If you took the photo and used AI tools to edit it, you generally still own the copyright in the photograph. Purely AI-generated images are the weak spot: in many places copyright needs a human author, so a prompt alone may give you no protectable rights. Say so in your contract, and have a solicitor check the wording.
Think of a spectrum, with a colour grade at one end and a prompt that makes the whole image at the other. The more the software decides what appears in the frame, the weaker your claim to own that part, and the more your contract needs to say about who carries the risk. Nothing here is legal advice, copyright rules differ between countries, and they are still moving.
Editing or generating: where your photo sits
A lot of AI use in photography is plainly assistive, but the line between assisting and generating is where disputes start. Here is the spectrum in seven steps, with the usual position on each.
| Step | Example | Who makes the creative choices | Usual ownership position |
|---|---|---|---|
| 1. Culling and sorting | AI ranks the set, flags blinks and duplicates | You shot it; the software only orders it | Yours; nothing about the pixels changes |
| 2. Colour, tone and presets | Auto exposure, a preset matched to your style | You, choosing or training the look | Yours |
| 3. Masking, noise reduction, upscaling | AI subject selection, denoise, enlarge | Software recalculates your own capture | Yours; this is assistive use |
| 4. Generative remove | Taking out a bin or a stranger; the gap is filled with generated pixels | Mostly your photo, with a small generated patch | Yours as a photograph; the generated patch is small, but disclose it if the client cares |
| 5. Generative fill or expand | Replacing a sky, extending the frame, adding objects | Large areas made by the model | Your capture and your arrangement are protectable; the generated areas may not be on their own |
| 6. Altering people | Changing an expression, swapping a face, adding someone | The model makes a likeness | As step 5, plus consent and likeness questions |
| 7. Text to image | A whole image from a prompt | The tool, steered by your words | May not be protectable at all; never call it a photograph |
Restyling your own photo with an image-to-image tool sits between steps 5 and 7, depending on how much the model changes. In a contract, treat it like step 5 or step 7, whichever is closer.
The clearest official guidance I found is from a national copyright office, published in January 2025. In summary: using AI tools to assist human creativity, rather than stand in for it, does not by itself remove protection from the human-made work. Prompts alone do not give a user enough control to count as authorship. Human expression that can be seen in the output, and a person's creative selection, arrangement or modification of AI output, can be protected. And each case is decided on its facts. That fits the table: steps 1 to 3 are assistive, steps 4 and 5 are mixed, and step 7 is where authorship is weakest.
Elsewhere the picture varies. A handful of countries' laws contain a special rule for computer-generated works, others insist on a human author, and some are reviewing the question. So the safe working assumption is uncertainty. Do not tell a client they will own a fully generated image outright, and ask a solicitor who knows copyright where you and the client are based.
What the tools' own terms say about the output
Ownership in a contract is only as good as the rights you have from the tool. Two patterns are worth knowing. OpenAI's and Anthropic's terms assign to you whatever rights they hold in an output, "if any". That wording cannot create copyright the law does not recognise, so it tells you the provider will not claim the image, not that anyone owns it.
Adobe takes a different line. Its Generative Fill page says the current Firefly model behind the feature was trained on licensed content, such as Adobe Stock, and on public-domain content, and answers yes to whether it is safe for commercial use. The same answer adds that content generated in "the Photoshop app" may not be used for commercial purposes, without saying which version it means, so check the terms for the app you actually use. Generative Fill in Photoshop on desktop also offers partner models, such as Google's Gemini and FLUX, and Adobe's commercial-safety claims are made for its own Firefly models, not for those, so note which model made each fill. Adobe also says Firefly outputs carry Content Credentials, a provenance label attached to the file that shows AI was used; check that your export settings keep it.
Next comes the question of who pays if an output infringes someone else's work. OpenAI's output indemnity in its service terms covers API and Enterprise, Edu and Healthcare customers, not ChatGPT Business, and editing an output or using it as a trade mark falls outside it. Anthropic's position on edited outputs is similar. Adobe's IP indemnity is enterprise-only. For a solo photographer on a standard plan, assume you carry that risk yourself, which is a good reason to keep generated content small and to say so in your contract.
Whatever tool you use, read the same six things in its current terms. Terms change, so for any significant commercial job, save a dated copy of the terms you relied on.
| Question for the terms | Why it matters to a photographer |
|---|---|
| Who owns or may use the output? | Decides what you can pass on to a client, and whether the provider keeps any right |
| Is commercial use allowed on your plan? | Free and paid plans can differ; client work needs a clear yes |
| Is there an indemnity, and who gets it? | Tells you who pays if a third party claims the output copies their work |
| Does it train on what you upload? | Client images, and especially children's, should not feed a model |
| How long are uploads kept, and where? | A client's privacy notice or your own may need to say so |
| Any limits on real people's likenesses? | Altering faces and bodies is where tools and consent rules meet |
Three things ownership does not settle
Consent comes first. Changing someone's face or body with a generative tool is a different act from retouching a blemish, and the person in the picture, not just the client, may have views. Releases and contracts should say whether images may be altered by AI, and children and other sensitive subjects need extra care. My notes on GDPR for photographers: client galleries, contracts and AI tools cover where those uploads go.
Labels come second. Platforms, clients and, in some places, the law expect AI-altered images to be marked. If you have customers in the EU, transparency duties for AI-generated or manipulated images that could pass as real have applied since 2 August 2026, and whether a particular edit counts is a question for your solicitor. On the technical side, the file can carry the label. Content Credentials is one route. The IPTC photo metadata standard has a digital source type field for the purpose, and Google Merchant Center asks that AI-generated product images keep metadata marking them as such, so a retail client's feed can depend on it. Never strip that metadata, and tell clients not to. For the social side of labelling, see how to label AI-generated images and video on social media.
Infringement is the third. A generated element can resemble someone else's work without anyone intending it, and you cannot easily check. That is why steps 5 to 7 deserve a sentence in the contract about what you promise and what you do not.
Nine clauses to adapt, in the order a contract usually needs them
These are drafting suggestions, not legal advice. Square brackets are choices for you, and your solicitor should check the final text against the law where you and your client work. Only promise what you actually do, because a clause you do not follow is worse than none. For the licence side in full, and for gallery wording, see photography usage rights: draft client licence wording with AI and AI disclosure for photographers: contract and gallery wording.
1. Ownership and licence
The Photographer owns the copyright in the images created under this agreement, subject to clause 6. On payment in full, the Client receives a [non-exclusive] licence to use the delivered images for [purposes] in [territory] for [period]. The licence [does / does not] include the right to modify the images, including with software or AI tools, beyond cropping and resizing.
The modification line is new. AI tools make altering an image trivial, so say whether the licence covers it, or you will argue about it later. If a client needs exclusivity, state it and price it.
2. Two definitions
"Captured Image" means an image recorded by the Photographer's camera. "AI-Generated Element" means any part of a delivered image that was created, extended or replaced by a generative AI feature, as distinct from an adjustment of the captured pixels.
Everything else hangs on these. They let you say exactly which parts you are promising rights in.
3. Everyday AI-assisted editing
The Photographer may use software tools, including tools with AI features, to sort, select, retouch, adjust colour and tone, reduce noise, mask, upscale and make similar adjustments to Captured Images. These tools adjust the Photographer's own images and do not need separate approval.
Steps 1 to 3 of the table. It stops the word "AI" from alarming a client about routine work.
4. Generative changes need approval
The Photographer will use generative AI features that add, replace, remove or extend content in an image (for example replacing a sky, removing or adding objects or people, or extending the frame) only where the brief calls for it or the Client has approved it. The Photographer will not use generative AI to alter a person's face or body beyond the retouching described in the brief without that person's and the Client's written approval. Images containing AI-Generated Elements will be identified to the Client on delivery.
Steps 4 to 6. List what counts as routine retouching in the brief (skin, stray hairs, flyaways) so that the line is clear.
5. Fully AI-generated images
Where the Client commissions an image created entirely or mainly by a text-to-image or similar system, the Photographer will identify it as AI-generated, will not describe it as a photograph, and delivers it on the basis set out in clause 6.
Step 7. If you never produce these, leave the clause out and say in the contract that you supply photographs only.
6. What you do not promise about AI-generated parts
The Photographer licenses or assigns rights only in Captured Images and in the Photographer's own creative selection, arrangement and edits. The Photographer makes no promise that copyright exists in any AI-Generated Element, or that any AI-Generated Element is exclusive to the Client, because the law on this is unsettled. [Optional: Where the Client requires an assignment, the Photographer assigns copyright in the Captured Images on payment in full, and AI-Generated Elements are delivered without warranty as to copyright or exclusivity.]
This is the clause that matters most for commercial clients, who often ask for outright ownership. It tells them in advance what you can and cannot hand over.
7. Labelling and metadata
Where law, a platform's rules or the Client reasonably requires AI-altered images to be labelled, the Photographer will mark the relevant files, for example in their metadata, as the Client requests. The Client is responsible for labelling the images in its own publications and must not remove metadata that identifies AI-Generated Elements.
Put the burden for publication where it belongs, with the publisher.
8. The Client's own use of AI
The Client will not use the delivered images, in whole or in part, to train, fine-tune or evaluate a machine-learning model, or to create synthetic likenesses of the people shown, without the Photographer's written consent. The Client will not upload the images to a service whose terms allow it to train on them or claim rights in them [, and will switch off automatic AI image enhancement features on advertising platforms for these images].
Clients and their agencies increasingly run images through generators and ad tools. This clause also covers headshots fed to avatar makers. The bracketed part is stricter, so keep it for commercial work.
9. Your own use of AI services
The Photographer will use AI services that process Client images only where the provider's terms do not allow it to train on those images or claim rights in them, and will not upload images of children or other sensitive material to such services without the Client's agreement.
Check that you can keep this promise before you make it. If you use an outsourced editing service, make the same terms bind them in writing.
Clause sets for a wedding, headshots and an advertising shoot
Clauses 1, 2 and 6 go into every version, because clause 1 points to clause 6 and the others rely on clause 2's definitions. For a wedding, add clauses 3, 4 and 9 to that core. You cull with AI, edit with assistive tools and use generative remove on two images, a bin and a stranger in the background. The clauses cover all of that, and clause 9 matters because guests' and children's faces sit in the galleries. A request to place a late relative into the family photo is exactly what clause 4 exists for. Agree it in writing, label the image as a composite, and decide whether you are comfortable doing it at all.
For corporate headshots, add clause 8. A company may well feed staff portraits into an AI avatar tool, and without that clause you have nothing to point to. Clause 4 matters too, because "open the eyes" and "take off the glasses glare" are face edits, and each person pictured should agree.
For an advertising shoot where a studio product photo is extended into a banner by a generative tool, clauses 4, 7 and 8 join the core, and clause 6 earns its place. Suppose an agency takes your delivered image, uses an ad platform's tool to extend the background, and later asks who owns the banner. If your contract said nothing, both sides argue. With clauses 1 and 6, the answer is already on paper: the licence says whether modification was allowed, and you have not promised exclusivity in the extended part. The agency may insist on an assignment, and that is the bracketed option in clause 6, priced accordingly.
Telling the client about one generative edit
Disclosure works best as a short, calm message before delivery. A sample (use your own wording):
Hi [first name], one note on your gallery. In three images I removed distracting objects (a bin, a sign) using a generative remove tool, which fills the gap with computer-generated pixels. Everything else is your original photograph with standard adjustments. If you would rather have these three left as shot or retouched by hand, tell me by [date] and I will swap them. [your name]
It names the images, says what the tool did, gives the client a choice and puts a date on it. A message like that also becomes your record of approval under clause 4.
An AI log, and a ten-minute audit of your last delivery
Keep a one-line log per job: file, tool and feature, what changed, who approved and how the file is labelled. It takes seconds as you work and saves an hour when a client asks.
| File | Tool and feature | What changed | Approval | Label |
|---|---|---|---|---|
| Image 0412 | Editor, generative remove | Removed a bin from the background | Client message, [date] | Per client preference |
| Image 0688 | Editor, generative expand | Extended the sky by a third for a banner crop | In the brief | Metadata marks the AI element |
| Image 0701 | Editor, noise reduction | Cleaned a low-light frame | Not needed (clause 3) | None |
Then audit your last delivery. Sort every image into the seven steps. Suppose you delivered 30 images: 24 were assistive only (80%), 4 used generative remove (13%), 2 used generative expand (7%) and none was fully generated. That tells you six images, 20% of the set, contain generated pixels, so six lines of the log need approvals and clause 4 applies to them. Only two images carry large generated areas, so those are the two where clause 6 and the labelling question deserve a second look. The audit takes ten minutes and shows whether your contract covers what you actually do.
What to ask your solicitor to check
- Which country's law governs the contract, and whether the copyright position on AI-Generated Elements differs there from where the client will use the images.
- Whether a licence or an assignment suits each type of client, and how clause 6 interacts with any rights the client's own terms claim.
- Whether you need an attribution or moral-rights wording where those rights exist.
- Whether the labelling clause matches any duties that apply to you or your client, including for audiences in the EU.
- Whether the warranty and indemnity wording is proportionate, given that the tool terms may leave much of the infringement risk with you.
- How your professional indemnity insurance treats AI-generated content, which is a question for your insurer as well.
- Whether subject releases and your privacy notice mention AI tools and where uploads are processed.
Take the clauses above, your AI log and the audit result to that conversation. A solicitor who can see what you actually do has far more to work with than one starting from a blank page. If you also want an AI-drafted contract template with a rights check built in, a photography contract template: let AI draft it, then check rights shows that routine.
Photographers' questions about AI and copyright
Can I enter AI-edited photos in competitions or stock libraries?
Rules differ and change, so read each competition's and agency's current AI policy before you submit. Some limit generative edits, and some require disclosure. Keep your AI log so you can answer honestly, and keep the original RAW files, which are your best evidence of what you captured.
Does AI culling or auto-editing change who owns my photos?
No. Culling and auto-editing are assistive, and the creative choices stay with you. What to check is the service's terms on training and storing your uploads, where it processes images, and whether you need to tell clients, especially for galleries that include children.
Can I stop AI companies using my published photos?
Not completely. You can state your terms in your licence and website footer, use metadata fields that say data mining is not permitted where your software supports them, and check whether the platforms you post on offer an opt-out. These are signals and settings, not guarantees, and the law on training differs by country.
What if my second shooter or editor uses AI?
Your agreement with them decides. Make sure freelancers, second shooters and outsourced editors assign or license their work to you in writing, agree to your AI rules and use only services whose terms do not train on your clients' images. Otherwise your promises to the client may not be ones you can keep.
Further reads
- Claude for Photographers: Contracts, Client Emails and Captions Tested — Using Claude to draft contracts, emails and captions, tested.
- How Photographers Cut Editing Time With AI and Keep Their Style — Cutting editing time with AI while keeping your style.
- Best AI Photo Editor for Photographers: Cull, Edit or Retouch First? — Where in the workflow each AI editor fits.
- AI Tools for Ecommerce Photography: Backgrounds, Shadows, Consistent Sets — Generated backgrounds and shadows for product work.
- AI for Freelance Designers: Faster Concepts Without Rights Issues — A designer's view of the same rights questions.
- How to Use AI in a Photography Business: Start With Admin, Not Editing — Why admin is the safer place to start with AI.
- Wedding Venue Instagram: Real-Wedding Posts and Couple Consent With AI — How a venue can post real weddings on Instagram with the couple's consent, the photographer's permission, checked supplier tags and AI-drafted captions.
- Does Dubsado Have AI? What to Use Alongside It for Client Admin — Dubsado lists AI form building, a call notetaker and email summaries. Here is what they do, what they cost, and what ChatGPT or Zapier adds beside them.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: a national copyright office's January 2025 report on the copyrightability of AI outputs and its announcement of it; OpenAI and Anthropic terms on output ownership and indemnity; Adobe's Generative Fill and Firefly pages (checked October 2026); IPTC and Google Merchant Center guidance on AI-image metadata. Not legal advice.