What should I look for in a stock footage metadata generator?
Eight checks for a stock footage metadata generator: per-agency sets, XMP or CSV, Latin names, Getty's vocabulary, review. Tools compared, tested 2026.

A good stock footage metadata generator writes a different title and keyword set for each agency, starts from what you actually filmed, checks facts such as species names, and delivers the metadata the way each agency reads it. Only Adobe Stock and Pond5 read keywords embedded in a video file (measured September 2026).
These numbers come from Seafootage, the catalogue of Hervé Berthou, whose son Thomas builds Rushlot. One account, not a market survey.
I build Rushlot, which does this job among others, so I am not neutral. The eight checks below come from what we measured on that account and from each agency's own rules, and they apply to any tool, ours included. A value marked measured was read on a live contributor account. The tool facts come from each vendor's own page, on the date shown.
What should you check in a metadata generator?
Eight things. Word count is not one of them.
| Check | Why it matters | How to test it |
|---|---|---|
| 1. Does it handle video? | Many tools are built for photos | Read its page, then feed it a real clip |
| 2. Does it start from your title? | You know what is in frame | Give it a precise title and see if it stays true |
| 3. One set per agency? | Four agencies, four shapes | Compare the Adobe and Shutterstock outputs |
| 4. Does it deliver the right way? | Two agencies ignore the file | Read the file back, open the CSV |
| 5. Does it keep the order? | Adobe weighs the first 10 most | Look at the first 10 keywords |
| 6. Does it check facts? | A wrong species is a false claim | Include a Latin name with a typo |
| 7. Does it know Getty's vocabulary? | Other words stay red | Count the green words in ESP |
| 8. Do you review before it sends? | Fixing later is slow | Find the point where you can still change a word |
Does it really handle video?
Read the tool's own page and look for the word "video" next to the word "file". Here is what six vendors said, on the date we read them.
| Tool | Read on | Video | Metadata inside the video file | Output | Free allowance and first price |
|---|---|---|---|---|---|
| MyKeyworder | 2026-10-28 | Its pages do not mention video | Its pages do not say | Lightroom plugin, web | Free, donation-funded |
| PhotoKeyworder | 2026-10-29 | Yes, several frames per clip | "Not embedded directly into video files" | CSV for video | 20 credits; 8.99 € per 30 days before VAT for 133 videos |
| SendStock AI | 2026-11-08 | Yes | Says it embeds in MP4 and MOV | File, CSV, FTP or SFTP upload | 100 requests; $6.99 a month |
| ClipMeta | 2026-11-08 | Yes, built for video; vision AI reads the frames | Its pages do not mention XMP | CSV per platform, FTP upload to Blackbox | 40 clips in the first 6 days, then 2 a day; from $9 a month |
| MetaMule | 2026-11-08 | Yes, 25 to 45 keywords per file | Its page does not mention it | One CSV per marketplace, FTP upload to three | 25 assets; from $19 a month |
| Rushlot | 2026-11-08 | Built for stock video | Yes, for Adobe Stock and Pond5 | XMP for those two, CSV for Shutterstock, Pond5 and Getty, upload to four | 20 clips at opening, no card; 19 € a month. Mac and Windows beta, waiting list |
Compare Rushlot with each of them one at a time: Rushlot vs MetaMule, Rushlot vs ClipMeta, Rushlot vs SendStock AI, Rushlot vs PhotoKeyworder, Rushlot vs MyKeyworder, or all comparisons.
Should a generator start from the picture or from your title?
From your title, with the picture as a second opinion. You know the species, the place and the behaviour. A model looking at frames sees a fish. It can name a plausible one that is wrong, and it cannot see where you were.
So give the tool a precise title and watch what it does with it. "Northern pike hiding in watermilfoil, Lake Salagou" should stay exactly that. If the output changes the species or adds a place you did not name, the tool is writing fiction. Rushlot uses the frames only for what a title cannot say, such as the shot type, the camera movement and the light, and raises a flag when they contradict you.
Does it write a separate set for each agency?
It has to, because the fields are different. Copying one list into four portals fails on the first one that disagrees.
| Agency | Title or description | Keywords | Order |
|---|---|---|---|
| Adobe Stock | Title up to 200 characters, under 70 advised; no description for video | Up to 49; 15 to 35 "often plenty" | First 10 weigh most |
| Shutterstock | One description, 200 characters in the CSV dialog measured, plus 2 categories | 7 to 50 | Stored alphabetically measured |
| Pond5 | Title 80 characters, description 500 | Up to 50; 10 to 20 advised | Kept from the file measured |
| Getty / iStock | Title and description | 5 to 50, from a controlled vocabulary | Nothing official |
Every limit has its source and date in how many keywords Adobe Stock allows. The free stock video title checker tests one title against three agencies at once.
Does it deliver the metadata where each agency looks?
This check decides whether your work reaches the agency at all. Adobe Stock and Pond5 read the title and keywords embedded in an MP4 as XMP. Shutterstock's panel showed 0 of 50 keywords for a file carrying them (2026-09-11) measured, and Getty's ESP read nothing either (2026-09-08) measured.
So "embeds metadata in your file" solves half of the problem. SendStock AI's page says embedded metadata "is read automatically by stock agencies". On our tests, that held for two agencies out of four. For Shutterstock and Getty you need a CSV or the form. For Pond5, watch one trap: a CSV keyword cell replaces the embedded list instead of adding to it measured.
If a tool writes into the file, check that it worked. Open your clip in the free reader below, in your browser. The file never leaves your computer.
And remember that any ffmpeg step after the write erases the XMP without an error. The metadata has to go in last.
Does it check facts, or only fill the field?
A generator that only fills fields will give you confident mistakes. These are the ones that happened to our own drafts, on the account of Hervé Berthou, my father:
- Language. A French keyword, tonne de chasse, reached Shutterstock on one September 2026 batch.
- Markup. One draft opened a description with an invented tag ending in
>. Shutterstock refused the row, Adobe used it as the title, and Pond5 showed the clip with no title. - Species. A misspelt or invented Latin name returns HTTP 204 from WoRMS (measured 2026-09-03). See stock footage keywords scientific names.
Rushlot now checks every Latin name, the language and the markup before anything leaves. Whatever you use, the agencies judge the words, not the tool: the rules are in what the agencies penalize.
Does it know Getty's vocabulary?
Ask, because Getty is the agency where a normal list fails most. Getty keeps only words from its controlled vocabulary. On one clip we sent 46 keywords: 19 were accepted, and 27 stayed in red, doing nothing for search (2026-09-11) measured.
Over 4,003 distinct keywords that Rushlot tried in ESP between 10 September and 23 October 2026, 1,263 (31.6%) were accepted and 2,740 (68.4%) came back red. That share is high because the tool tests candidates on purpose. It does not mean two-thirds of a normal list fails. What it shows is that a generator with no knowledge of Getty will leave you to fix the red words by hand.
Shutterstock can surprise you too: on 9 of 15 clips of one lot, it kept 37 to 48 of the 49 keywords sent, and the words it dropped were not a length cut.
Where does the time go, even with AI?
In everything after the draft, as far as I can tell. My father used ChatGPT to write his metadata before Rushlot existed. Putting 60 videos on four marketplaces still took him about 40 hours. With Rushlot it takes him about 1 hour. That is his own estimate, and he is my father, so read it as one contributor's experience.
Will better metadata make your clips sell?
We cannot prove it, and you should distrust anyone who says they can. We found no agency that publishes what metadata is worth. On Seafootage, with sales read between 21 and 27 October 2026, no sale had been matched to the 167 clips in Rushlot's database that were uploaded at least once, from 4 September to 25 October. Another 33 clips are not in that database, one Adobe sale may come from another profile (unverified), and 131 of the 193 clips the database holds, uploaded or not, are architecture. It is too early to conclude anything. The most a generator can honestly promise is less time and fewer mistakes.
How do you test a generator before paying?
Run it on five hard clips, not on your best one:
- A precise species. Does the Latin name validate?
- A vague title. Does it invent details?
- A place you named. Does it keep the place, and only that place?
- A landmark or a person. Does it warn you about releases?
- A clip with a typo in its title. Does it copy the mistake or flag it?
Then read the first 10 keywords, count the repeated stems with the stock video keyword cleaner, and check the output against each agency before you upload. If you cannot change a word before it is sent, stop there.
What have we not measured?
- How accurate any of these tools is on the same set of clips. We have not run them side by side.
- Whether a list from one tool sells better than a list from another.
- Whether any agency detects AI-written metadata as such. None says it does.
Once your metadata is right, the next question is the send: how to automate stock footage uploads. Rushlot runs on Mac and Windows: the download page has the waiting list.
Frequently asked questions
It is software that writes the title, description and keywords a stock agency needs for a video clip, usually with AI. Some only suggest words to copy; others write a CSV, embed the metadata in the file, or upload it too.
It can draft it. Check the output against each agency's limits and Getty's vocabulary yourself, and verify any species name. Hervé Berthou, whose son builds Rushlot, used ChatGPT and still spent about 40 hours on 60 videos for four marketplaces, by his own estimate.
Yes, in practice. Adobe has a title and no description for video and weighs the first 10 keywords most. Shutterstock uses a description, 7 to 50 keywords and categories, Pond5 a title of 80 characters at most, and Getty keeps only words from its vocabulary.
It varies. PhotoKeyworder analyses several frames but says metadata is 'not embedded directly into video files'. MyKeyworder's pages do not mention video. SendStock AI says it embeds metadata in MP4 and MOV. Read the tool's own page and test it on a real clip.
None of the four main agencies forbids or allows them in the pages we read in October 2026. They penalize what the words say: irrelevant terms, repetition, brands and spam.
Nobody can promise that, and we have no proof of it. Good metadata helps a buyer find a clip, but we found no agency that publishes a figure. On one account, Seafootage, no sale had been matched to the 167 uploaded clips in Rushlot's database when sales were read in October 2026, which is too early to conclude anything.
As many as the clip supports, not as many as the field allows. Adobe calls 15 to 35 often plenty, Pond5 advises 10 to 20, and Getty asks for quality over quantity.
Sources
- PhotoKeyworder home and pricing pages, read 2026-10-29
- MyKeyworder home, Updates and Lightroom pages, read 2026-10-28
- SendStock AI home, features and pricing pages, read 2026-11-08
- MetaMule home page, read 2026-11-08
- Adobe help, Tips for effective titles and keywords (updated 2026-08-18), read 2026-10-01
- Adobe Artist Hub, Maximize Metadata part 3 (undated), read 2026-10-01
- Shutterstock help, Contextual Metadata (updated 2026-02-26), read 2026-10-27
- Pond5, Tagging & Pricing Your Files (modified 2026-01-29), read 2026-10-27
- Getty Images, Content Partners guide, article 6064, read 2026-10-01
- WoRMS REST web service, AphiaRecordsByName
- Measured on live accounts: embedded XMP read by Adobe (2026-09-07) and Pond5 (2026-09-11), not by Getty ESP (2026-09-08) or Shutterstock (2026-09-11)
- Measured on the live Seafootage accounts: Getty keyword verdicts 2026-09-10 to 2026-10-23; Shutterstock keywords kept on one lot, September 2026; sales matched against Rushlot uploads, read 2026-10-27
- Hervé Berthou (Seafootage), the author's father: 60 videos on four marketplaces, about 40 hours before and about 1 hour with Rushlot (his own account, 2026-10-28)




