Over 150 million people use AI image generators monthly, producing around 80 million images a day (Morphed, 2026).
Most of those people believe they own what they generate.
In the United States, they do not.
On 2 March 2026 the Supreme Court declined to hear Thaler v. Perlmutter.
That left the lower ruling standing: AI cannot be an author, and autonomously generated work cannot be copyrighted (Norton Rose Fulbright, 2026).
Writing a detailed prompt does not change that. 🖼️
Here is what it means in practice, and what it does not mean.
🎁 Compare Creative Tool Trials →
🧾 主要发现概览
| 措施 | 数字 | 来源 |
|---|---|---|
| Monthly users of AI image tools | Over 150 million | Morphed (2026) |
| Images generated daily | Around 80 million | Morphed (2026) |
| Organisations using AI somewhere | 88% | SQ Magazine (2026) |
| Large enterprises using generative AI | 82% | SQ Magazine (2026) |
| Assets produced by Adobe Firefly | 24 billion | Axis Intelligence (2026) |
| Market size estimate for 2026 | $4.8bn to $12.4bn | Axis Intelligence (2026) |
| Copyright for AI-only output, US | Not available | Norton Rose Fulbright (2026) |
| Prompting counted as authorship | 不 | Garrett Ham (2026) |
This article explains published rulings and is not legal advice. Rules differ by country and change. Take advice for your own situation.
⚖️ Two Different Questions
Most confusion here comes from merging two separate issues.
⚖️ Owning it and being allowed to use it
Ownership
Copyright protects human creativity. Purely machine-made output has no human author.
So it falls outside protection entirely in the United States.
That means anyone may copy and reuse it without asking you.
Permission to use
This is a contract question, set by whichever tool you used.
Most paid tiers permit commercial use of what you generate.
That permission says nothing about whether you own the result.
Why both matter
You may legally use an image in your advertising and still have no exclusive claim to it.
A competitor could use the identical image without breaching anything.
For most marketing material that is tolerable. For some things it is not.
🚫 Where This Genuinely Bites
The practical consequences are narrow but sharp.
| 使用 | Problem? |
|---|---|
| A social post | No real issue |
| A blog illustration | No real issue |
| An advertisement | 通常很好 |
| Your logo | Serious problem |
| Artwork you sell as a product | Serious problem |
| Illustrations licensed to clients | Serious problem |
| Book or album covers | Consider carefully |
The three flagged rows all depend on exclusivity.
Never use a purely AI-generated image as your logo.
Why logos are the clearest case
A trademark must be distinctive and defensible.
Without copyright in the underlying artwork, that defence is weakened considerably.
我们的 design tool analysis makes the same point about stock elements.
The client work problem
If you sell design services, clients expect to receive rights.
You cannot transfer ownership you never had.
Say so in writing before delivering AI-generated work.
✍️ What Counts as Human Input
AI-assisted work can be protected, and the bar is higher than people assume.
| Contribution | Likely to count? |
|---|---|
| Writing a detailed prompt | 不 |
| Generating many versions and picking one | 不 |
| Substantial editing afterwards | Possibly |
| Combining outputs into a new composition | The arrangement may qualify |
| Using output as one element among many | Likely, for the whole work |
| Your own text alongside generated images | Text is protected |
The Zarya of the Dawn case illustrates the split clearly.
Protection was granted for the text and arrangement, and refused for the generated images (Editors Weblog, 2026).
The prompt misconception
Many people assume a sufficiently elaborate prompt makes them the author.
The published position is that prompting is not enough, however detailed.
Control over expressive elements is what matters, not instruction.
What this suggests practically
If you need to own the result, treat generated output as raw material.
Edit substantially, combine it with your own work, and keep records of what you did.
That record is what would support any later claim.
🏢 Why Training Data Became a Selling Point
One commercial detail explains a lot about enterprise adoption.
Adobe Firefly launched with fully licensed training data.
That removed a specific worry for large organisations (Axis Intelligence, 2026).
| 关注 | Why it matters to a business |
|---|---|
| What the model trained on | Possible infringement exposure |
| Whether the provider indemnifies you | Who pays if there is a claim |
| Output resembling existing work | A practical risk |
| Terms changing later | Check periodically |
Indemnity is the row most worth checking before committing.
Some providers will defend you against claims. Many will not.
The distinction for small businesses
A sole trader faces far less exposure than a large advertiser.
Still, the questions are the same and the answers take minutes to find.
Read the commercial terms of whichever tool you use.
The resemblance risk
Occasionally generated output closely resembles an existing work.
That is a separate problem from ownership and can be a real one.
Check anything you plan to use prominently before publishing it.
📈 How Fast This Became Normal
The adoption figures explain why the legal question matters to so many people.
📈 The scale of everyday use
Why the gap opened
Tools reached hundreds of millions of people within a few years.
Copyright doctrine moves in decades, not years.
So millions of images exist in a category the law had not anticipated.
The concentration nobody mentions
One ecosystem is reported to account for roughly 80% of AI-created imagery.
That means most output shares a small number of underlying models.
Similar prompts across different tools can produce similar results.
What that means for distinctiveness
If everyone uses the same few models, outputs converge in style.
Generated images increasingly look generated, and recognisably so.
That is an aesthetic problem before it is a legal one.
🧭 A Practical Decision Rule
Rather than memorising doctrine, ask one question about each image.
| 问题 | If yes | If no |
|---|---|---|
| Do I need to stop others copying this? | Do not use AI alone | Generated is fine |
That single question resolves nearly every case correctly.
Exclusivity is the only thing you actually lose.
Applying it
A blog illustration does not need exclusivity. Use whatever you like.
A logo needs it absolutely. Commission something original.
Product artwork you sell needs it. So does anything you licence onward.
The middle cases
Book covers, packaging and signage sit in between.
Ask how much the visual identity carries the business.
Where it carries a lot, treat exclusivity as necessary.
Keeping the option open
Generated images can be a starting point that a designer then reworks substantially.
That path may produce something protectable while saving time.
Keep records of what the human contributed at each step.
📋 Questions to Ask Your Tool
Terms vary considerably, and they are checkable in minutes.
| 问题 | 为什么这很重要 |
|---|---|
| Does my tier permit commercial use? | Free tiers often do not |
| What was the model trained on? | Affects infringement exposure |
| Is there an indemnity? | Who defends a claim |
| Can others generate the same image? | Usually yes |
| Do I keep rights if I cancel? | Check before relying on it |
| Are outputs stored or reused? | Privacy consideration |
The free tier question catches people most often.
Many free plans permit personal use only.
The cancellation question
Some terms grant rights only while your subscription is active.
Images already published could become a problem after you leave.
Check that clause before building a campaign around generated work (Red Escuela, 2026).
Where to find the answers
Look for a page titled licensing, terms of use, or commercial rights.
Search the document for the word commercial and read that section fully.
我们的 审判目录 lists tools whose terms are worth comparing.
🌍 Beyond the United States
Every ruling discussed above is a United States decision.
| 地区 | General position |
|---|---|
| 美国 | Human authorship required |
| 英国 | Has a computer-generated works provision |
| European Union | Emphasis on human creative choices |
| Elsewhere | Varies considerably |
This is a rough summary of a moving picture, not a statement of law.
If you trade internationally, do not assume one answer applies everywhere.
Why the divergence matters
An image unprotected in one country may have some protection in another.
That complicates anything sold or licensed across borders.
It is another reason to commission original work for anything foundational.
The direction of travel
Cases continue and legislatures are considering the question actively.
Today’s position may not hold in two years.
Decisions made now should be robust to that changing (Picwand, 2026).
🧾 Disclosure and Who Expects It
Separate from law, there are expectations about saying when work is generated.
🧾 Where disclosure is expected
The product photograph problem
A generated image of a product implies the product looks like that.
If it does not, that is potentially misleading regardless of copyright.
Consumer protection rules apply whether or not an image is generated.
Where disclosure costs nothing
Decorative illustrations on a blog post mislead nobody.
Few readers expect those to be photographs of real events.
Disclosure there is optional and rarely expected.
Client work again
Tell clients before delivering, not after they ask.
Some are entirely comfortable and some are not.
Finding out afterwards damages the relationship more than the method would have.
💷 What These Tools Cost
Pricing varies widely and the free tiers carry the most restrictions.
| 层级 | Typical position |
|---|---|
| 自由的 | Often personal use only |
| Entry paid | Commercial use usually permitted |
| Bundled in design software | Terms follow that product |
| Enterprise | Indemnity sometimes included |
The free tier restriction catches small businesses regularly.
Using free-tier output commercially may breach the terms you accepted.
Bundled generation
Many design tools now include image generation in their subscription.
That is usually the simplest route, because one set of terms covers everything.
Check whether the generation feature carries different terms from the rest.
Why enterprise tiers cost more
Part of the price buys indemnity and licensed training data.
For a sole trader that protection is rarely worth the difference.
For anyone advertising at scale, it may be the main reason to pay.
🛠️ Using Generated Images Sensibly
A short checklist covers most situations without needing legal knowledge.
| 步 | 为什么 |
|---|---|
| Check your tier permits commercial use | The commonest oversight |
| Never use it for a logo | Exclusivity cannot be secured |
| Do not imply it is a photograph | Separate rules apply |
| Look for resemblance to existing work | A real if uncommon risk |
| Tell clients before delivering | Avoids a worse conversation later |
| Keep a note of what you edited | Supports any later claim |
Five of six steps take under a minute each.
The checklist matters more than understanding the doctrine.
Checking for resemblance
Run a reverse image search before using anything prominently.
It occasionally surfaces a close match to an existing work.
That is worth knowing before publication rather than afterwards (Garrett Ham, 2026).
Keeping records
Save your original generation, your edits and the finished file.
If you ever need to show human contribution, that sequence is the evidence.
It costs nothing at the time and cannot be reconstructed later.
👤 The Case for Commissioning Instead
Generated imagery is cheap, and sometimes the alternative is better value.
| Need | Generated | Commissioned |
|---|---|---|
| Blog illustration | Fine, and instant | Overkill |
| Social graphics | Fine | Rarely necessary |
| Logo | 不 | Yes, once |
| Product illustration you sell | 不 | 是的 |
| Photographs of your actual work | Cannot substitute | Take them yourself |
| Brand illustration style | Inconsistent | Worth paying for |
Two rows need commissioning and one needs a camera.
No generated image can show your actual finished work.
The photography point
Customers want to see what you really did, not an idealised version.
Real photographs of real jobs outperform polished generic imagery.
A phone camera and decent light is enough for that.
The consistency problem
Generated images vary in style between sessions.
Building a recognisable look from them is difficult.
A commissioned illustration style can be reused indefinitely (Axis Intelligence, 2026).
🔮 What to Watch Next
Several things could change the picture within a year or two.
| Development | Possible effect |
|---|---|
| Further court decisions | May clarify the human input threshold |
| Legislation on generated works | Could create new protection |
| Cases about training data | Affects which tools are safe |
| Mandatory labelling rules | Disclosure may become required |
| Provider terms changing | Recheck periodically |
The training data cases are the ones with the widest consequences.
An adverse ruling could affect tools people already rely on.
How to stay robust to change
Prefer tools that disclose what they trained on.
Avoid building anything foundational on generated work.
Keep your own photographs and commissioned assets as the backbone (SQ Magazine, 2026).
The reasonable summary
Use these tools freely for material that does not need to be exclusively yours.
Commission or photograph anything that does.
That division survives whichever way the law develops.
🗂️ A Note on Text and Code
The same principle extends beyond images, and the answer is similar.
| Output type | Ownership position |
|---|---|
| Purely generated images | Not protected |
| Purely generated text | Same principle applies |
| Text you substantially rewrote | Your contribution may qualify |
| Arrangement of generated pieces | Arrangement may qualify |
Human creative contribution is the deciding factor throughout.
The medium changes, the underlying question does not.
What this means for written content
Marketing copy rarely needs copyright protection to do its job.
Nobody is going to copy your service page and profit from it.
So the practical concern is much smaller than it is for a logo.
Where it does matter for text
Books, courses and anything you licence to others.
Those depend on being able to stop copying.
Treat them the way this article treats logos and product artwork.
One thing that has not changed
Using a machine to help make something has never disqualified the result.
Cameras, synthesisers and design software all raised similar questions once.
What matters is whether a person made the creative choices.
That principle is old, and the current rulings apply it rather than inventing it.
🚫 这些数据无法告诉你什么
This is United States law. Other countries differ, some considerably.
The position is still developing. Cases continue and rules may shift.
Adoption figures come from vendors. Read the enthusiasm accordingly.
Market estimates vary threefold. $4.8bn to $12.4bn depends on definitions.
Nothing here is advice for your situation. Take proper advice for anything significant.
The jurisdiction point
Some countries take a different view of computer-generated work.
If you trade internationally, the answer may not be uniform.
Do not assume one country’s rule applies everywhere.
The definitional problem in the market data
Estimates range from $4.8 billion to $12.4 billion for the same year.
The gap depends on whether features embedded in other software are counted.
Any single figure quoted without that context is close to meaningless.
🏁 简短版
150 million people generate 80 million AI images daily, and most assume they own them.
In the United States, purely AI-generated images cannot be copyrighted.
The Supreme Court declined to revisit this in March 2026.
Prompting is not authorship, however detailed the prompt.
Being allowed to use an image and owning it are separate questions with separate answers.
For social posts and blog illustrations, none of this causes a problem.
For logos, products and client work, it matters enormously. 🖼️
🚀 Browse Creative Trials by Category →
❓ 常见问题
Can I copyright an AI-generated image?
Not in the United States, if it was produced without meaningful human creative input. The Copyright Office position is clear.
Does a very detailed prompt count as authorship?
No. Published guidance states that entering prompts, however detailed, is not sufficient.
What happened in March 2026?
The Supreme Court declined to hear Thaler v. Perlmutter, leaving in place the ruling that AI cannot be an author.
Can I still use AI images commercially?
Usually yes. That depends on your tool’s terms, and it is a separate question from ownership.
What is the practical risk?
Anyone may copy an unprotected image. For marketing that rarely matters. For a logo it matters greatly.
Can any AI-assisted work be protected?
Yes, where a human exercises meaningful creative control. Text and arrangement were protected in the Zarya case.
Should I use AI images for my logo?
No. A trademark needs to be distinctive and defensible, and unprotected artwork weakens that position.
What about selling AI artwork as a product?
Consider carefully. You cannot transfer ownership you do not hold, and buyers often assume they receive it.
Why does training data matter?
Some providers licensed everything they trained on, which reduces infringement exposure for commercial users.
What is indemnity and should I care?
It means the provider defends you against claims. Some offer it, many do not. Check before relying on the tool.
Do these rules apply outside the United States?
Not necessarily. Several countries treat computer-generated work differently, so check your own jurisdiction.
📚 参考资料
Norton Rose Fulbright. (2026). AI in litigation series: an update on AI copyright cases in 20262026年8月8日检索自 https://www.nortonrosefulbright.com/en/knowledge/publications/ce8eaa5f/
Editors Weblog. (2026). Can you copyright AI-generated art? What the law says in 20262026年8月8日检索自 https://editorsweblog.org/2026/06/30/can-you-copyright-ai-generated-art-2026
Garrett Ham. (2026). Can AI-generated works be copyrighted? 2026 legal guide2026年8月8日检索自 https://garrettham.com/ai-generated-works-copyright/
Red Escuela. (2026). AI and copyright: what you can and cannot do with AI-generated content2026年8月8日检索自 https://redescuela.org/ai-guides/ai-copyright/
Axis Intelligence. (2026). AI image generation statistics 2026: market size, platform data and industry adoption2026年8月8日检索自 https://axis-intelligence.com/ai-image-generation-statistics/
Morphed. (2026). AI image generation statistics 2026: market, users and adoption2026年8月8日检索自 https://morphed.app/stats/ai-image-generation-statistics
SQ Magazine. (2026). AI image generation statistics 2026: market size, adoption and risks2026年8月8日检索自 https://sqmagazine.co.uk/ai-image-generation-statistics/
本站相关阅读
我们的 design tool analysis covers stock image licensing and the same logo warning. See also our video marketing research，以及 免费试用会员目录 for testing creative tools free.
关于本次分析
This article separates two questions that are routinely merged: whether AI output can be owned, and whether it may be used commercially. The first is settled by copyright law and the second by the tool’s contract, and they have different answers. Rulings cited are United States decisions and are identified as such. Nothing here is legal advice, and anyone relying on generated imagery for a logo, product or client deliverable should take proper advice. Checked on August 8, 2026.
-
AI Image Generation: Adoption and Legal Reality - 10 月 30, 2026 -
Every Tube Magic Feature, and Which You Will Actually Use - 10 月 30, 2026 -
Design Tools: Who Uses What, and What It Costs - 10 月 27, 2026