Small Business Using AI Art: Licensing, Disclosure and Safe Uses

Yes, a small business can usually use AI-generated art in its marketing, provided the tool's current terms allow commercial use on your plan, the image copies no one else's work, and you do not mislead customers with it. The catch is ownership: a purely machine-made image may have no copyright protection at all.

That catch changes how you should use the images. A tool's terms can say "you own the output" and still leave you unable to stop a competitor reusing it, because the terms can only hand over rights that exist. So the practical question is not "may I?" but "which jobs are safe for something I may not be able to protect?" Short-lived decoration usually is; a logo, a mascot or a packaging print usually is not, and a two-minute log per image covers you for the rest.

Follow me on Instagram@sagnikteaches

Three separate questions, three different answers

Most confusion comes from mixing up three things that sound alike. Keep them apart and the rest is simple.

Connect on LinkedInSagnik Bhattacharya

• Permission to use. What the tool's terms let you do with the output, for example in an advert, on packaging or on a client's site. This is a contract between you and the vendor.
• Protection. Whether copyright law gives anyone, including you, exclusive rights in the image. This is law, and a contract cannot create it.
• Exposure. Whether the image copies or closely resembles someone else's work, face, brand or character, so that you could be on the receiving end of a complaint.

A tool can give you full permission while the image has no protection and still carries exposure. Each of the three needs its own check, and each is covered below.

Subscribe on YouTube@sagnikteaches

Check one: what the tool's terms actually say about your plan

Read the terms for the plan you use today, not a review from last year. Three things vary between vendors and plans, and they are the three that catch small businesses out.

Who owns the output, and with what limits

OpenAI's terms of use say that, as between you and OpenAI and to the extent the law allows, you own the output, and that OpenAI assigns you whatever rights it has in it "if any". The same terms warn that output may not be unique and that other users may receive similar results. Midjourney's terms say you own the assets you create, with exceptions, including that a company or employee of a company with more than $1,000,000 a year in revenue must be on a Pro or Mega plan to own them. Midjourney also states that your content is publicly viewable and remixable by default, which matters for unreleased work, and that ownership survives if you later downgrade or cancel.

Notice the pattern. Every one of these phrases is hedged by "applicable law" or "if any". The vendor is passing you what it has. It is not promising the law will recognise it.

Whether the vendor stands behind you if there is a claim

Some vendors offer a legal indemnity, meaning they will cover certain claims against you if the output infringes someone's rights. Adobe's Firefly plans page says outputs from its own Firefly models are safe for commercial use because they are trained on licensed and public-domain content, but it leaves you responsible for judging outputs from the partner models offered in the same app, and it offers IP indemnification only to eligible enterprise customers for select Firefly outputs. OpenAI's output indemnity in its service terms covers API and Enterprise-type customers, not ChatGPT Business. If you are on an ordinary consumer or small-team plan, assume you carry the risk yourself.

Where the image was made inside a larger product

Rules can also differ between apps from the same vendor. Adobe's own pages for Firefly say outputs from its Firefly models can be used commercially, while a Photoshop generative fill FAQ has carried wording that content generated in the Photoshop app may not be used for commercial purposes. When two pages from one company disagree, do not guess: write to support, keep the reply, and use a different route for anything important until you have it in writing.

Check two: what copyright law lets you protect

In the US, the Copyright Office's January 2025 report on copyrightability reached several conclusions that matter to any small business. Copyright does not extend to purely AI-generated material, or to material where a human has insufficient control over the expressive elements. On the technology available at the time, prompts alone do not give enough control. Human authors do keep copyright in what they add: their own work visible in the output, the creative selection, coordination or arrangement of AI outputs, and creative modifications of them. Assisting a human creator with AI does not stop that creator's work being protected.

In the UK, the Copyright, Designs and Patents Act 1988 has an unusual provision for computer-generated works, treating the author as the person who made the arrangements necessary for the creation. Whether and how that applies to modern image generators has been argued over, and the government's consultation on copyright and AI, which closed in February 2025, asked about it. Its report published in March 2026 went further: it said the provision sits uneasily with the originality test, and proposed removing protection for wholly computer-generated works while keeping copyright for works made with AI assistance. That is a proposal, not yet a change in the law, so check whether legislation has followed before you rely on either position, and do not build a plan on the provision. Ask a qualified adviser if protection is central to your business.

The working rule for both: treat a raw AI image as unprotected, and treat your edits, layout and selection as the part you can protect.

What that means in money terms

Take a small online clothing store that commissions a human illustrator for its signature print, which sells across 40 products. If it used a raw generated image instead, a rival could print the same picture on its own shirts and the store would struggle to stop it. Compare the stakes: the illustrator's fee is a one-off cost, perhaps a few hundred dollars for a print, while the print drives the range. For things that carry your brand's identity, protection is worth paying for. For a Tuesday email header nobody would bother copying, it is not.

Check three: exposure from what the model produced

Models can produce images that echo living artists, famous characters, real people or recognisable logos. Courts in several countries are hearing cases about how the models were trained, and none of that gives a small publisher a safe harbour for a particular image. You can only control what you publish, so inspect it for these items before it goes anywhere.

• A face that resembles a real person, especially a public figure or a customer, which brings in privacy and endorsement problems.
• A character, mascot or costume that resembles a film, game or cartoon property.
• Any visible brand mark, including on clothing, packaging or shopfronts in the background.
• Garbled signature marks, watermark remnants or stock-photo style overlays, which are a sign the model has drifted close to a source image.
• Text in the image that is misspelt, or names a product or claim you have not approved.

A two-minute licence log for every published image

Keep one spreadsheet. One row per published image, filled in at the moment you publish. Here is an illustrative row for a craft brewery supplier's product page.

Date published:   14 Nov
Where:            Hop varieties guide, header image
Tool and plan:    Image generator, Pro plan (company revenue under the plan threshold)
Terms checked:    14 Nov, commercial use allowed, output assigned to me "if any"
Prompt (saved):   "Flat illustration of hop cones on a wooden table, muted green and cream, no text"
Edits by us:      Cropped, colour matched to brand palette, added headline text in our layout
Exposure check:   No faces, no brands, no signature marks. Checked at 200% zoom.
Reviewer:         Owner
Protection:       Treat as unprotected raw art; layout and text are ours

That takes about two minutes per image once the habit forms. Its value shows up the day a platform queries an image, a client asks "where did this come from?", or you need to explain a change of terms.

Sorting every use into safe, careful and avoid

The table below is a starting point, not legal advice. The sorting depends on whether the image is something you need to protect, something a viewer could be misled by, or something that could confuse your brand with another.

Disclosure: when a plain label is the right call

There is no one disclosure rule for every business. What exists is a patchwork: platform settings, advertising standards that ban misleading content, and transparency duties being introduced for some synthetic media. You will need to check the ones that reach you when you publish, because they are changing. Three principles hold up regardless of where the rules land.

1. Label where a reasonable person would feel misled. A realistic photo of a "customer" holding your product, a before-and-after that implies a result, or an image of a workshop that does not exist: all of these need either a clear label or replacement with the real thing.
2. Use the platform's own setting where it has one. Several social platforms offer an AI-content label at upload or detect and label it themselves. Using the setting is cheaper than arguing about it later. See the tutorial on labelling AI images and video on social media for the mechanics.
3. Never use AI to manufacture proof. Fake reviews, invented customer photos and staged results are the cases that cause real trouble, and consumer-protection rules in both the US and UK now target fake reviews. An illustration of a concept is fine; an illustration passed off as evidence is not.

A one-line caption can be enough: "Illustration created with AI." It costs nothing for decorative images and removes any argument. For images where the origin does not affect the buyer's decision, you can reasonably choose not to label, as long as the image does not imply anything untrue.

Worked example: auditing a clothing store's spring campaign

An illustration, with made-up numbers. A small online clothing store plans a spring campaign with 12 images: 5 for email and social backgrounds, 3 product-page lifestyle shots, 2 for packaging inserts and 2 homepage banners that show models wearing the clothes. It uses a generator on a paid plan and the owner runs each image through the log above.

First the terms. The owner opens the plan page and the terms, sees commercial use is allowed, notes the revenue threshold clause (the store turns over well under it) and records the date. Second, the protection sort. Backgrounds and banners are low stakes. The packaging inserts will be printed 5,000 times, so the owner decides they will be redrawn by a freelance illustrator for roughly $300 each rather than printed raw. Third, the exposure check at 200% zoom. Two lifestyle shots contain a garbled logo on a background tote bag. One banner model has a face close enough to a known actor that the owner regenerates it.

Then the product honesty check. The three lifestyle shots would show the dresses in colours the generator invented, softer than the real fabric. The owner replaces them with real photos on a phone, which costs an afternoon. Only the two banners which show models are kept, each captioned "Illustration created with AI", because customers would otherwise assume real people wearing the real garments.

Result, in numbers: 12 images planned, 5 used raw, 2 redrawn by a person for about $600 in total, 3 replaced with real photos, 2 banners kept with labels. Out of the original 12, none went out unlogged and none promised something the garment could not deliver.

A sample prompt and what to fix in the output

A vague prompt invites resemblance. A specific one steers away from it. Here is a prompt for a specialist food importer's newsletter header, and the sort of thing it returns.

Create a wide banner illustration of olive oil bottles and fresh herbs on a stone
counter, warm morning light, flat vector style, muted greens and cream.
No people, no text, no logos or brand names, no labels with writing.

An illustrative result: three bottles, a bunch of rosemary, a soft light from the left. On inspection at full size, the bottle labels carry gibberish lettering that reads almost like a real brand, and the rosemary has an extra stem. What you would fix: regenerate or paint over the labels so they are blank, remove the stray stem, and add your own headline in a text layer. The prompt's "no labels with writing" instruction reduced the gibberish but did not remove it, which is typical. Models follow exclusions loosely, so the check always has to be visual.

Before and after: turning a risky brief into a safe one

When a logo goes wrong: a realistic case

A kitchen fitter wants a logo quickly and generates a stylised hammer-and-tape emblem. It looks fine, goes on the van, the website and 200 business cards. Six months later they try to register it as a trade mark. The professional search finds a similar emblem registered for tools and building services, and the fitter has no copyright to point to for the image itself because a person did not draw it. The van wrap and the cards must be redone.

The warning sign was visible at the start: a logo is the one asset you must be able to own and defend. The fix is to use AI for sketching ideas, then have a designer draw the final mark, run a trade-mark search before printing anything, and keep the design files. The tutorial on designing a logo with AI covers the process in detail.

Quick sum: what the check costs against what it saves

Suppose you publish 20 AI images a month. Logging each takes two minutes and the visual check takes three, so about 100 minutes a month. At an owner's time value of, say, $40 an hour, that is roughly $67 a month. Against it: a single takedown complaint, a platform removing an ad account's creative, or a reprint of 5,000 packaging inserts at $0.40 each is $2,000. The check pays for itself the first time it stops one bad image in about two years of normal use.

Ask your suppliers and freelancers the same questions

Designers, agencies and freelancers increasingly use AI inside their workflow. If you commission work, put three lines in the brief and the contract.

• State whether AI tools were used, and which, for each deliverable.
• Confirm the tool's terms allow the commercial use you need, and that the supplier assigns you whatever rights it has.
• Require the original working files, because a person's layered file is the clearest evidence of human authorship you could hold.

The same questions suit stock libraries and print-on-demand services, which now sell AI-made assets alongside photographs. The tutorial on AI images for print on demand goes through the products where this matters most.

A short policy your staff can follow

Write it on one page and pin it in the team channel. An illustrative version for a six-person shop:

1. Use only the tools and plans on the approved list, and keep the terms page bookmarked.
2. Never use a real person's face, name, a known character or a named artist's style in a prompt.
3. Log every published image in the licence sheet before it goes live.
4. Logos, packaging and brand mascots go to a person for final drawing.
5. Pictures of products show the real product. Pictures of people are real, with consent, or clearly illustrative and labelled.
6. Anything doubtful goes to the owner before publishing, not after.

If you would rather build this policy around your own tools and the images you publish, that is the sort of thing my AI implementation consultation covers.

Re-checking when terms change

Terms move. Plans get renamed, thresholds change and apps get folded together. Put a recurring calendar reminder every quarter to reopen the terms pages for each tool you use, compare them with the "Terms checked" date in your log, and note any change. If something material changes, such as the revenue threshold, the commercial-use wording or a new restriction inside an app, review the images that rely on it. It is a half-hour job and it is the habit that makes the whole system trustworthy.

For choosing a tool with these questions in mind, the comparison of Adobe Firefly and Midjourney for client-safe images sets out where each one stands, and the pre-publish tutorial on checking AI-generated images for errors gives the visual checklist in more detail.

Questions owners ask before publishing AI art

Can a competitor copy my AI-generated logo?

Possibly, yes. Where copyright does not protect a purely machine-made image, you have little to stop someone reusing it, and the tool may hand out similar results to other users. Logos should be drawn or heavily reworked by a person, and ideally registered as a trade mark after a professional search.

Do I have to tell customers an image is AI-made?

There is no single rule for every business. Platforms have their own labelling settings, some regulators are bringing in transparency duties, and any image that could mislead a buyer about a product needs a plain label or should not be used. Check your platforms and advertising rules at the time you publish.

Is a paid plan always needed for commercial use?

No, but check. Some tools allow commercial use on free tiers with conditions, some tie ownership to a paid plan once your company passes a revenue threshold, and some restrict outputs made inside particular apps. The only reliable answer is the current terms page for the plan you actually use.

What if the AI copies a famous character by accident?

You can still be responsible for publishing it. Treat any recognisable character, logo, celebrity face or artist's signature style as a reason to discard the image and regenerate with a different brief, not as something the tool has cleared for you.

Should I keep records of my prompts?

Yes. Save the prompt, tool, plan, date and the final file name for every image you publish. If a platform queries an image or someone challenges it, a short log shows what you did and which terms applied at the time.

Further reads

• Can You Legally Use AI-Generated Images in Your Marketing? — The wider legal question for marketing images, from the series' first pass.
• Brand Guidelines for AI Images: Colours, Style and Limits — Set a house style so AI images look like yours.
• Can You Use AI Images on Etsy? Photos, Mockups and Disclosure — How disclosure works for AI images on a marketplace shop.
• AI for Freelance Designers: Faster Concepts Without Rights Issues — Rights questions from the designer's side of the brief.
• ChatGPT for Tattoo Artists: Admin Jobs and Designs to Avoid Copying — Four admin jobs ChatGPT does safely in a tattoo studio, two it should never touch, and the originality rules for any design that goes near AI.
• AI Model Photos for a Clothing Brand: Cost per Look and Disclosure — How much do AI fashion model photos really cost per look? Compare studio shoot expenses against synthetic models, fit accuracy, and disclosure rules.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: US Copyright Office, Copyright and Artificial Intelligence Part 2 (Copyrightability); OpenAI Terms of Use and Service Terms; Midjourney Terms of Service; Adobe Firefly plans page and Photoshop Generative Fill FAQ; UK Copyright, Designs and Patents Act 1988; UK government copyright and AI consultation and its Report on Copyright and Artificial Intelligence (March 2026); FTC final rule on fake reviews; CMA guidance on the fake reviews banned practice.