01Collectors
Key facts
- Pay is USD 5–15 per hour of accepted footage. It varies by project and is not guaranteed, and you see the exact rate before you accept.site FAQ; Data Collector Agreement §2.1
- Review usually takes about a week (contract limit 30 days). After acceptance you're usually paid within 7 days, in US dollars through Wise; the contract limit is 60 days.Data Collector Agreement §2.1, §3
- You don't need to speak English: instructions and support are available in Spanish and Portuguese.TaskCapture
- There is no fee to apply, and you never pay for equipment: the project provides it, or the task needs no special equipment.TaskCapture
- Every identifiable person on camera must sign a release, and commercial sites need written permission from the owner or manager.Privacy Policy §2
- TaskCapture collectors must be 18 or older.Privacy Policy §10
What makes footage useful for AI?
An AI model learns from examples. If it's going to learn how people load a dishwasher or pack an order, it needs many recordings of real people doing that, in real places. For the basics, read what AI training data is.
A recording is worth learning from when three things hold. The first is point of view. Video filmed from the worker's own eyes, called egocentric or first-person video, shows the hands, the tools and the order of the steps. Research teams build large collections of it: the Ego4D paper describes 3,670 hours of daily-life video "captured by 931 unique camera wearers from 74 worldwide locations and 9 different countries". The second is completeness. If a task stops halfway or your hands leave the frame, the model sees only part of the job. The third is a clear record of how the footage was made. The authors of Datasheets for Datasets propose that every dataset come with a datasheet that "documents its motivation, composition, collection process, recommended uses". That's why projects ask who recorded what, where, and with whose permission.
Every TaskCapture project puts these needs in writing, as a list of requirements and equipment you see before you accept. Your footage also carries metadata: the device, timestamps, and location or GPS data where the recording embeds it.
What do you wear?
For physical work like cooking, a warehouse shift or a repair, you wear a small camera on a head strap or a cap. It films what you see as you work and leaves your hands free. Researchers use the same setup: according to its research paper, the EPIC-KITCHENS-100 collection recorded unscripted kitchen activity in 45 environments "using head-mounted cameras".
For gaming, nothing goes on your head. A gaming project records the game screen and every keyboard, mouse or controller input, in sync. It doesn't collect your voice, chat, gamer handles or data from the game itself. Researchers pair screen and inputs the same way: OpenAI's Video PreTraining study trained an agent on gameplay video paired with keyboard and mouse actions. Apps that pay you to play are something else, and the guide to getting paid to record gameplay explains the difference.
Residential, commercial or gaming: what changes
TaskCapture collects three kinds of footage. What you wear, what ends up in frame and whose permission you need all depend on which kind a project asks for.
| Residential | Commercial | Gaming | |
|---|---|---|---|
| Where | Your own home | Your workplace, during a normal shift | Your gaming setup |
| What you wear | A small camera on a head strap or cap | A small camera on a head strap or cap | Nothing: the screen and inputs are recorded |
| What is in frame | Your hands and the chore: cooking, dishes, laundry, cleaning | Your hands and the job: prep lines, shelves, repairs | The game screen, with every keyboard, mouse or controller input in sync |
| Permissions | A signed release from anyone identifiable who appears; a person under 18 only with a parent's or guardian's release | Written permission from the owner or manager, plus a release from every identifiable person | No camera, voice, chat or gamer handles are collected |
Each project says which kind it needs, and the requirements for that kind, before you accept.
The types of recording work pages describe one activity at a time, such as a shop floor, a workshop or a home kitchen.
What does a good recording show?
A good recording meets the project's own requirements, because those are what reviewers check your footage against. A few habits help on top of that:
- Record the whole task. Start before the first step and stop after the last one. A recipe without the final plating, or a shelf half restocked, leaves the ending out.
- Keep your hands and the work in frame. Look at what you're doing, as you naturally would. If you turn away to talk, the camera turns too.
- Work at your normal pace. The project wants to see how the work happens on an ordinary day.
- Keep private things out of frame: screens, documents and anyone who hasn't signed a release.
For framing, light and a checklist to run before you press record, see the guide to recording first-person video of your work. For the review side, read what gets footage accepted. Research teams count permission as part of a good recording too. The EPIC-KITCHENS acquisition guidelines say "The participant should have signed the attached consent form, before recording."
If anything in the project instructions is unclear, ask before you record.
What you must never record
Some footage can never be used, however good it looks. Never record:
- a person who has not signed a release, if they can be identified;
- a person under 18, unless a parent or guardian has signed a release (otherwise the footage is rejected and deleted);
- a workplace, shop, restaurant or other commercial site without written permission from its owner or manager.
These rules protect the people around you. We also limit what happens to the footage afterwards. TaskCapture doesn't run face or voice recognition, and footage reaches buyers only after quality review. Datasets never include collectors' names, contact or payment data, and buyers are contractually barred from identifying anyone. Buyers may use the footage only to develop, train, evaluate and improve AI and machine-learning models. The full rules are in the Privacy Policy.
Researchers think about bystanders too. The EgoBlur paper describes a protocol in which recorded video "is processed by an AI anonymization model" to protect the privacy of people who happen to be filmed. Before you record anyone at work, read recording at work: privacy and consent.
How review and payment work
We arrange each project with you directly by email, and there's no account or app to set up today. When a project fits where you work, it goes like this:
- 01
Project details
We email you the project, with its requirements, its equipment and the pay, before you accept.
- 02
You accept
You see the exact rate before you accept, and the pay is fixed once you do.
- 03
You record
Record the task the way the project describes, with the releases and permissions it needs.
- 04
You hand it over
The project details tell you how to send your footage to TaskCapture.
- 05
Review
People and automated checks review your footage. Review usually takes about a week, with a contractual limit of 30 days.
- 06
Revision, if needed
If something needs fixing, you get a revision request and 5 days to send a corrected version.
- 07
Payment
You're paid in US dollars through Wise, usually within 7 days of acceptance. The contractual limit is 60 days after acceptance.
On a project you work as an independent contractor under the Data Collector Agreement. Sections 2 and 3 cover pay and review.
A worked example: one chore at home, revised once
Ana is an invented person. Every rule in her story comes from the sections above.
Ana lives in a shared apartment and applies with her country and "home". When a residential project fits, we email her the details: the task (recording herself doing the dishes), the project's requirements and equipment, and the rate per hour of accepted footage. She reads them, checks the rate and accepts. From then on, the rate is fixed.
Her roommate sometimes walks through the kitchen, so before recording, Ana asks her to sign a release. Over two evenings she records the dishes with the head camera, from the first plate to the dried rack, and sends the footage the way the project details describe.
About a week later she gets a revision request: in one session her hands left the frame while she reached into a high cupboard. She sends a corrected recording within the 5 days allowed, and the footage is accepted. She's usually paid within 7 days of acceptance, in US dollars, at the rate she accepted, and never later than 60 days after acceptance.
Text AI training jobs vs video data collection
The two kinds of work share a name and not much else. For a longer comparison, read data collection compared with annotation jobs.
| Typical text AI training job | Video data collection | |
|---|---|---|
| What you do | Write, rate or label text and AI answers | Record a real task you already do, from your own point of view |
| Where | At a computer | Where the task happens: your home, your workplace or your gaming setup |
| Skills that matter | Writing, or expertise in a subject | Doing the task well and following the project's instructions |
| How work is checked | Reviewers grade your answers | People and automated checks review your footage against the project's requirements |
The first column shows how text-based AI training work is usually advertised, and the details vary from company to company. TaskCapture collector work doesn't involve rating chats or writing.
Common mistakes
Before you record, check the project's requirements. Then watch for these mistakes:
- Wait until you've accepted before you record. The project details set the task, the equipment and the rate, so record only once you've agreed to them. See how collector projects work.
- Keep the task in view. If your hands or the work leave the frame, or the recording stops early, part of the job is missing. What gets footage accepted has more tips to check against your project instructions.
- Get the releases first. Every identifiable person needs one, and a workplace needs the owner's or manager's written permission before you start. Recording at work: privacy and consent has a permission checklist.
- Watch for private information. Screens, documents and mail end up in frame more often than you'd think, and the tips for recording first-person video show how to keep them out.
- Check your email after you hand footage over. A revision request gives you 5 days to send a corrected version.
Is it legit? Five checks
Fake AI job offers are common, so check any company before you apply. The US Federal Trade Commission's job scams guidance puts it plainly: "Honest employers, including the federal government, will never ask you to pay to get a job." Run these five checks on TaskCapture or anyone else offering AI training work:
- You should see what a project pays before you commit to it. TaskCapture shows you the exact rate before you accept, and the pay is fixed once you accept.
- The terms should be in writing. TaskCapture collectors work as independent contractors under the Data Collector Agreement.
- The money should reach you through a real payment route. TaskCapture pays in US dollars through Wise, usually within 7 days of acceptance; the contractual limit is 60 days after acceptance.
- A serious data company tells you who may appear on camera and what permission you need.
- You shouldn't have to pay anything. There is no fee to apply to TaskCapture, and you never pay for equipment.
For the warning signs of task scams and fake offers, read how to spot AI training job scams.
Questions
How much does it pay?
Pay is USD 5–15 per hour of accepted footage. It varies by project and is not guaranteed. You see the exact rate before you accept, and it's fixed once you do. Collectors are paid in US dollars (USD).
When do I get paid?
Review usually takes about a week, with a contractual limit of 30 days. Once your footage is accepted, you're usually paid within 7 days, in US dollars through Wise. The contractual limit is 60 days after acceptance.
Can I do this without any AI experience?
Yes. There is no fee to apply, and you don't need any coding or AI background. What counts is knowing the task you record, like a kitchen prep shift or a game you play, and following the project's requirements. You don't need to speak English either: instructions and support are available in Spanish and Portuguese.
Do I need my own camera?
No. You never pay for equipment: the project provides it, or the task needs no special equipment. Each project lists its requirements and equipment before you accept.
Can I do it from home?
Some of it. Residential projects happen in your own home, and gaming projects record your game screen and inputs. Commercial projects happen at your workplace, with written permission from the owner or manager. The footage has to show a real task in the place where it normally happens.
What happens if my footage needs changes?
People and automated checks review your footage. If something needs fixing, you get a revision request and 5 days to send a corrected version.
Do I get paid for rejected footage?
No. Rejected footage is not paid, because pay is per hour of accepted footage. If you get a revision request, use it to fix the problem.
Who sees my footage?
Footage reaches buyers only after quality review. Datasets never include collectors' names, contact or payment data, and buyers are contractually barred from identifying anyone.
Do I need an account or an app?
No. We arrange projects directly by email, and there's no account or app today. The project details say how to hand over your footage.
To see how other services' published terms compare, with the same questions asked of each, read our recording-platform comparison.