Two of the three largest stock agencies do not accept AI-generated submissions from contributors at all. Getty bans them outright; Shutterstock does not take contributor uploads of AI content while selling AI generation to its own buyers. Adobe Stock does accept them, with labelling and release conditions most AI portrait work cannot meet.
The pitch is everywhere: generate images for pennies, upload them to stock agencies, collect royalties forever. It is one of the most confidently repeated AI side-hustle claims, and the part it leaves out is that most agencies will not take the images.
This guide covers what the major agencies actually permit, the release requirement that quietly rules out the highest-earning category, what the images have to survive technically, and the places where selling generated images does work.
It is not an encouraging article. It is an accurate one.
Generating commercially either way? Rangy runs the models on your own API key, so you keep the files and the terms are the provider's, not a middleman's.
Try it freeWhich agencies accept AI-generated stock?
Fewer than the side-hustle videos suggest. Getty Images stopped accepting submissions created with generative models, citing copyright concerns about training data. Shutterstock does not accept AI-generated content uploaded by contributors, while offering AI generation to its own customers. Adobe Stock does accept it, under conditions.
That asymmetry is worth understanding rather than resenting. An agency's product is not images, it is indemnified images — the promise that a buyer can use a picture commercially without being sued. Generative models complicate every part of that promise, from training-data provenance to whether a face resembles a real person.
Two practical consequences follow:
- Check before you generate, not after. Policies differ per agency and have been revised repeatedly since 2022. Read the contributor agreement of the specific platform you intend to upload to, today.
- Uploading AI content where it is not permitted risks the account, not just the individual file. Contributor accounts get closed for policy breaches, taking the back catalogue's earnings with them.
What does Adobe Stock require?
Labelling, restraint in the prompt, and correct categorisation. Content made with generative tools must be marked as such at submission using the "Created using generative AI tools" checkbox, photo-realistic results are submitted as Photos and everything else as Illustrations, and you must confirm your generator's licence permits commercial stock use.
Adobe's generative AI content guidelines are the authoritative version and worth reading in full before you upload anything. The parts that catch people out:
- You cannot prompt with other people's IP. Work created with prompts referring to or based on artists, real known people, fictional characters, places or property is not permitted without the legal right to use them.
- The label is not optional. Mislabelling generated content as photography is a policy breach rather than a shortcut.
- Your generator's licence has to allow it. Some tools' terms do not permit broad commercial redistribution, which is a separate question from the agency's rules.
- Keywords and titles must describe the image, not the prompt. Prompt text is not metadata.
Why can you not sell AI photos of people?
Because of the model release. Adobe's guidelines state that content depicting, based on, or intended to portray an identifiable person requires one — and a generated person has nobody who can sign it. That rules out most of the highest-earning stock category before you have generated anything.
This is the single most consequential thing in the article and almost nobody mentions it, because the side-hustle framing never gets past "generate images, upload images".
The tension is genuinely awkward rather than merely bureaucratic:
- People sell. Business teams, doctors with patients, families, hands holding products. That is what buyers licence most.
- A generated person cannot sign anything. There is no subject to obtain a release from.
- Resemblance is a risk you cannot audit. Models are trained on photographs of real people, and you have no way to prove an output resembles nobody in particular.
What is left is the category without people in it — backgrounds, textures, objects, abstract concepts, illustration. Lower competition to shoot, and correspondingly lower demand.
Would the images even pass review?
Often not, and the failures are invisible until you open the file. The corporate scene at the top of this article looks entirely convincing at web size. At 100%, the hand reaching over the laptop has fused fingers with no separation, and the papers on the desk are illegible with the pen merging into a hand.
That image took one generation and five cents, and it is genuinely good — the lighting is right, the composition leaves copy space, the skin has texture. It would still be rejected, because stock review happens at full resolution and every reviewer knows where to look.
The recurring rejection triggers in generated images:
- Hands. Still the first place a reviewer looks, and still the most common defect.
- Text and signage. Screens, papers, whiteboards, labels. Illegible text reads as a defect even when nothing else is wrong.
- Object logic. Chair legs that do not meet the floor, cables going nowhere, a pen that becomes a finger.
- Repeated faces or patterns in a crowd or a texture, from the model reusing the same element.
- Over-smoothing. Skin and fabric that have no pores or weave at full size.
Audit at 100%, not at fit-to-screen. Open every candidate at full size and look at every hand, every piece of text, and every place two objects meet. This is the same check that catches problems in generated ad creatives, and it takes about thirty seconds per image.
What do AI stock images actually earn?
Very little per image, which is true of stock photography generally rather than a quirk of AI. Individual downloads pay cents to low single-digit dollars depending on the licence and the contributor's tier, so income depends almost entirely on portfolio size and how often files get found.
The arithmetic that makes stock work has never been about the quality of one image. It is about thousands of files, each earning rarely, discovered through metadata. Generation changes the cost of producing files and changes nothing about discovery.
Which creates a problem the enthusiasm rarely acknowledges: if generation makes supply nearly free for you, it makes supply nearly free for everyone. A category that fills up with cheap near-identical images does not become more profitable to enter.
Exact rates vary by agency, licence type, contributor level and region, and they change — so no figures are quoted here. Look up the current royalty schedule of the platform you are considering rather than trusting a number in an article.
What kinds of AI images do sell?
The ones without recognisable people, without readable text, and without an obvious claim to document something real. Abstract backgrounds, textures, patterns, conceptual illustrations, product-free scenes and stylised graphics avoid the release problem and the artefact problem at the same time, which is why they are the categories that survive review.
The viable list is short but real:
- Backgrounds and textures. Gradients, paper, fabric, marble, bokeh. No hands, no faces, no text, and buyers genuinely need them.
- Conceptual illustration. Ideas that are hard to photograph — data, growth, connection — where a stylised treatment is expected.
- Patterns and tiles. Repeating decorative work for design use.
- Landscape and architectural fiction, clearly stylised rather than passing as documentary.
Notice what these have in common: they are the categories where nobody was going to ask "is this a real photograph". The moment an image implies it documents something real, the whole indemnity question wakes up.
Where else can you sell generated images?
Places that sell a product rather than a licence. Print-on-demand puts the image on a physical object, digital-product marketplaces sell it as an asset pack, and your own storefront sets its own rules. All three avoid the indemnity problem that makes agencies cautious.
| Route | What you are selling | Main constraint |
|---|---|---|
| Print on demand | A printed object | Resolution and disclosure rules |
| Digital asset packs | Textures, backgrounds, game art | Marketplace terms vary widely |
| Your own store | Whatever you decide | You provide the traffic |
| Client work | A commissioned result | Disclose that AI was used |
Marketplace policies on AI-generated content differ and have changed repeatedly; confirm the current terms before listing.
Of these, print on demand is the most established path for generated images and has its own resolution requirements, which are worked through in the print-on-demand guide. Asset packs are covered in the game assets guide.
When is this a bad plan?
When the plan is passive income from volume. Uploading thousands of generated files to agencies is the version of this that most often ends in rejected batches or a closed account, and it is the version most commonly recommended. Treat generated stock as a small supplement, not a business.
Stated plainly, because this guide is published by a company that sells image generation:
- Two of the three biggest agencies will not take the files. That is not a workflow problem you can prompt around.
- The best-selling category is closed to you by the release requirement.
- Bulk uploading invites scrutiny. Agencies have tightened review specifically because generation made spam cheap.
- Copyright in purely AI-generated images is unsettled and differs by jurisdiction, which matters if your business depends on owning something exclusive.
- Real photographers still have the advantage in exactly the categories that earn: real people, real places, real releases.
The honest version of the opportunity is narrower and duller than the videos: generate backgrounds and conceptual work for the one major agency that accepts them, label everything correctly, and expect supplementary income rather than a replacement one.
Frequently asked questions
Can you sell AI-generated images on stock sites?
On some. Getty Images stopped accepting submissions made with generative models over copyright concerns, and Shutterstock does not accept AI-generated content uploaded by contributors while offering AI generation to its own buyers. Adobe Stock does accept it, provided the content is labelled as generative AI at submission and meets its prompt and release conditions.
Do AI-generated people need a model release?
Adobe's guidelines state that content depicting, based on, or intended to portray an identifiable person requires a model release. Since a generated person has nobody who can sign one, this effectively closes the people category — which is the highest-earning part of stock photography. It is the single biggest limitation on the whole idea.
How do I label AI content when submitting?
On Adobe Stock there is a "Created using generative AI tools" checkbox in the submission flow, and photo-realistic results are filed as Photos while everything else is filed as Illustrations. Labelling is a requirement rather than a courtesy, and describing a generated image as photography is a policy breach.
Why do AI stock submissions get rejected?
Usually artefacts found at full resolution rather than anything conceptual. Hands with fused or extra fingers, illegible text on screens and papers, objects that do not physically connect, repeated faces in crowds, and skin or fabric with no texture. Review happens at 100% zoom, so defects invisible at web size are the ones that fail.
How much can you earn selling AI stock images?
Per image, very little — which is true of stock generally, not just AI. Income depends on portfolio size and discoverability rather than the quality of any one file. Rates differ by agency, licence, contributor tier and region and change over time, so check the current royalty schedule where you plan to contribute rather than trusting any published figure.
What AI images are safest to submit?
Anything without identifiable people, readable text or a claim to document something real. Backgrounds, textures, patterns, gradients and conceptual illustration all avoid the release requirement and are less likely to carry the artefacts that fail review. They also face the most competition, because everyone else reached the same conclusion.
Is it better to sell AI images somewhere other than stock?
Usually yes. Print on demand sells a physical object rather than a licence, digital marketplaces sell asset packs, and your own storefront sets its own rules — all of which sidestep the indemnity question that makes agencies cautious. Each has its own disclosure rules and resolution requirements worth checking first.
The bottom line
Check the contributor agreement before you generate anything, expect the people category to be closed, audit every file at 100% for hands and text, and treat agency stock as one narrow outlet rather than the plan. Print on demand and asset packs are the more realistic routes.
None of this is a judgement on the images. The corporate scene at the top of this article is good work for five cents, and the quality of generated stock is no longer the limiting factor.
The limiting factor is that stock agencies sell a legal promise as much as a picture, and generated images make that promise harder to keep. Until that changes, the honest advice is to sell generated images where what you are selling is the image itself.
"This software has increased my workflow speed tenfold, and the output quality it has delivered in my work has been exceptional."
Generate on your own terms
Rangy runs the image models on your own provider account, so the licence terms are the model provider's and every file is saved to your disk — useful whichever route you sell through.
Download Rangy Free Or read the print-on-demand guide →Mac & Windows · Free plan, no credit card · 5 generations a day
The agency positions are drawn from each company's own published policy: Adobe's generative AI content guidelines for what it accepts and requires, and Getty's and Shutterstock's public statements on contributor submissions. Those positions have changed before and will change again, so the contributor agreement of the platform you are uploading to is the authority, not this page. No royalty rates are quoted, because they differ by agency, licence, contributor tier and region and are revised regularly. The corporate scene was generated with GPT Image 2 at 2K for $0.05 and is shown unretouched; the magnified crops are from that file, and the defects described were found by inspecting it at full resolution. The agency diagram deliberately names no company, because the positions it illustrates are a snapshot.
This guide is published by Rangy, which makes one of the tools it describes, so it is not a neutral source. It also argues against a use case that would sell more of that tool — the reasoning is in when this is a bad plan.