Contributor tools split into two groups, and the split is the whole answer. Desktop tools like Xpiks and Microstockr do their work on your machine, so your files never move. Cloud services — Microstock+, PhotoTag.ai, Wirestock, Blackbox — require you to upload before anything happens. Once a file is uploaded, where it is processed, how long it is kept and whether it reaches an AI model depend entirely on what that vendor has written down.
The question is whether the file leaves your machine
Every other privacy question is downstream of that one. If a tool runs locally and reads the image from your drive, there is no retention policy to worry about, no processing jurisdiction, no subprocessor list and no training question — because nothing was transmitted. The moment a file is uploaded, all four of those become live, and the only honest way to answer them is to read what the vendor committed to in writing.
That turns out to be harder than it sounds, because "desktop tool" and "local processing" are not the same claim. A desktop application can still send your image to a server for the keywording step. Xpiks does exactly that through one of its own plugins, and its licence agreement separately describes a cloud upload path with a stated storage period. So the right unit of analysis is the feature you are about to use, not the product you installed.
Where six contributor tools process your files
Compares six third-party contributor tools on four things each vendor documents or declines to document: whether processing is local or cloud, any stated retention period, what the policy says about AI training, and whether a royalty share applies. Every cell comes from the vendor's own site, licence or privacy policy; unknown means the vendor does not state it.
What each vendor's own documentation says
Xpiks — local by default, with two documented exits
Xpiks is desktop software for macOS, Windows and Linux, and its licence agreement is unusually specific about the exceptions. Use its Cloud upload feature and you grant a limited right to store your files on the author's servers for up to one month; the agreement also states the author acquires no ownership of your files. Use the in-app Auto Keywording plugin and it uploads a significantly downscaled copy of the photo to the server and any involved vendor — that vendor is not named. A separate free web tool on the same site works differently again, sending your own API key, prompts and images directly to OpenAI with no other external services involved. Three features, three data paths, one product.
StockSubmitter and Microstock+ — a desktop app on a cloud platform
These are sister products from the same company, and the distinction matters. StockSubmitter is a Windows desktop application whose own page states that all passwords are stored on your computer only. Microstock+ describes itself as a cloud-based portfolio management platform that works on any operating system with a browser, and its documentation says you must first upload files to the M+ interface before sending them to agencies. The shared privacy policy says personal information is stored in the EEA and explicitly covers storing the credentials used to authenticate on agency sites on your behalf. Both statements are published by the same vendor, and they do not sit comfortably together — which is itself worth knowing before you decide which product you are actually using.
PhotoTag.ai — the most candid policy, and the least reassuring
PhotoTag.ai is a web service with an API and a Lightroom Classic plug-in, so processing happens on its servers. Its privacy link resolves to the operator's company-level privacy policy, and that document is refreshingly direct about its own limits. It says information may be stored or processed in the United States and other countries where its providers or users operate, and that it does not represent that data is stored in any particular country. On training it says submissions may or may not be used to train or improve provider models, and that provider practices and available controls vary and can change. Residual copies may remain in backups until those backups expire. The AI providers are named only as a category; the actual subprocessor list is reserved for customers with a data processing agreement.
Wirestock — cloud by design, with an opt-out you have to find
Wirestock is a managed pipeline: you upload, it submits onward, and it hosts your portfolio. Its FAQ states contributors receive 85% of royalties earned through partner marketplaces, with Wirestock taking 15%, that Extra Channels may exceed 15% but never more than 50%, and that you remain the sole owner of your content. The line to read twice is the Dataset Deals programme, which licenses your content to selected third parties and applies automatically unless you opt out in account settings. The FAQ does not say what those third parties do with the content, and the privacy policy does not address image storage, name a processing country, or mention AI training at all.
Blackbox and Microstockr — the two ends of the range
Blackbox is a video intermediary: footage goes up through its portal and Blackbox submits it to agencies on the creator's behalf, taking 15% of net sales, with free membership and no subscription fee, and you keep your copyright. Its published FAQ does not address retention, processing location or AI use. Microstockr is at the other end — an earnings tracker rather than a keywording tool, whose FAQ states it does not store your credentials on its servers, that they are encrypted and stored on your device, and that the only thing checked server-side is subscription status. It is the clearest local-data claim of any tool here, though whether its image-matching feature uploads files is not stated.
Rastock AI — cloud processing, delivery over SFTP, terms in the DPA
For the sake of a complete list: Rastock.ai sits in the cloud group. Files are processed on its side, metadata is generated and checked against each agency's rules, and delivery to 10+ agencies happens over FTP or SFTP. What it does commit to publicly is the commercial side — no revenue share, no claim of ownership over your files, and IPTC, CSV and XML export available so your metadata is portable out again. The specifics a careful reader should want — retention periods, subprocessors, processing locations — are in the data processing agreement and the security page rather than paraphrased here, because a paraphrase is not what you should be relying on for this question from any vendor, including this one.
The layer underneath: whose AI is reading your photo
Most keywording tools do not run their own vision model. They call someone else's, which adds a second hop your file takes. The major API providers are clearer about this than the tools that use them. OpenAI's developer documentation states that data sent to the OpenAI API is not used to train or improve OpenAI models unless you explicitly opt in, with abuse-monitoring logs retained up to 30 days by default. Google's Cloud Vision data-usage page says Google does not use the content you send to train and improve Cloud Vision features, and that for online requests the image data is processed in memory and not persisted to disk. Anthropic states that by default it will not use inputs or outputs from its commercial products, the API included.
One of the four is different, and it is the difference that should shape how you ask the question. Amazon's Rekognition FAQ states that the service may store and use image and video inputs to improve and develop the quality of Rekognition and other Amazon machine-learning services, and that you may opt out of that use — through an AWS Organizations opt-out policy, which is an account-level configuration the vendor applies, not something a tool's end user can set. So the default is not uniform across providers: three of the four say no training unless you opt in, and one says training unless the customer opts out.
Which produces the uncomfortable conclusion. Not one of the tools above names the AI provider it uses. Xpiks' plugin page says the server and any involved vendor. PhotoTag.ai's policy names a category of providers and puts the actual list behind a data processing agreement. That means a contributor cannot verify the chain end to end: you can read the provider guarantees, and you can read the tool's policy, but you cannot confirm which guarantee applies to your file. If the work is sensitive, that gap is the thing to plan around — the same reasoning that drives evaluating whether a tool is safe for unreleased work.
When a local tool is the right answer, and when cloud is fine
Use a local tool when the constraint is contractual rather than practical. Work under client NDA, unreleased commercial work, embargoed editorial, anything where a third party's data handling would breach an agreement you signed — for these the question is not whether a vendor is trustworthy but whether you are permitted to transmit the file at all. Xpiks with cloud features switched off, and a tracker like Microstockr for the reporting side, keep the files where your contract expects them.
Use a cloud tool when the constraint is volume. A local app cannot keyword four hundred files and then deliver them to eleven agencies while you sleep; that is what server-side pipelines are for, and the time saved is real. The decision then moves from "local or cloud" to "which cloud, on what terms" — which is where the royalty share question belongs, since a managed pipeline like Wirestock or Blackbox charges a percentage of earnings rather than a subscription. Whether that trade suits you depends on volume, and the tools that take no revenue share are the alternative side of that comparison.
The middle case is the most common and the least discussed: ordinary stock work, no NDA, where you would simply rather your archive not end up in a training set. Here you are not choosing on risk but on disclosure quality. A vendor that publishes a retention period, names its processing region and gives you a subprocessor list on request has told you something. A vendor whose policy says your files may or may not be used to train provider models has also told you something — just not what you wanted to hear.
Should this batch leave your machine?
Four questions that sort a batch into local-only processing, cloud processing with documented terms, or a managed pipeline that charges a royalty share. Based on the distinctions in this article: contractual restrictions on transmitting files, what a vendor documents about retention and processing, and whether you need delivery as well as metadata.
Five questions worth asking before you upload anything
Ask where files are processed, and in which country or region. Ask how long they are retained after a job finishes, and whether deletion is immediate or waits for a backup cycle. Ask whether the vendor's own policy excludes your content from model training, or merely declines to address it. Ask which AI subprocessors are used and whether the list is available. And ask what export you get on the way out, because metadata you cannot take with you is metadata you are renting.
A vendor that answers all five in writing is not necessarily safer than one that does not — but it is accountable, which is the only thing you can actually act on. The two follow-on questions are what happens when you leave a tool and take your metadata with you, and how to hand over agency access in the first place without handing over more than the job needs, which is covered in the safest way to give a tool access to your stock accounts.
If you want to test a cloud pipeline against your own material before committing anything, Rastock.ai runs a 14-day trial with no card required. Start with a batch you would not mind being wrong about, read the terms before you upload the batch you would, and keep an IPTC export of everything either way.
Frequently asked questions
Do stock keywording tools upload my photos to a server?
Most do. Cloud services such as Microstock+, PhotoTag.ai, Wirestock and Blackbox require an upload before they can do anything. Desktop tools such as Microstockr work on files where they sit. Xpiks is desktop software but has two documented paths that send data out: a cloud upload feature and an in-app keywording plugin that uploads a downscaled copy of the image.
Will my images be used to train AI models?
It depends on the tool, and most do not say. The major API providers are clearer than the tools built on them: OpenAI, Google Cloud Vision and Anthropic all state that API inputs are not used for training by default, while AWS Rekognition documents that inputs may be used for service improvement unless the customer applies an account-level opt-out policy. Since no contributor tool names its AI provider, that chain cannot be verified from the outside.
Which contributor tool keeps files entirely on my own machine?
Microstockr makes the clearest local-data claim, stating that credentials are not stored on its servers, are encrypted on the device, and that only subscription status is checked server-side — though it is an earnings tracker, not a keywording tool. Xpiks keeps files local if you avoid its cloud upload feature and its auto-keywording plugin, both of which transmit data by design.
How long do these services keep my uploaded files?
Only one publishes a period. Xpiks' licence agreement grants a limited right to store files on its servers for up to one month when cloud upload is used. Wirestock and Microstock+ tie retention to the life of your account. PhotoTag.ai's policy states retention depends on data type, service and provider, and that residual copies may remain in backups until those expire. Blackbox does not address it.
Does a managed pipeline mean giving up part of my royalties?
Generally yes, and the percentages are published. Wirestock's FAQ states contributors receive 85% of royalties from partner marketplaces, with 15% to Wirestock, and that Extra Channels may exceed 15% but never more than 50%. Blackbox states a 15% commission on net sales with free membership. Subscription tools such as StockSubmitter, Microstock+, Xpiks and Rastock AI charge a fee instead and take no share of earnings.
What should I ask a vendor before uploading sensitive work?
Five things: where files are processed and in which region, how long they are retained and whether deletion waits for a backup cycle, whether the policy actually excludes your content from model training rather than staying silent, which AI subprocessors are used and whether that list is available, and what metadata export you get if you leave. A written answer makes a vendor accountable, which is the part you can act on.