Why original art stays tied to a person
Image generators make finished-looking pictures in seconds, which is why people keep asking whether AI will replace artists. The fine art job is wider than image output. A painter, sculptor or illustrator decides what to make, agrees on a brief and a budget with a client or dealer, then makes the object by hand in paint, clay, stone, ink or mixed media. The buyer is often buying the person: a signature, a body of work, a story about how the piece came to exist.
Most of the day is physical. Stretching and priming a canvas, mixing color, carving, firing, framing, mounting and installing a finished piece in a room are not prompt problems. The robotics panel on this page places that kind of work at the mobile-robot tier, which is hardware that is still expensive and rare outside logistics and manufacturing.
The commercial edge is where pressure shows. Decorative images, stock-style illustration and quick concept work can be generated cheaply, and the cost table above shows how far tool spend sits below paying a person for the same hours. That squeezes paid commissions at the low end first. The Bureau of Labor Statistics projects employment for this group to fall 2.9% between 2025 and 2035, from a base of about 11,220 US jobs, with median pay of $55,490 (BLS, 2025 projections).
What AI makes, what it assists, and what it leaves to the artist
In the group where AI can do the work on its own, the page puts 3% of task time. That covers generating reference images and researching subject matter, plus producing rough designs and sketches for a client to react to before anything is committed to canvas or clay.
The assist group accounts for 30% of task time. Here the tools speed up steps the artist still owns: running color and composition variations on an agreed idea, and preparing files for print, submission or a gallery proposal. Writing up a commission quote or a grant application sits in the same place.
Work that still needs a person is 67% of task time. That includes applying the medium by hand, installing and framing finished pieces, meeting clients, dealers and art directors to settle subject, budget and deadline, and teaching classes or workshops. Our coverage score, the share of task time AI can handle today, prints as 23 out of 100; the coverage method page explains how that is built.
What has actually been tested
Not much, for this job specifically. The evidence grade for quality parity prints as D, and a D means no direct, measured test of AI against working fine artists exists yet. So this page gives no parity number, and you should treat any site that quotes one for painters or sculptors with care.
A real test would look like this: a set of live commissions with briefs, revisions and deadlines, delivered by qualified artists and by AI-led workflows, then judged blind by the people who normally buy, commission or exhibit the work. Tracking what working artists actually earn from commissions over several years would settle it faster than any lab benchmark. Until something like that is published, the honest read is task erosion at the commercial end rather than the end of the occupation. How quality parity is graded sets out the bar.
When the picture could change
Most likely after 2037 (8 in 10 of our scenarios). Two things could pull that forward. First, continued price pressure on paid illustration and concept work, where clients accept a generated image for jobs that used to go to a junior artist. Second, better editing control, so a buyer can get consistent characters, style and revisions without hiring anyone.
Two things hold it back. The physical side of making and installing work has no cheap machine behind it, and the hardware tier involved is not close to studio-ready. And a large share of the value here is authorship: provenance, exhibition history, licensing and the right to say who made the thing. Copyright disputes over training data push buyers of high-value work toward documented human authorship, not away from it. For how the window is modeled, see the replacement-year method.
How to stay needed as a painter, sculptor or illustrator
Lean into the parts of the job the task list above keeps with people. Make work physically, in materials that read differently in person. Handle the client and gallery relationship yourself, including the brief, the budget and the installation. Teach: workshops and classes pay, and they build the audience that buys later.
Two skills are worth adding. One is documentation and provenance, so every piece has a clean record of how it was made, which matters more as generated images spread. The other is working with the tools openly for drafting and variations, so you keep the speed without handing over authorship. Our guide to AI skills employers want covers the practical end of that.
What to do: price and pitch the commissioned, physical and taught parts of your practice first, and treat generated images as a sketching step rather than a product.
Nearby jobs move differently. Compare this page with Craft Artists, Art Directors and Special Effects Artists and Animators, or put any two side by side on the compare tool. The wider picture sits on the art and design workers family page and the arts and entertainment sector page, and the list of jobs that mostly need a person shows where this one falls among the rest.
This job’s headline Still needs a human score prints above as 75 out of 100 (higher is safer). Every figure here comes from open data and published studies; the method shows the sources and the math.