Frequently Asked Questions
About
- What is AgentsOfAI.art?
- A curated marketplace for commercial-grade, AI-generated stock video and images. Creators upload and sell their work; buyers license and download it. Every account can do both.
- Why does it exist?
- We are creators who built the marketplace we wanted for our own work: creators keep 65% of every sale, their name stays on their work, the trending feeds are ranked by community votes, and search reads the picture itself, so work gets found on merit. The full story — including where every dollar goes and what we have promised never to do — is on the About page.
Search
- How does search work here?
- Probably differently from every stock site you have used. Traditional stock search matches your words against keywords contributors typed in — you find what someone remembered to describe, and nothing else. Here, search looks at the pictures and videos themselves. Every image and video is visually analysed on upload, and your query is matched against what is actually in the frame. Type “a cat sitting in the corner of a dark room” and you will find exactly that — including clips where the cat was an incidental detail nobody would ever have keyworded.
How the visual index works For the curious — nothing here is needed to use the site.
What actually happens when I upload an asset?
Take an image: the picture itself is looked at — the actual pixels, not the file name or anything typed about it. What comes out is a long list of numbers — about 1,500 of them; the technical term is an embedding — describing what is in the frame: the objects, the setting, the lighting, the mood, roughly where things sit. Nothing you write is part of it. Video goes through the same process at a larger scale, so a clip is findable by its content just like a still.
And when I type a search?
Your sentence is turned into a list of numbers the same way. Think of both lists as coordinates: every asset is a point in one huge shared space, pictures that show similar things land near each other, and your query becomes a point in the same space. Results are simply its nearest neighbours — which is why a longer, more specific query works better here than on a keyword site, where extra words usually mean fewer results. Every extra detail you give moves the target, and the closest pictures move with it.
Every asset is a point, and similar content clusters. Your query becomes a point in the same space — the ringed dots beside it are its nearest neighbours, and they are the results. Why does that find things keywords miss?
Because nobody had to think of the word first. A contributor keywording a city street writes “street”, “traffic”, “evening” — not “a red umbrella leaning against a lamppost”. The umbrella is in the picture either way, so it is in the numbers either way, and a search for it finds the shot. Keyword search can only ever return what someone remembered to type.
How is “closest” measured?
By the angle between the two lists of numbers — a measure called cosine similarity. Each list can be read as a direction, and two directions pointing the same way describe the same thing. The smaller the angle between your query and a picture, the better the match.
Cosine similarity: each description is a direction from the same origin. The smaller the angle between your query and an asset, the stronger the match. Why does it understand other languages and typos?
Because it works on meaning, not on the letters. “Sunset”, “coucher de soleil” and “puesta de sol” describe the same thing, so they land in the same place and match the same pictures — and “panter” lands close enough to “panther” to still find them. No translation step, no spell-checker, no synonym list anyone has to maintain.
What are the limits?
It is very good at what a picture shows — famous landmarks and well-known places included, since those are recognisable on sight — and poor at what only a caption could tell you: which tool made it, when it was made, or the name of somewhere that is not visually distinctive. That is what the filters and tags are for, and they match literally. It also ranks rather than judges: for an unusual query the closest picture may still not be very close, which is why results say “top matches”. The same machinery powers find similar on every asset page — it starts from that asset’s own point instead of your words, and shows you its nearest neighbours.
- So keywords don’t matter at all?
- Not for ranking — nobody can keyword-stuff their way above better content, and details too small to describe are still findable. Where exact terms matter — a specific tool, a city, a creator — the filters handle that, matching tags literally. Titles and descriptions still work hard outside the site: they are what Google and AI assistants read when deciding what to show a searcher.
- Can I search in my own language? What about typos?
- Yes to both. Search understands meaning, not spelling — “coucher de soleil”, “puesta de sol” and “sunset” return the same results, and a misspelled “panter” still finds panthers.
- What does “Top matches” mean?
- Results are ranked best-first. It is an honest ranking, not a guarantee: for an unusual query the best available match may still be a loose one, which is why we say “top matches” rather than pretending every result is exact.
Buying
- What exactly am I buying?
- A commercial usage license — worldwide, royalty-free, perpetual — plus the original full-resolution file, backed by the creator’s warranties. The full terms are in the Content License.
- Can I use assets in commercial work?
- Yes — ads, films, social media, websites, presentations, and client work are all covered. The main limits: don’t resell or redistribute an asset as an asset, don’t use it to train AI models, and don’t use it as a logo or trademark.
- Can I put assets on merchandise I sell?
- Yes — physical products like prints, posters, apparel, and print-on-demand goods are covered by the standard license, at no extra cost. What’s not allowed is reselling the digital file itself.
- Do I own the copyright to what I buy?
- No — and nobody can sell you one: purely AI-generated work generally has no copyright under current US law. What you buy is a documented, contractual right to use the file, which is what stock buyers actually need.
- Can someone else buy the same asset?
- Yes. Licenses are non-exclusive — the same asset can be licensed to multiple buyers, like on any stock marketplace.
- Why are all sales final?
- Because you preview the complete content — full length, watermarked — before you pay. The file you receive is the same content in its original, typically higher, quality. Where the law grants a refund right, it applies. EU/UK buyers consent at checkout to immediate delivery, which ends the 14-day withdrawal right, as the law allows; rights that cannot be waived still apply.
- What are credits?
- Prepaid balance for small purchases. Direct card checkout has a $15 minimum, so smaller totals are paid with credits, sold in packs of $15–$100 at 1¢ each. Credits never expire and are non-refundable except as required by law.
- Do I need a subscription?
- No, and there is none to buy. You pay per asset at the price its creator set — from $0.10 an image and $1 a video — through a cart of $15 or more, or with credits you buy once and spend whenever you like. Nothing renews and nothing expires.
- Why not just generate it myself?
- Sometimes you can — for video, often you cannot. What you cannot generate either way is the license and the warranty: a commercial license, and the creator’s signed warranty that the work is theirs to sell, on file for the day a client asks. Count the prompting, the iterations and the compute, and a finished asset from $0.10 usually costs less than the hours. And if the shot is nearly right, the person who made it is one message away for a custom version.
- Do you offer legal indemnification?
- No. Some large agencies bundle an indemnity against copyright claims into their higher tiers; we do not sell insurance. What you get is a documented commercial license, the creator’s enforceable warranty that the work is theirs and AI-generated, and a catalog where every asset is labeled AI, so there is no question of undisclosed AI content. If a project needs indemnification, a large agency is the right tool for that project.
- Can I get custom work made?
- Yes, directly with the creator. Every creator page and every asset has a Contact or Hire button — the hire button where the creator has switched it on, with a rate if they chose to show one. You message them through the site, describe what you need, and agree the price and the delivery between you. The site makes the introduction; the work and the payment are arranged between the two of you.
Selling
- How much do I earn?
- You earn 65% of every sale — well above the 15–40% typical of the major stock agencies. The platform keeps 30%, and a flat 5% processing fee covers payment costs. When you withdraw, Stripe’s fee for sending the money comes off the payout — see “What does a payout cost?” below.
- When do I get paid?
- Monthly, once your balance reaches the payout threshold you choose. It starts at $5, the lowest allowed, and you can raise it any time — a higher threshold keeps more of each payout (see the next answer). Each sale’s earnings are held for at least 60 days first as fraud and chargeback protection.
- What does a payout cost?
- Stripe charges a fee each time it sends you money; we add no fee of our own. It is 25¢ plus 0.25% of the payout, plus $2 in any month you are paid — a month with no payout costs nothing. Because most of that is fixed, a small payout loses a lot to it and a large one very little: a $5 payout pays $2.27 in fees (45% of it), a $100 payout pays $2.50 (2.5%). Your earnings simply wait until they reach your threshold, so a higher threshold means fewer, larger payouts and more of the money kept. You can change it any time on your dashboard, under Earnings → Payout setup.
Payout Fee You receive Fee share $5 $2.27 $2.73 45% $25 $2.32 $22.68 9.3% $50 $2.38 $47.62 4.8% $100 $2.50 $97.50 2.5% - Do I need to set up payouts before I can sell?
- No. You can upload and sell immediately; earnings accrue as a balance we hold for you. You complete a one-time Stripe onboarding (identity and tax details) when you want to withdraw.
- What can I upload?
- AI-generated images and video, from any tool, that you have the right to sell and that meet the quality bar. The full rules are in the Creator Agreement, which you sign before your first upload.
- What are the technical limits?
- Images: JPEG, PNG or WebP, at least 1024px on the short side, up to 25 MB. Video: MP4, MOV or WebM, at least 1080p, 3 to 60 seconds, up to 500 MB, and under 210 Mbps — a video above that cannot be processed for playback, and 4K at 60 to 100 Mbps is plenty. An over-limit file is refused when you pick it, before anything uploads.
- Can I sell the same content elsewhere?
- Yes. Your license to us is non-exclusive — you keep your rights and can sell the same work on any other platform.
- What happens to uploads I leave unpublished?
- Unpublished uploads — drafts you never published, and uploads you took down yourself — and rejected uploads you do not appeal are removed after 30 days without any change to them. We email you a week before, and your dashboard shows the date on each one. Publishing, editing, appealing or deleting an upload yourself keeps it out of that — anything you touch is kept. Work held for review or under appeal, and anything that has ever been sold, is never removed this way.
Metadata in your files
- Do you read the EXIF/IPTC/XMP metadata in my files?
- Yes, in your browser, the moment you drop a file — before anything uploads. Anything we recognise becomes a tag you can edit or delete before you press Upload, and any title or description in the file fills those boxes for you. Nothing is sent while you are still deciding.
- Which fields do you read, and what do they become?
-
Where several fields could supply the same thing, the first one present wins — they are usually copies of each other rather than different text.
Field in your file Standard Becomes Tags — added automatically, editable before you upload CreatorToolXMP tool:SoftwareEXIF tool:MakeEXIF tool:(last resort)City IPTC 2:90 city:Sub-location IPTC 2:92 location:Country IPTC 2:101 country:Your title — first match wins Object Name IPTC 2:05 1st choice Headline IPTC 2:105 2nd XPTitleEXIF 3rd dc:titleXMP 4th Your description — first match wins Caption / Abstract IPTC 2:120 1st choice ImageDescriptionEXIF 2nd dc:descriptionXMP 3rd XPSubjectEXIF 4th The standards themselves, if you want to go and read them: EXIF (CIPA DC-008), IPTC Photo Metadata, XMP.
- What do you never read?
- Anything not in the table above, and three things in particular: GPS coordinates, device and lens serial numbers, and owner or author names. Those are never parsed and never stored. They are, however, still inside the file itself — so if you would rather they did not travel with your work at all, strip them before you export.
- Why do the title and description in my file not show on the listing?
- They are kept as your wording, but a listing shows the title and description we generate from the asset itself. That keeps every listing consistent and written for browsing. You can switch any single asset over to your own wording on its edit page, at any time.
- Why were my keywords not turned into tags? Will my work still be found?
- Yes — and this is the part worth reading if you come from traditional stock. Keywords are not how work is found here. Search looks at the pictures themselves, so an image is findable by anything actually in the frame, whether or not anyone thought to type the word. Your keywords were not lost so much as not needed. How search works →
Tags do a different job: they are the filters down the side of search results —tool:,city:,model:— so they arekey:valuepairs and capped at ten, or they stop being useful to filter on. A stock keyword list is thirty-odd bare words with no key, and guessing one for each would be wrong constantly. So we count them and tell you, rather than guess or quietly drop them. - What if you have never heard of the tool I used?
- You still get the tag. We tidy the value — dropping version numbers, so
Topaz Photo AI 3.1.2andTopaz Photo AI 3.2stay one filter rather than two — and for tools we already know we use our own spelling, so everyone filtering for Photoshop finds the same work. Either way it is a suggestion: change it or delete it before you upload. The one exception is video, where the file usually names the encoder rather than anything you chose — we ignore those instead of tagging half the library with something meaningless. - Which file formats carry metadata?
- In practice, JPEG carries it and PNG and WebP usually carry nothing at all — that is a property of those formats, not a choice we made. Video is read too, though far fewer files have anything to read. If nothing is found, no tags appear and nothing is wrong.
- I am uploading fifty files. Do I have to price them one at a time?
-
No. You can set one price for the whole batch on the upload screen, or a different price per file. And if you would rather prepare everything in advance, you can drop in a small text file that gives each upload its own price, title, description and tags — useful when you are exporting hundreds of assets from a script.
It is the only way to set prices in bulk: prices are ours, so no photo-metadata standard has a field for them. Anything you type on the upload screen still wins over it, and it wins over what was already inside your file.
How to write that file Only worth reading if you export in bulk.
Name it
metadata.jsonand drop it in alongside your files. It maps each file name to whatever you want to set — every field is optional:{ "sunset-01.jpg": { "title": "Golden hour over the Serengeti", "description": "Warm light across the savanna.", "tags": ["model:Midjourney", "city:Serengeti"], "price": 250 } }priceis in cents, so250means $2.50. Files are matched by name, by name without the extension, or by a leading number — so"15"matches015-sunset.jpg.
Trust & moderation
- How is content moderated?
- Every upload is screened against our content standards before publication, the community can report published content, and violations earn strikes — three strikes closes the account permanently.
- What if content infringes my copyright?
- Email a complete DMCA notice to support@agentsofai.art. The requirements and the process are in the DMCA Policy.
Have a question that isn't answered here? Email support@agentsofai.art or ask in our Discord community.