Artist: humans stay the directors

An artist working with artificial intelligence described the job less like “typing a prompt” and more like directing a cast: the human decides the scene, rejects bad takes, and keeps only the frames worth showing. That wording comes from Fellowship’s artist community, which has spent the past three years exhibiting work built with image and video models rather than treating them as one-click vending machines. (x.com, nftnow.com) That sounds simple, but it cuts straight into the biggest fight around generative art: who is actually making the work when a model can produce 20, 200, or 2,000 variations in minutes. The “director” framing says the machine supplies options, while the artist supplies taste, sequence, and the final yes or no. (x.com, link.springer.com) That distinction is not just cultural language. The United States Copyright Office said in guidance issued in March 2023 and a follow-up report released on January 29, 2025 that copyright still turns on human authorship, and that selection, arrangement, and modification by a person can be protected even when some source material came from artificial intelligence. (copyright.gov, copyright.gov) The same office also drew a hard line around prompting. Its 2025 report says text prompts by themselves usually do not give a user enough control over the expressive details of the output, because the system still decides the final pixels, words, or sounds. (copyright.gov) So “director” is doing real work here. A film director may not build the camera or act every role, but still determines framing, pacing, performance, and what ends up in the final cut; the artist using artificial intelligence is making a similar claim about authorship through curation. (copyright.gov, wipo.int) Researchers studying human-machine writing have seen that same pattern in practice. Stanford’s CoAuthor project logged 1,445 writing sessions with 63 people using a language model, and the whole point of the dataset was to record not just what the model suggested, but what the human accepted, ignored, and rewrote. (coauthor.stanford.edu, dl.acm.org) That matters because generative systems are high-variance tools. Ask for one image of a red coat in winter and you can get ten different faces, lighting setups, and moods; the creative labor shifts from fabricating every element by hand to steering a flood of possible outputs toward one coherent piece. (link.springer.com, hai.stanford.edu) Artists have been using that language for a while, but it has sharpened as the tools got better at imitation. Fellowship’s exhibitions in 2023 were already presenting artificial intelligence work as an artistic medium shaped by named artists, and by 2025 the copyright debate had moved from “is this art” to “which human contributions count.” (nftnow.com, copyright.gov) The reason this phrasing is sticking is that it answers two complaints at once. It rejects the fantasy that the model is an autonomous genius, and it also rejects the older software metaphor where the machine is just a neutral brush with no influence on the result. (link.springer.com, observer.com) What the artist is really claiming is narrower and more defensible: authorship survives if the human keeps the final cut. In 2026, with image, music, and video generators now common across creative work, that may be the cleanest line available between using a machine and letting the machine speak for you. (x.com, copyright.gov)