7 Ways to Use AI Without Stealing Anyone's Creative Work

Ways to use AI without stealing anyone's creative work usually turns into a guilt trip: don't do it, full stop. That advice doesn't hold up, because the technology isn't going anywhere, and pretending otherwise doesn't protect a single artist.

The more useful question is the one most listicles skip: how do you tell whether a tool respects the people whose work trained it? It turns out you can, in a few minutes, with checks anyone can run. Adobe built its Firefly model on licensed Adobe Stock content and public domain material only, and set up a compensation fund to share value with the contributors behind it. That kind of transparency exists. Most people just never think to look for it.

So this list isn't seven reasons to feel bad. It's seven concrete habits, each one something you can start using today.

1. Check Whether Your Image Generator Licensed Its Training Data

Renata runs a small ceramics studio and started using an AI tool to mock up product photography backgrounds before a client shoot. She'd never once looked at where the tool's training images came from until a customer asked her directly. The answer took four minutes to find on the company's own site, and it wasn't the answer she assumed.

Most AI image and video tools now publish some version of a training data policy, even if it's buried a few clicks deep. Some, like Firefly, lead with it. A growing number attach Content Credentials, a tamper-evident label showing exactly what generated or edited a file. Others say nothing at all, which tells you something too.

Start today: Before you generate anything for a paying client, search "[tool name] training data" or look for a Content Credentials label or "responsible AI" page on the company's own site. If you can't find one in five minutes, treat that as your answer.

2. Ask AI Whether Your Own Prompt Is Imitating a Specific Living Artist

It's easy to reach for "in the style of [name]" without thinking about it, the same way you'd reach for a reference photo. The difference is a reference photo doesn't get sold as a shortcut around someone's decades of work.

The model itself is a decent check here. Paste your prompt in before you run it and ask directly whether it names or closely mimics a specific working artist's signature style. Most will tell you plainly, and most will offer a rewrite.

Start today: Take your last few image or writing prompts and ask an AI model whether any of them reference an identifiable living artist. Replace any name with the actual descriptive qualities you wanted: the color palette, the linework, the mood, not the person.

3. Know Which Tools Pay the Artists Behind the Style You're Borrowing

"AI-generated" doesn't automatically mean nobody got paid. It depends entirely on which tool you picked. Adobe's contributor compensation fund is one concrete example: creators whose Adobe Stock images trained Firefly receive royalties and bonus payments tied to the model's use. Shutterstock's Contributor Fund runs on a similar principle, paying an average 20% corporate royalty rate back to the photographers and illustrators whose libraries licensed the underlying data.

Start today: Before assuming the worst about a tool, check its contributor or creator page directly. If it has one, that's a signal worth factoring into which tool you reach for next time.

4. Turn Your Own Back Catalog Into New Work, Never Someone Else's

The cleanest way to avoid the whole question is to point AI at material you already own the rights to. A photographer with two hundred of their own past shoots, a writer with a decade of their own published essays, a designer with a full portfolio: all of that is a usable base for generating new variations that are unmistakably built from your own hand.

This is the one item on this list worth setting up properly instead of doing manually each time. BYOBot can organize and index your own content library so a new piece draws only from work you already own, no outside training data involved.

Start today: Gather twenty to fifty pieces of your own past work into one folder. That folder is the starting spec for a workflow built entirely on material that's already yours.

5. Read the License on a "Royalty-Free" AI Asset Before You Use It Commercially

Owen sells home goods on Etsy and used a "free" AI-generated background image on twelve product listings before he read the fine print. The license covered personal and editorial use only. Commercial resale wasn't included, and he'd been technically violating it for three months without realizing.

"Royalty-free" and "commercially clear" are not the same phrase, and AI asset libraries mix both freely.

Start today: Before publishing any AI-generated asset commercially, open the actual license tab, not just the download button, and look specifically for the words "editorial use only." If you see them, don't use the asset for a product listing, ad, or anything you're selling.

6. Add One Honest Disclosure Line Instead of Hiding the AI-Assisted Parts

The instinct is to hide it. The better move is usually the opposite. Clients and audiences who discover AI involvement after the fact, without being told, tend to feel misled even when the work itself is fine. Clients told upfront, with the human parts made clear, tend to trust the relationship more.

Start today: Pick one piece you're about to publish or deliver this week and add a single honest line: what was AI-assisted, what was you. Watch how the reaction compares to work you didn't disclose.

7. Set Your Own Personal Rule for What You'll Never Ask AI to Generate

This last one doesn't need AI at all, and that's the point. The honest version of this whole topic isn't a universal rule everyone has to follow. It's a line you draw for yourself, once, so you're not re-deciding it under deadline pressure with a client waiting.

Start today: Write one sentence defining your own boundary, something like "I won't deliver final client-facing art without at least one human pass," and save it somewhere you'll genuinely see before your next project starts.

Pick One

Don't try all seven this week. Pick the check that made you wonder what you'd find if you looked, and go look.

Sunday, All Things Agentic reports what changed in AI that week. Wednesday, The Loop builds one workflow end to end. This column hands you the ideas in between, and the full directory of agent workflows is there whenever one of these turns into something worth setting up for good.

Frequently Asked Questions

• Is it ever okay to use AI-generated art commercially?+
  Yes, when the tool trained on licensed or public domain material and you check the specific usage terms on the asset you generated. The line isn't AI versus no AI, it's licensed versus scraped.
• How do I know if a tool trained on stolen work?+
  Look for a public training data policy or Content Credentials on the tool's own site. If a company can't say where the training data came from, that silence is itself the answer. BYOBot's workflow directory only recommends tools that publish this openly.
• Doesn't disclosing AI use hurt my credibility with clients?+
  Usually the opposite. Clients who find out later that a piece was AI-assisted and weren't told feel misled. Clients who are told upfront, with the human parts made clear, tend to trust the relationship more, not less.
• What if I can't tell whether my own prompt is imitating someone?+
  Ask the model directly. Paste your prompt and ask whether it references a specific living artist's identifiable style. Most models will tell you plainly and suggest descriptive language to use instead.