruments at launch, split between free and paid versions.

That alone would be a fairly ordinary software release. The part that caught my attention is Nova Builder, which Lunacy is still developing. The plan is to let people construct their own instruments from effect blocks and written instructions. They will be able to alter tools made by other people and eventually sell what they create through the same platform. The Verge has a useful account of the launch.

The Builder is not available yet, so the larger idea remains unproven. We do not know how creators will be paid or what happens when one person builds heavily on somebody else’s instrument. A cloud instrument also depends on the company continuing to support it. These are fairly large unknowns.

Even so, I think the model is interesting. It treats the instrument as a creative work in its own right. A musician can use it to produce any number of pieces, while the person who designed its behaviour still has something identifiable to sell.

Could this work for generative art?

I keep imagining a visual version of the same idea. An artist could build a peculiar drawing system, a new kind of shader or a way for an image to react to a viewer. Other people might use that system within clear limits, with the original artist credited and paid.

There is an obvious danger. A strong visual language could quickly become another menu of effects. Once a system is easy to copy, a recognisable idea can spread until it loses any connection with its maker. We have already seen how quickly AI products can flatten an artist’s work into a style option.

The model becomes more interesting if the artist controls what can be changed. Some parts might remain fixed. The system could also retain a visible history of who made it and how it has developed. That would make it feel closer to a living edition than a disposable preset.

For generative artists, this could open another form of collecting. A collector might own the right to run a system or take part in its development. A gallery could commission a version for a particular space. The artist would still have to decide where the artwork ends, which is already one of the central questions in long-form generative work.

Claude moves closer to the working project

Anthropic also made a useful change on 16 September. Claude is bringing its chat and Cowork functions together. Its document and presentation tools now sit inside the same conversation, alongside Claude Design. The work remains editable and can be exported into common formats. Anthropic shows the new tools here, and Reuters reported the rollout.

I can see myself using this around an artwork. A project could begin as a loose studio document and gradually become an exhibition proposal without being rebuilt from scratch each time. That would save a lot of tedious work.

I am wary of giving one assistant too much control over the presentation. If it writes the proposal and designs the deck, the whole project can begin to feel as though it belongs to the software. The answer is probably to give it much stronger material at the start and to be ruthless about anything that sounds generic.

A smaller institutional signal

Serpentine has announced the first group in its Future Art Ecosystems R&D Fellowship. Four practitioners receive £10,000 and six months to research subjects that include world models and XR.

It is a good programme and I will follow the work, especially Alice Bucknell’s research into world models. I see it as a lower signal for my own practice because it does not offer a new capability or route to market. Its main value is showing that institutions are prepared to support art technology while the questions are still unresolved.

Worlds with something inside them

The Hayward Gallery announced Cao Fei’s largest UK exhibition on 16 September. It will run from 20 April to 25 July 2027 and includes RMB City, Asia One and the first UK presentation of The Birth. Wallpaper has an early description of the five-part show.

This feels relevant as I move further towards XR. Cao Fei built RMB City in Second Life between 2007 and 2011, long before the current rush towards generated worlds. Her virtual spaces hold traces of work and urban life. They also contain the expectations people once placed on technology.

That is a much harder standard than immersion. A digital world needs something worth finding inside it. Scale and spectacle can get somebody through the door, but they will not give the world a reason to stay with them.

Keeping track of how a work was made

The 2026 Brisbane Portrait Prize opened this weekend after several artists who had boycotted its acceptance of AI returned to the competition. Of the 49 finalists, only one declared using generative AI. The prize requires AI-assisted work to comply with copyright law and places weight on the artist retaining control. The Guardian’s report includes serious concerns about First Nations cultural material being absorbed into training data.

The small number of declared AI works is revealing. Allowing them into a prize has not led to a sudden flood. It has made the questions around source material and authorship more specific.

I think artists who use code or machine learning should keep a plain record of how important works were made. That does not need to become part of the artwork. It should be enough to explain what tools were used and where the source material came from. It also gives the artist something solid to point to when the process is questioned later.

Nova is the development I will remember from this week. Its full marketplace may never work as planned, but the proposition is useful. The system itself can carry creative value. Generative artists have known this for a long time. We may be getting closer to having better ways to present that value, and perhaps to sell it.