Where to draw the line with AI in photography
A client asks whether you used AI on their gallery. The honest answer depends on what you mean by AI.
Five questions
Ask these before you reach for the tool
Does it change what the picture claims happened?
Removing dust is cleanup. AI-generated people or objects, or AI reconstruction that changes what was in the scene, alters the record.
Who needs to know?
The client, the contest, the platform, and the law each have their own expectations.
What happens to the photograph while the tool is working on it?
Does it upload? How long is it retained? Can the provider use it to improve models? Client and NDA work make this a first-class question—terms differ by product, so check before you send.
What remains yours?
Your authorship, your creative decisions, and the parts of the work you actually created. AI-generated material can raise separate copyright questions.
What stays yours to do?
The cull on jobs that matter, the caption, the final look—and a human still reviewing the delivered gallery.
Assistive, generative, and admin are different tools
Culling software, noise reduction, and masking speed up the work. AI-generated people or objects, or AI reconstruction that changes what was actually in the scene, changes what the photograph says happened. Most arguments about AI in photography are two people using one word for two different things.
Accurate as of October 2026. Contest rules, platform labels, and laws change, and nothing here is legal advice.
“Autofocus already uses AI” is true and beside the point. Autofocus helps you capture the frame. Generative fill changes what is in it.
There is a second axis: what the tool does to the image versus what it does with the image. Local noise reduction on your machine and uploading a client wedding gallery to an external AI service can look similar in the result and very different as a privacy and business decision. Check processing, retention, training, and privacy terms before client, NDA, editorial, corporate, or conservation work leaves your studio.
| Bucket | Examples | The test |
|---|---|---|
| Assistive | Culling, noise reduction, masking, matching your look, skin and flyaway cleanup | Helps you make the picture you captured |
| Generative | Fill, expand, sky replacement, AI-generated people or objects, AI reconstruction that changes what was in the scene | Changes what the picture shows |
| Admin | Emails, shot lists, first drafts, scheduling | Never touches the picture |
The line moves with the job
Weddings, documentary, and news work depend on the picture being a record, so clients, contests, and newsrooms are strict, and many disqualify generative edits outright. Commercial and stylized work has more room, but the contract still decides. Fine art is yours to define, as long as you say what you did.
Who needs to know
Disclosure is not one rule for everyone. Match the audience to the step. Provenance in the file is not the same as telling a person.
| Audience | What they need | Practical step |
|---|---|---|
| Client | What your editing includes | Define reconstruction vs retouching in your editing policy, and tell the client when a composite matters—e.g. combining elements from multiple frames of the same scene |
| Contest or publication | Whether generative edits were used | Read the rules before you enter, and assume generative fill is not allowed unless they say it is |
| Platforms and viewers | What the image’s provenance or label communicates | Some tools add Content Credentials or other provenance information. Treat that as provenance, not a substitute for whatever disclosure a contest, client, platform, or law requires |
| The law | Whether specific AI-generated or manipulated content must be disclosed | Rules vary by jurisdiction and use. In the EU, Article 50 transparency rules apply from August 2026, including specific rules for deepfakes and certain public-interest content; assistive standard editing is largely outside marking obligations. Get professional advice for contracts or regulated work |
What you can protect
In the US, copyright needs a human author. Your photograph is yours. Material a model generates on its own may not be protected, and there is no clear rule yet on how much human input is enough. You can retain copyright in the photograph while incorporating material whose copyright status is different. If you plan to register or license work, keep the original capture and your edit history.
The creativity question
Does AI make you less creative? It can, and the order matters. Letting a tool write the first draft tends to make results look alike and leaves you with less sense that the work is yours. Using it after you’ve made your own choices tends to help.
In practice: make your selects first, write your caption first, then ask a tool to tighten it. Your voice is the product. Captions, an About page, and emails written entirely by a model read like everyone else’s, and clients hire the difference.
Honest verdict
Use assistive tools freely. Keep your own judgment where trust is the product: the cull on a wedding, the caption on a documentary frame, and a human looking at the final set. Check what a tool does with the files before client work goes through it. Disclose generative edits before anyone has to ask.
Assistive tools are craft. When a tool changes what the picture says, say so.
Speed up the work. Don’t rewrite what happened.
Assistive tools are craft. When a tool changes what the picture says, say so.
Also in Guides: AI can build software. It can’t run your studio. · Can AI build a photography website? · Public vs private work