In 2022, an image created with Midjourney, “Théâtre D’opéra Spatial” by Jason M. Allen, won first prize in the digital art category at the Colorado State Fair. The win sparked fierce debate — and the US Copyright Office later refused to register the work because it contained too much AI-generated material without sufficient human authorship. The episode captured the tensions of AI art: astonishing capability, contested legitimacy.

How image generators work

Text-to-image systems such as Midjourney, DALL·E, Stable Diffusion, Imagen and Flux are mostly diffusion models. They are trained on huge numbers of image–caption pairs to reverse a noising process: starting from random noise, they gradually produce an image that matches a text description. They learn statistical associations between words and visual features — including the styles of artists whose work appears in the training data.

Why many artists object

  • Consent and compensation: training datasets were largely scraped from the web, including artists’ portfolios, without permission or payment.
  • Style imitation: prompts can request images “in the style of” living artists, competing directly with them.
  • Market impact: illustration, concept art and stock imagery markets face cheaper substitutes.
  • Attribution: generated images carry no credit to the artists whose work shaped them.

In 2023, a group of artists sued Stability AI, Midjourney and other companies in the US, alleging copyright infringement; the case has moved slowly through the courts, with some claims allowed to proceed.

Why others embrace it

Many artists and designers use AI as a tool: for brainstorming, exploring compositions, generating textures, or integrating it into mixed-media work. Supporters compare it to photography or digital painting, which were also once dismissed as not “real art”. Accessibility is another argument: people without traditional drawing skills can visualize ideas.

In the US, the Copyright Office’s position is that works must have human authorship. Images generated from prompts alone are not protected; human contributions — selection, arrangement, substantial modification, or human-made elements — can be. Other countries’ rules differ, and some offer protection for computer-generated works.

Tools for protecting artists

Researchers at the University of Chicago developed Glaze, which adds subtle changes to images that make it harder for models to mimic an artist’s style, and Nightshade, which aims to “poison” training on unlicensed images. Many platforms also offer opt-out settings, and some AI companies honor opt-out requests or train only on licensed images.

Licensed and “commercially safe” models

Some companies train image models on licensed or public-domain content — for example stock libraries — and offer indemnities to business customers. These models appeal to brands that want to reduce legal risk.

Where the debate is heading

The likely outcome is a mix of court rulings, licensing deals and regulation (such as the EU’s transparency requirements for training data). In the meantime, both the tools and the arguments continue to evolve quickly.

How to use image generators responsibly

  • Avoid prompting with the names of living artists to imitate their style.
  • Disclose AI use when it matters to your audience, clients or competitions.
  • Respect competition and publication rules, many of which now address AI.
  • Prefer tools trained on licensed content for commercial work.
  • Credit and pay human collaborators whose work you build on.

What to watch next

  • Court rulings in artists’ lawsuits against image generator companies.
  • Licensing markets for training images.
  • Transparency obligations for training data in the EU.
  • Adoption of provenance labels for AI images.

Key takeaways

  • Image generators are mostly diffusion models trained on image–caption pairs.
  • Artists object to training without consent, style imitation and market impact.
  • In the US, images generated from prompts alone are not copyrightable.
  • Glaze and Nightshade aim to protect artists’ work from unauthorized training and mimicry.
  • Licensed training models offer lower legal risk for commercial users.

Frequently asked questions

Is AI art real art?

It is a philosophical question with no single answer. Many people distinguish between prompting alone and works with substantial human creative input.

Can I sell AI-generated images?

Often yes, depending on the tool’s terms, but you may not be able to stop others from copying them if they lack copyright protection.

How do I stop AI from training on my art?

Use platform opt-out settings, robots.txt for your website, tools like Glaze, and check whether AI companies honor opt-out registries.