AI Art Generators: The Ethics of MidJourney and DALL-E

I used to think AI art was just a fun toy. Something you open, type a few words into, laugh at the weird hands, and move on.

Then I started seeing full campaigns, book covers, and stock libraries quietly switching over to MidJourney and DALL-E, and the ethical questions stopped being theoretical.

The short answer: AI art generators like MidJourney and DALL-E raise serious ethical issues around copyright, consent, artist compensation, bias, and transparency. They are trained on huge datasets that likely include copyrighted work and styles from artists who never agreed to be included. Right now, the law is behind the tech, so a lot of what they do sits in a gray zone: not clearly illegal, but also not clearly fair. If you use them for hobby projects, the risk is low. If you use them for commercial work, you have to think about consent, disclosure, bias, and how much you are replacing human artists with models that learned from those same artists without clear permission or payment.

Let us unpack that properly, because the surface-level “AI is bad / AI is good” argument does not help anyone.

What AI art generators like MidJourney and DALL-E actually do

When people say “AI art,” they usually mean one of a few tools:

• MidJourney: runs mainly through Discord, famous for visually striking, stylized images.
• DALL-E: from OpenAI, integrated into products like ChatGPT and Bing, known for prompt understanding.
• Stable Diffusion: open-source model that anyone can run, modify, or fine-tune.

Under the hood, they all follow similar ideas. There are differences, but at a high level the process looks like this:

Stage	What happens	Why it matters ethically
Data collection	Millions or billions of images scraped from the internet, with text descriptions.	Raises questions about consent, copyright, and how artist work was gathered.
Training	The model learns patterns between text prompts and pixels.	The “learning” comes from human work that may not be credited or paid.
Generation	You enter a prompt, the model outputs a new image, built from learned patterns and noise.	This new image could resemble real artists, photos, or brands in ways that matter.

AI art models do not store full images like a folder. They store mathematical patterns from millions of images, and that is where the ethical tension starts: patterns still come from real people.

That “patterns from real people” detail is important.

Because the defenders of AI art say: “The model does not copy images; it learns like humans do.”

And critics answer: “Humans are not scraping billions of images and monetizing the patterns at industrial scale.”

Both sides are part right. That is why this topic feels messy.

Where the training data comes from (and why that is controversial)

If you ask MidJourney or DALL-E exactly which images they trained on, you will not get a full list. Companies often say things like “licensed data, public domain data, and data created by human trainers.”

That is technically true, but not very helpful.

Most large image models started from huge web-scale datasets. One of the best known is LAION, a dataset built from links scraped from the public web. That includes:

• Portfolio sites
• Online galleries
• Art communities (earlier versions included things like DeviantArt and ArtStation)
• Photography collections
• News and media images

Many artists never gave permission for their work to be included. They had no idea their art would help train a model that might someday replace them for certain types of jobs.

“Public” does not mean “free to use for anything” in law or in ethics. Public just means “visible.”

From a legal angle, companies argue:

• Training counts as “fair use” or something similar under their local law.
• The model does not store or redistribute the original images as files.

From an ethical angle, artists respond:

• My art is feeding this system.
• I did not consent to it.
• I am not credited or paid.

Both can be true at once: the law might allow something that still feels unfair to the people affected by it.

Consent and opt-out vs opt-in

For a long time, the default posture from AI companies has been: “We will train on it unless you make noise or we are forced to stop.”

Later, under pressure, a few things started to appear:

• Some datasets added opt-out lists for artists.
• Certain platforms blocked scraping or model training in their terms.
• Newer image models from some companies claim to use more licensed material.

But there is a core problem: opt-out means your work has probably already been used. You are just asking to not be part of future training rounds.

Ethically, consent works better as opt-in. Practically, companies have built their models on opt-out or no-choice-at-all.

If you run a business and you use MidJourney or DALL-E heavily, you are benefiting from this history, even if you personally did nothing wrong. That does not automatically mean you must stop. But it does mean you should not pretend the cost is zero.

Are AI-generated images “stolen” or “original”?

This is the question that drives half the arguments on social media. Is an AI-generated image just remixing old content, or is it genuinely new work?

Let me split this into three angles:

• How the models actually build images
• When outputs cross into clear copying
• What “style” means in law vs ethics

How MidJourney and DALL-E build new images

Most of these tools use a diffusion process:

1. Start with random noise.
2. Gradually remove noise guided by the text prompt and what the model learned during training.
3. Output a final image that matches the prompt as closely as it can.

The key point: there is no hidden “copy this picture” list. The model uses probabilities learned from all its training samples.

For example, if you type:

“An orange robot watering a plant, studio lighting, 4K, cinematic”

The system has learned things like:

• What a robot usually looks like.
• What “orange” looks like on objects.
• What “studio lighting” often means: soft shadows, controlled highlights.
• How cinematic compositions tend to frame subjects.

It does not pull from a single robot photo. It pulls from statistical patterns across thousands of robot images, studio portraits, cinematic frames, and so on.

That is where advocates say: “This is like humans learning. Artists also study other art.”

The catch: humans are not trained on billions of images at once and sold as subscription services.

When AI art crosses into direct copying

There are some clear red lines.

Cases that are hard to defend ethically:

• Outputs that match a particular artist’s piece almost 1:1.
• Recreations of copyrighted characters or logos for commercial use without permission.
• Training custom models on one living artist’s portfolio then selling “in their style” packages.

We have already seen examples where Stable Diffusion or other models reproduced near-identical watermarks, logos, or composition from training images. That suggests at least some capacity to echo, not just learn patterns abstractly.

If a system can spit out something that looks like it came from a specific living artist, down to their visual quirks, that is a serious ethical problem, even if lawyers find a way to call it legal.

On the legal side, courts are still figuring out how to treat this. Some early decisions suggest that AI-generated work itself may not be protected by copyright in certain jurisdictions, because there is no human author in the strict sense. At the same time, using AI to infringe on someone else’s copyright is still infringement.

So you end up with a strange situation: you might produce an image that you cannot fully own, but that still violates someone else’s rights.

That is not a great place to be as a business.

Is copying an artist’s style ethical?

Here is where things get subtle.

Law tends to say: style is not protected, specific expression is. So “cubism” is fine. A specific cubist painting is protected.

If you say to DALL-E:

“Portrait in the style of Van Gogh”

Most people accept that, because Van Gogh is long dead and his works are public domain.

Now change the prompt to:

“Fantasy warrior in the style of [living artist name]”

There is a good chance the model will produce something very close to that person’s current work. Maybe so close that a casual fan would think that artist made it.

From a strict legal angle, some lawyers will argue that this is still style, not a direct copy of a single work.

From an ethical angle, many artists feel this is parasitic. Their brand took years to build. Now anyone can approximate it with a sentence.

Ethics is not just “can I get sued?” It is “am I benefiting from someone’s work in a way that feels fair to them?”

This is where I push back on some people who say: “Well, artists copy styles too.” Yes, artists learn from each other. But they also:

• Invest years of practice.
• Carry the cost of their exploration.
• Do not instantly scale to millions of copies at zero extra cost.

When you type a prompt and generate a close imitation in seconds, that changes the power balance.

If you are a company, a safe and more respectful line is:

• Use broad style descriptions (“watercolor”, “impressionist”, “digital painting”) rather than specific living artists.
• Avoid prompts that name current, identifiable artists unless you have their permission.

Labor, jobs, and the value of human artists

I hear two extreme stories:

• “AI will replace all artists; we will not need them anymore.”
• “AI is just another tool; nothing changes at all.”

Both are wrong in different ways.

Real life sits in the middle.

What is already happening in the market

We already see concrete changes:

• Concept art and ideation: studios are using MidJourney for quick visual drafts.
• Stock images: some stock sites now accept AI-generated images; others ban them or limit them.
• Marketing content: small businesses swap illustrators for DALL-E to create blog visuals and social posts.
• Book covers: indie authors generate covers instead of paying designers.

Some of this is neutral or even helpful. For example:

• A writer who could never afford a custom cover can now present something more polished.
• Small teams can test many directions visually before hiring a human artist to refine the best ones.

But there is also a downward pressure on prices and expectations.

If a client thinks “I can get 10 images for a few dollars,” they may question why a designer charges hundreds.

AI art does not remove the need for taste, storytelling, or visual strategy. It does reduce the perceived value of pure execution in some contexts.

Over time, I expect:

• Low-budget, one-off illustration work gets squeezed hardest.
• High-level art direction, branding, and style-defining work remains more resilient.
• Artists who combine AI tools with their own skills may stand out, but that is not comfortable for everyone.

If you run a team, using MidJourney or DALL-E to replace all your junior artists looks attractive short-term. Long-term, you lose the pipeline of people who would learn, grow, and become your senior creatives.

So even in cold business terms, going all-in on AI-only art has side effects.

Compensation: who gets paid for what?

One of the strongest ethical claims from artists is: “My work trained your model; I should share in the upside.”

Right now, most commercial AI art tools follow a simple pattern:

• Users pay for subscriptions or API calls.
• Companies keep the revenue.
• Original artists receive no share and often do not even know they are in the dataset.

From a fairness perspective, that feels off.

There are potential alternatives, at least in theory:

• Training on fully licensed libraries where artists are paid.
• Revenue sharing models tied to source datasets.
• Artist registries where creators can explicitly opt in and get compensated.

Right now, these are mostly ideas or partial experiments, not the norm.

If AI art is built on human art, then long-term legitimacy probably requires a way for those humans to benefit beyond “your work inspired the model for free.”

If you run a brand or product, you might not control how MidJourney or DALL-E handle this at a global level, but you can:

• Favor tools that are transparent about training data and licensing.
• Pay human artists for higher-impact pieces, even if you use AI for rough drafts.
• Be honest internally that AI art is lowering your costs partly because artists did unpaid training labor at some point.

Bias, stereotyping, and harmful outputs

There is another ethical layer that gets less attention in the “AI vs artists” talk: bias.

Remember the training data? It is a mirror of the internet. And the internet has plenty of bias.

If you ask DALL-E or MidJourney for:

“CEO giving a presentation”

You are more likely to see:

• A man
• Often white or from certain ethnic groups, depending on defaults
• In a Western-style corporate setting

Ask for:

“Nurse taking care of a patient”

You may see:

• A woman
• Again, often fitting common stereotypes from photography and stock images

AI models do not invent social stereotypes. They amplify what exists in their data, which often came from biased hiring, media coverage, and visual culture.

This matters because images shape expectations. If your product images, campaign visuals, or illustrations always repeat the same unbalanced picture, you reinforce that pattern.

Some tools now add:

• Filters for explicit content.
• Guidelines to avoid certain offensive outputs.
• Optional settings for more diverse outputs.

But these controls are often partial.

As a user, you can:

• Write prompts that invite diversity: mention different ages, genders, body types, and backgrounds.
• Review outputs critically: do these images reinforce clichés?
• Refine or discard sets that feel biased, not just ugly.

Ethics here is less about “Is AI allowed to be biased?” and more about “Am I paying attention to what I am publishing?”

Ownership, licensing, and legal risk when you use AI art

Let us talk about risk, because this is the part that can affect your website, your product, or your brand very directly.

With MidJourney or DALL-E, people often assume:

“I generated it, so I own it. End of story.”

That is not always true.

What the tools usually claim in their terms

You have to read the terms of service (no one likes that, but here it matters). These terms vary, but the patterns often look like this:

• You are granted rights to use what you generate, especially if you pay.
• The company keeps some rights to your prompts and images for research or improvements.
• They give no guarantee that your images do not infringe on existing copyrights.

That last bit is crucial. MidJourney, DALL-E, and similar tools often put the legal responsibility on you. If you generate something that looks too close to a known brand, character, or an artist’s piece, you are the one exposed.

These tools give you the ability to create, but they do not give you a 100 percent clean bill of legal health for every output.

If you are using AI art casually on a personal blog, the real-world risk is relatively low, though not zero. If you are putting images on product packaging, in ads, or inside client deliverables, you should have higher standards.

Some safer habits:

• Avoid prompts that reference protected characters, logos, or exact brands.
• Avoid “in the style of [living artist]” prompts for commercial use.
• Run basic reverse-image searches on critical images to check if an AI result looks too similar to something existing.

Is this perfect? No. But it reduces the chance that you ship something obviously infringing.

Copyright status of AI images

Different countries handle this differently, and the situation is not stable.

Common patterns:

• In some jurisdictions, works created entirely by AI without human creativity may not be eligible for copyright protection.
• Some legal systems focus on whether there is enough human creative input in prompts and editing.
• Courts are only starting to see cases about this; many questions remain open.

Why this matters for you: if your AI-generated image does not qualify for copyright, someone else might be free to reuse it. Your ability to stop them could be limited.

You might be tempted to say “Fine, I do not care, I will just generate more.” That can work for low-stakes uses. For brand-defining visuals, you might want more control and protection. That often pushes companies back toward human-created work, or at least heavy human post-processing on top of AI output.

Privacy, deepfakes, and misuse

Another ethical layer: these tools can be used to manipulate images of real people very convincingly.

Some models, especially ones built for editing or face-swaps, have been used for:

• Non-consensual explicit images
• Political misinformation
• Harassment and defamation

MidJourney and DALL-E have stricter rules on this than many unknown models. They try to block:

• Public figure deepfakes
• Certain forms of explicit content
• Violent or hateful imagery

Still, determined users often find ways around filters. And other tools with weaker safeguards exist.

The same capability that lets you put your friend on a sci-fi poster also lets someone else put a real person into a harmful scenario without consent.

Ethically, there are some clear lines you should not cross:

• Do not generate harmful or explicit content of real identifiable people.
• Do not use AI art to mislead people about events that did not occur.
• Do not hide AI edits in contexts where trust really matters, like news reporting.

That last point connects to transparency.

Transparency: should you tell people an image is AI-generated?

This one is interesting. Many brands and creators are torn: if they say “This is AI-generated,” will audiences value it less?

From an ethical standpoint, there are strong arguments for disclosure, especially when:

• You are in journalism or education.
• You are showing images of real people or events.
• You want to build trust with your audience over time.

For entertainment and fictional content, disclosure is still helpful, but the stakes are a bit lower.

A simple pattern you can follow:

• Add a short note: “Header image created with MidJourney” or “Illustration generated with DALL-E and edited by our team.”
• Internally, document which assets came from where, so you have a record.

Transparency is not about shaming yourself for using AI. It is about giving people enough context to interpret what they see.

From a marketing angle, honest disclosure can actually become a positive. It shows you are aware of the tools, aware of the trade-offs, and not trying to impress people with something you did not actually hand-draw.

Practical guidelines if you want to use MidJourney or DALL-E ethically

Let me be direct here. If you use AI art tools like these, you are already inside an ethical gray area created by the way the models were trained. You cannot completely fix that alone. But you can make better or worse choices on top of it.

Here is a simple mental checklist.

1. Respect living artists

• Avoid prompts that directly name living artists without their consent, especially for commercial projects.
• Do not market AI art as if a specific artist created it, unless they actually did.
• If an AI output looks very close to a real person’s work, do not use it. Try a different direction.

2. Use AI where it replaces stock, not people you hired

If you have a long-term relationship with human designers or illustrators, dropping them overnight for AI art sends a message.

A more balanced approach:

• Use AI for moodboards, early concepts, and low-stakes internal visuals.
• Use human artists for brand-defining work, unique styles, and final polish.
• Consider combining both: AI outputs as raw material, refined and reworked by a human creative.

The easiest ethical win is when AI replaces generic stock shots, not the specific people you are telling stories with.

3. Be cautious with commercial and client work

For anything tied to real revenue, contracts, or clients:

• Check the terms of the tool you are using about rights and restrictions.
• Do not include protected logos, brands, or characters unless you have explicit rights.
• Keep a record of prompts and outputs in case questions appear later.

If a client assumes every image you delivered is fully original and human-made, it may be fair to clarify your process upfront, especially if that client cares a lot about originality or artist support.

4. Pay attention to bias in what you publish

When generating people or roles:

• Vary genders, ethnicities, ages, and body types.
• Review results through a diversity lens, not just a “does this look cool” lens.
• Delete or fix sets that lean too heavily on stereotypes.

Ethics here is less about the hidden model and more about your own curation choices.

5. Be transparent when it actually matters

At minimum:

• Label AI-generated or AI-edited images in articles where accuracy is important.
• Use clear captions or alt text that mention AI when the origin might affect trust.
• In any research, reporting, or educational role, disclose AI generation as standard practice.

For casual blog posts or internal slides, you can still be transparent, but you might adjust the level of formality.

What MidJourney, DALL-E, and others could do better

You and I do not control how these platforms work, but it is still worth naming what ethical improvements from their side might look like.

Some meaningful steps companies could take:

• Publish clearer explanations of training data sources and licensing models.
• Offer true opt-in datasets for artists who want to participate and earn money.
• Provide better tools for artists to check if their work was included and to request removal.
• Strengthen guardrails around style mimicry of living artists.
• Invest more in bias measurement and mitigation for generated people and scenarios.

Right now, much of the burden falls on individual artists and small users, while the biggest gains flow to large companies. That imbalance is the core ethical tension in AI art.

There is no single switch that fixes everything. But there are many small levers that can reduce harm and increase fairness.

Living with the ethical gray zone

I am not going to tell you “never touch MidJourney or DALL-E.” That is not how real people or real businesses work. The tools exist. Competitors use them. They save time and money.

At the same time, pretending there is no ethical cost is not honest.

So where does that leave you?

If I had to condense the practical stance:

• Accept that current AI art tools sit on ethically shaky ground, especially around consent and compensation.
• Use them thoughtfully, not as an excuse to ignore human artists or lower standards.
• Draw clear personal and company lines: no deepfakes, no style theft, no deceptive use in sensitive contexts.
• Pay human artists where the work matters, even if AI could give you “good enough” quickly.
• Stay informed, because the law and the norms around AI art will change over the next few years.

You do not have to solve the whole ethics of AI art alone. But you are responsible for how you choose to use these tools, where you save money, and whose work you replace or support along the way.