Free tools Windows power users keep installed
One-click scans. No signup required.
ThatPainter is reader-supported. When you buy through links on our site, we may earn an affiliate commission. Learn More
The strongest case against AI art is not that machines “have no soul,” or that every generated image is a literal copy. It is that much commercial image AI has been built through an opaque, extractive pipeline: creative work is collected and copied at industrial scale, often without individual consent or payment, then used to produce cheap substitutes for the artists whose work helped make the systems effective.
That argument does not make every use of generative software equally objectionable. AI-assisted retouching, a model trained on an artist’s own work, a licensed-data tool, and a fully generated image sold as a human-made illustration are materially different cases. The useful question is not simply whether AI was involved. It is whether the system and its use respect consent, compensation, authorship, provenance, and the people affected by the result.
What people mean by “AI art”
“AI art” can describe several different practices:
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCrashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minute#1 Best Overall
- text-to-image systems such as Midjourney, Adobe Firefly, Stable Diffusion-based tools, and similar services;
- image-to-image generation, inpainting, outpainting, background replacement, and generative fill;
- AI used as a limited assistive tool inside a predominantly human-made painting or illustration;
- fully generated images presented as if they were drawn, painted, photographed, or otherwise created by a person;
- systems trained on licensed, public-domain, user-uploaded, scraped, or undisclosed material.
The ethical objection changes with the category. A painter who generates rough background ideas and then substantially redraws them is not doing the same thing as a company replacing commissioned illustrators with a model trained on unlicensed portfolios. A responsible criticism must distinguish those cases rather than treating all computational assistance as morally identical.
The central objection: creative extraction without meaningful consent
Many artists did not receive a practical opportunity to approve, refuse, negotiate, or receive compensation for their work’s inclusion in training datasets. An image being publicly viewable is not the same as its creator consenting to commercial model training.
Training a large image model may involve downloading or storing enormous quantities of visual material. Dataset documentation often does not provide an itemized account of whose work was included, and opt-out systems—where they exist—may be difficult to discover, prospective only, or incapable of removing material from a model that has already been trained.
The U.S. Copyright Office’s report on generative-AI training identifies consent, compensation, licensing feasibility, and liability as central unresolved issues. It describes the tension between technological innovation and the possibility that unlicensed training could damage the creative ecosystem.
Recommended Free Tools
This is an ethical and economic argument before it is a legal conclusion. Lack of consent does not automatically prove copyright infringement. Copyright law, fair use, contracts, dataset copying, model behavior, and output similarity are separate questions. But a business model can still be objectionable even when a court has not found it unlawful.
Why “it learns like a human artist” is not enough
Supporters often compare model training with the way artists study existing work. Human creativity is indeed shaped by influence. Artists absorb visual traditions, observe other artists, and build on earlier techniques.
But the analogy leaves out the features critics consider decisive:
- A person encounters and studies work at human scale. A commercial model can ingest billions of files through industrial infrastructure.
- A human artist cannot normally reproduce the market output of thousands of competitors instantly and cheaply.
- Model developers can capture the commercial value of collective cultural production while individual contributors have little bargaining power.
- The resulting system may compete directly with the people whose labor made its capabilities possible.
The distinction is not between “influence” and “no influence.” Influence is everywhere in human creativity. The distinctive concern is industrialized extraction combined with automated substitution.
That labor-and-extraction argument is developed in the AI Art is Theft: Labour, Extraction, and Exploitation paper. Its importance is not that it settles every legal dispute, but that it explains why scale, ownership, and competitive effects matter alongside the question of whether a model reproduces a particular image.
The legal case is serious—but not simple
Arguments about AI art often fail because they collapse several legal theories into one slogan. At least four questions should be separated.
1. Copying while creating a dataset
Training may require making copies of source works. Whether those copies are lawful depends on jurisdiction, licensing terms, fair-use analysis, contracts, and the facts of the particular system. “The model learned from it” does not by itself answer the legal question.
2. Memorization and near-duplicate outputs
Some systems may reproduce or closely approximate particular training examples, especially when an image is unusual or a prompt points toward a specific work. A close reproduction is a stronger infringement argument than the mere ability to generate a generic image in a broad genre.
3. Style imitation
An artist’s general style is often difficult to protect under copyright because copyright generally protects expression rather than abstract styles, methods, or techniques. That does not make style imitation harmless or legally irrelevant. It may raise questions about unfair competition, false endorsement, consumer confusion, moral rights in some jurisdictions, reputational harm, market dilution, or platform rules.
“Style theft” should therefore not be presented as automatically equivalent to copyright infringement. A company using a living artist’s name to market a recognizable imitation may create a different problem from a painter making work influenced by a broad artistic tradition.
4. Outputs and derivative works
An image that resembles a category of art is not automatically an infringing derivative work. Courts and commentators have considered whether particular outputs are substantially similar to specific works, rather than merely similar in mood or genre. The legal analysis of the Andersen litigation discusses training-data copying, inducement, substantial similarity, and the difficulty of connecting a general output to an identifiable source.
Artists have sued companies including Stability AI, Midjourney, Runway, and DeviantArt over alleged training and related conduct. Litigation is evolving, so individual claims and rulings should not be treated as a final answer for every model or output.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →The copyrightability paradox
There is an important tension in the commercial pitch for generative images. Providers may present their systems as powerful enough to replace substantial portions of human creative work. Yet the customer may not receive the same copyright protection that a human artist would have supplied.
On January 29, 2025, the U.S. Copyright Office stated that AI-assisted outputs may qualify for copyright protection when a human determines sufficient expressive elements. Merely entering prompts is not enough. Human selection, arrangement, modification, or incorporation into a larger human-authored work may qualify, depending on the facts; machine-determined expression alone does not.
The accurate conclusion is not that “AI art cannot be copyrighted.” It is that copyright protection depends on the degree and nature of human authorship. A buyer may obtain an image that is inexpensive to generate but harder to protect, enforce, or distinguish from competing outputs.
This also shows why copyrightability does not answer whether training was ethical or lawful. The rights in an output and the treatment of the works used to build a model are related but separate issues.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
The labor case: substitution, wage pressure, and lost pathways
The claim that AI will eliminate every creative job is too broad. The stronger economic argument concerns bargaining power and career structure.
Generative images can affect entry-level illustration, concept art, stock imagery, advertising variations, production work, book and game assets, and routine commissioned sketches. The consequences may include:
Rank #3
- fewer paid opportunities for junior artists to build portfolios;
- downward pressure on rates and shorter deadlines;
- less demand for routine variations and production tasks;
- pressure on artists to adopt systems they do not support;
- deskilling or transformation of artists into reviewers and quality-control workers;
- greater concentration of creative infrastructure in a small number of technology companies.
The harm may therefore be job substitution in particular tasks, wage suppression, and the loss of apprenticeship routes—not the total disappearance of artistic labor. A profession can remain alive while becoming less accessible, less fairly paid, and less able to support newcomers.
This is also why “artists can simply use AI too” is not a complete answer. Adoption may help an individual compete in the short term while normalizing a market in which fewer people are paid to develop the underlying creative skills.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Market flooding and the value of attention
Generative systems make it possible to produce vast quantities of visually competent material: marketplace thumbnails, social posts, advertising variants, book covers, game assets, concept pitches, and low-cost stock imagery.
More images are not inherently bad. The concern is what happens when supply expands faster than audiences can evaluate it. Human artists may find discovery harder, commissioned work may lose scarcity value, and marketplaces may fill with content that has little accountability for its source or process. Search systems and audiences may reward volume and algorithmic visibility rather than craft.
This is a market-structure argument, not a claim that every generated image is aesthetically inferior. A technically polished image can still contribute to a system that makes original work harder to find and harder to sell.
Authenticity, disclosure, and deception
Private experimentation is different from deceptive presentation. The central question is often not “Was AI involved?” but “Was the audience entitled to know?”
Potentially objectionable uses include:
- submitting generated work to a competition requiring human authorship;
- selling an image while implying it was hand-painted or individually illustrated;
- using a living artist’s name to obtain a recognizable style;
- publishing a synthetic editorial image without disclosure;
- presenting generated images as documentary evidence;
- using synthetic portraits, signatures, or artwork to impersonate a creator;
- selling AI-generated assets to an audience that reasonably expects human craftsmanship.
Disclosure does not cure unauthorized training or labor displacement. It does, however, prevent one additional harm: misleading people about authorship, provenance, labor, and evidence.
Authorship is part of the value of art
The objection to AI art is not merely romanticism about tools. In many forms of painting and illustration, the process matters. A viewer may value the decisions, limitations, revisions, physical labor, lived experience, and accountability behind the image.
A human-made work has an identifiable relationship between creator and audience. There may be a story about why the subject was chosen, how the materials were handled, what was changed, and who stands behind the result. A generated image can be visually successful without providing that same provenance.
That does not mean a machine-generated image has no aesthetic value. It means its value should not be described as identical to the value of a commissioned painting, a hand-built illustration, or a work whose process is itself part of the artistic proposition.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsCultural homogenization and visual sameness
Critics also worry that image models encourage familiar visual patterns: fashionable color grading, cinematic compositions, popular fantasy and anime conventions, polished editorial surfaces, and prompt-pleasing arrangements.
Optimization for engagement and recognizable results may favor what is already familiar. Local, idiosyncratic, technically difficult, or culturally specific features can be smoothed away. The result may be a visual culture that is abundant but increasingly repetitive.
This is a tendency and a risk, not a universal technical fact. Some artists use generative systems deliberately to produce unusual combinations. The question is whether commercial incentives reward experimentation—or mostly reward images that resemble the dominant visual language already represented in the data.
Bias and representation
Image generators can reproduce or amplify stereotypes involving race, gender, beauty, occupations, sexuality, disability, and culture. They may default toward whiteness or Western visual conventions, sexualize subjects, misrepresent non-Western settings, or distort historical figures.
Recommended Free Tools
Different tools have different data, safeguards, and failure patterns. A responsible evaluation should test concrete prompts rather than generalize from one system. Useful comparisons include:
- occupations and leadership roles;
- family scenes and professional settings;
- medical contexts;
- children and adults;
- disability representation;
- historical figures;
- non-Western places, clothing, and traditions.
The objection is not that human art is free from bias. It is that automated systems can reproduce bias at enormous scale while presenting the result as neutral or objective.
Privacy, likeness, and abuse
The case against AI art extends beyond professional artists. Synthetic imagery can create nonconsensual sexual images, fake portraits, misleading political or news images, commercial likenesses, and impersonations involving people who never agreed to participate in a visual generation system.
The Copyright Office’s earlier AI report on digital replicas addressed unauthorized realistic digital replicas and recommended a federal law concerning them. Copyright alone cannot address every harm caused by a fabricated likeness, especially when the person depicted never owned the relevant copyright or never intended to be represented at all.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
The environmental argument is about scale
Image generation requires data centers, electricity, cooling, hardware, and repeated computation. The total impact depends on model size, hardware efficiency, resolution, number of discarded generations, the energy mix, cooling systems, and usage volume.
For that reason, universal claims such as “one image uses this exact amount of water” are unreliable without a specific lifecycle study. The stronger argument is about scale and rebound. When an image becomes cheap and instant, users may generate hundreds or thousands of discarded variations, and companies may produce far more visual material than the human workflow would have supported.
A more efficient model can reduce the impact per generation while still increasing total demand. Environmental criticism should therefore ask what the system replaces, how much additional production it encourages, and who bears the infrastructure cost.
The accountability gap
A traditional commission has relatively clear participants: an artist, a client, a contract, a revision history, and an identifiable party responsible for delivery. An AI-generated image may involve a training dataset, a model provider, hosted software, third-party components, a prompt writer, an editor, a publisher, and an end user.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →When the result contains plagiarism, a trademark, a false likeness, defamation, or a harmful stereotype, responsibility can become unclear. Before using a system commercially, ask:
- Who is responsible if the output is disputed?
- Does the provider offer meaningful indemnification?
- What commercial uses are actually covered?
- Are prompts, source images, and generation records retained?
- Can a disputed output be traced or challenged?
- Does the provider offer a takedown or appeal process?
A commercial-use license may reduce some legal uncertainty, but it is not proof of ethical sourcing. It does not by itself resolve labor displacement, cultural homogenization, bias, or disclosure.
The strongest defenses—and why they do not end the debate
A fair argument should acknowledge the best responses:
- Human artists also learn from existing art.
- AI can improve accessibility and reduce repetitive work.
- Some models use licensed or public-domain material.
- Not every output resembles a particular training image.
- Bans may restrict beneficial assistive uses and experimentation.
- Regulation may be more effective than a total prohibition.
These points matter. They show why “all AI art is illegitimate” is too broad. They do not answer whether a particular model obtained its training material with consent, whether artists were compensated, whether a customer is being misled, or whether a tool’s commercial advantage depends on replacing the people whose work supplied its training signal.
How to evaluate an AI image tool
Before choosing a system, evaluate it against the following questions:
- Dataset provenance: Are the training sources licensed, public domain, user-provided, or undisclosed?
- Consent: Can creators opt out before training, and does the mechanism have meaningful effect?
- Compensation: Is there licensing, revenue sharing, or another creator-payment system?
- Attribution: Does the provider disclose sources and attach useful provenance information?
- Output safeguards: Does it limit near-duplicates, living-artist imitation, trademarks, and personal likenesses?
- Commercial rights: Are publishing and resale rights stated clearly?
- Indemnification: Does the provider offer meaningful protection for commercial customers?
- Disclosure: Can users label AI involvement clearly?
- Human contribution: Is the tool assisting a human-made work or replacing a commission entirely?
- Social effect: Does the use preserve creative labor and bargaining power, or maximize substitution?
Adobe, for example, says that its Firefly models are trained on licensed and public-domain content and that Content Credentials are automatically attached for transparency. Its U.S. plans page has advertised Firefly Standard at US$9.99 per month and Firefly Pro at US$19.99 per month, though prices, plan contents, and promotional terms can change. Those claims address provenance and workflow concerns; they do not make generative substitution ethically neutral.
Other systems require their own scrutiny. Midjourney is known for stylized generation, but readers who require documented creator-consented training data should check its current terms and disclosures. Canva AI is convenient for social and presentation graphics, while image generation through ChatGPT offers broad accessibility; neither convenience nor a commercial plan substitutes for fine-grained dataset and provenance control.
For readers who object to automated substitution itself, human commissioning remains the clearest alternative. A direct commission provides an identifiable creator, agreed scope, revisions, provenance, and direct economic support. It generally costs more and takes longer, but those are part of what the transaction supports.
A narrower position is the stronger position
Society does not need to reject every generative feature to reject an extractive business model. A defensible standard would require, at minimum, consent or a legally and ethically defensible licensing basis; creator compensation; meaningful dataset transparency; effective opt-out and deletion mechanisms; safeguards against close imitation and likeness abuse; honest provenance and disclosure; clear commercial warranties; and protection for human creative labor.
The best case against AI art is therefore a case against unconsented extraction, opaque ownership, automated substitution, and deceptive presentation. It survives the obvious counterarguments because it does not depend on claiming that humans never learn from predecessors, that every output is copied, or that every AI-assisted image is worthless.
The question is who benefits, who bears the risk, and whether the people whose creative work made the system possible had any meaningful say in what happened to it.
Quick Recap
