An unexpected image can loosen a fixed idea. A different colour relationship, an awkward composition or a surprising detail may suggest a direction we had not considered. This is where I find the metaphor of AI as a muse useful: a source of prompts and possibilities in a creative process led by a person.

Generative models can draw on patterns learned from large collections of images. Their range can make unfamiliar combinations available very quickly. But this should not be confused with an artist’s visual experience—the knowledge built through looking, making, remembering and living. A model’s training is not a human memory, and the scale of its training data does not make its judgement superior.

In practice, the exchange can be simple: describe an idea, examine the results, change the brief and try again. A few useful variations may be enough to move the work forward. Producing hundreds is not a goal in itself; too many similar options can make a weak idea harder to recognise.

The next step remains a matter of judgement. I can select a composition, redraw it, alter its rhythm or use it as a starting point for work in another medium. What matters is the relationship between the emerging image and the intention behind it, rather than the novelty of the tool that produced the first draft.

This way of working raises questions about authorship, but it does not make intention disappear. The creator still decides what is worth pursuing, what to reject and when the work says what it needs to say. For me, the promise of an AI muse is not that it replaces imagination. It is that an encounter with an unexpected result can help imagination move.