Midjourney
processed over 15 million image generation requests per day in early 2025,
making it one of the most widely used creative AI tools in the world and one
of the most contested. The platform, which generates images from text
descriptions using a diffusion model trained on billions of images scraped
from the internet, has become both a symbol of AI’s creative potential and
the focal point of a copyright dispute that involves thousands of artists,
major visual media companies, and courts in multiple jurisdictions. The
central legal and ethical question is whether the images used to train
Midjourney and similar tools were used with the consent and appropriate
compensation of their creators, and whether the outputs of these models
constitute infringement of the creative rights of the artists whose work was
in the training data. On both questions, the current legal position is
disputed, evolving, and consequential for the future of both AI and creative
industries.
The scale of the training data issue is important context.
Midjourney, Stability AI’s Stable Diffusion, and other image generation
models were trained on datasets including LAION-5B, which contains
approximately 5.85 billion image-text pairs scraped from the internet without
licence or consent from image creators. The creators of these images, from
professional illustrators and photographers to graphic designers and fine art
practitioners, had no knowledge their work was included, no opportunity to
opt out, and no compensation. When they subsequently discover that AI tools
can generate images “in the style of” their work with considerable
accuracy, they understandably feel that something significant has been taken
from them without their knowledge or agreement.
The Legal Landscape
The copyright litigation against AI image generation companies is
the most advanced in the US, where several cases are working through the
federal court system. A class action lawsuit filed by artists including Kelly
McKernan, Karla Ortiz, and Sarah Andersen against Stability AI, Midjourney,
and DeviantArt alleged copyright infringement through the training process
and through the generation of images that are substantially similar to
specific copyrighted works. Getty Images filed a separate case against
Stability AI specifically, alleging that its images were used without licence
and that Stable Diffusion outputs sometimes include visible Getty watermark
artefacts, suggesting direct reproduction rather than mere stylistic
influence.
The legal arguments turn on whether training constitutes
infringement and whether AI outputs are infringing copies. The Getty
watermark evidence is particularly significant for the output infringement
question, because visible watermarks in AI-generated images are difficult to
explain without some form of reproduction of training data occurring in the
model. The US Copyright
Office has published guidance stating that AI-generated images
without human creative selection and arrangement are not eligible for
copyright protection, a position with significant implications for the
commercial value of AI-generated content that companies like Midjourney are
commercialising.
The Style Question and Its Limits
Copyright law does not protect artistic style; it protects
specific original expression. An AI model that generates images in the style
of a specific living artist is not necessarily infringing copyright under
current law, even if the artist finds this deeply objectionable on economic
and moral grounds. The economic harm to artists whose distinctive style can
be replicated on demand, reducing the value of their original work and their
paid commissions, is real but may not be legally cognisable under the current
copyright framework. This gap between legal protection and economic harm has
led artist advocacy organisations including the Association of
Illustrators to call for new rights frameworks that go beyond
copyright, including a specific right of artistic personality that protects
distinctive style from commercial AI replication.
Industry Responses
Responses from AI image generation companies to the copyright
controversy have ranged from defensive legal argument to commercial
accommodation. Adobe’s Firefly, trained exclusively on licensed content,
represents a model of licensed data procurement that avoids the consent
problem at the cost of a smaller and more expensive training dataset. Getty
Images has launched its own AI image generation tool trained exclusively on
its licensed collection, with automatic compensation for contributing
photographers when their images are used in training. Shutterstock has signed
a commercial data licensing agreement with OpenAI and introduced a
contributor compensation fund. These alternatives to scraping demonstrate
that licensed AI training data procurement is commercially viable, which
strengthens the argument that the scraping approach was a choice to
prioritise cost over creator rights rather than a technical
necessity.
What This Means for You
If you are a visual artist, illustrator, photographer, or graphic
designer, your work has almost certainly been included in AI image generation
training datasets without your consent. The legal remedies currently
available are limited and contested; the legislative remedies that would
provide more comprehensive protection are being advocated for by creator
organisations but have not yet been implemented in the UK or US. Registering
your copyright, opting out of AI training datasets through services like Have
I Been Trained, and supporting the legal and legislative advocacy work of
artist organisations like the Association of Illustrators and the Design and
Artists Copyright Society are the most direct actions available. The
international dimension of the AI art copyright crisis varies significantly
by jurisdiction in ways that affect what creators can do to protect their
interests. In the EU, the opt-out mechanism under the Copyright in the
Digital Single Market Directive provides a legal basis for creators to
require that their work not be used for commercial AI training, and several
major image rights organisations have filed collective opt-outs on behalf of
their members. In the UK, the debate about whether to introduce a commercial
text and data mining exception is ongoing, and creator organisations are
actively lobbying for an opt-out mechanism equivalent to the EU provision. In
the US, the litigation-based approach means creator rights depend on courts
rather than specific legislative protections, creating uncertainty that will
persist until significant cases are resolved. Creators in different
jurisdictions therefore have different tools available to protect their
interests, and engaging with the advocacy organisations active in your
jurisdiction is the most direct way to ensure your interests are represented
in the policy processes currently determining the outcome. For related
analysis, see our coverage of AI
training data legal battles and AI
and creative activism.
The moral rights dimension of the AI
art copyright debate deserves acknowledgment alongside the economic copyright
questions. In many European jurisdictions, artists have moral rights
including the right of integrity, which protects against modifications to
their work that harm their honour or reputation, and the right of
attribution. These rights cannot be waived by contract in the way that
economic copyright can be licensed. Whether training AI models on artists’
work without attribution constitutes a violation of moral rights is a legal
question that has not yet been definitively addressed in UK or European
courts, but it is one that artist advocates are raising with increasing
specificity. The Design and
Artists Copyright Society represents UK visual artists in these
discussions.
The practical toolkit for artists
seeking to protect their work includes not only legal and advocacy routes but
technical ones. Services including Glaze and Nightshade, developed by
researchers at the University of Chicago, allow artists to apply
imperceptible modifications to images before uploading them online that
disrupt AI style learning without affecting the human-visible appearance of
the work. These tools represent a technical response to a governance failure,
and their adoption by artists who cannot afford litigation provides a
meaningful layer of self-protection while the legal and regulatory framework
develops.
About the Author
Stuart Kerr is a technology correspondent at LiveAIWire, covering
artificial intelligence, digital innovation, and the social impact of
emerging technologies. Follow LiveAIWire for daily analysis at liveaiwire.com.