Buy AI training data, collected to your requirements

Existing datasets versus commissioned data

An existing dataset is quick to obtain, but it was collected for someone else's model. Its coverage, capture setup and licence are fixed. When the gap between what it contains and what your model needs matters, such as a missing activity, the wrong camera height or too few regions, more of the same data does not close it.

Commissioned data starts from your requirements. You choose what is recorded, how, by whom and where, and the licence is settled before collection begins.

What you can buy through GoKuna

Real-world media recorded on phones: video of everyday objects from two sides, ambient household sound recorded without voices, photographs of handwriting samples, first-person household clips filmed at head height, and long-form point-of-view footage from a head or chest mount.

Recruitment for first-person household clips is open worldwide; long-form footage is collected in North America to start. Any task can be limited to specific countries.

What a purchase includes

A brief turned into tasks with your capture instructions, clip duration limits, audio requirement, accepted file formats, maximum file size, number of submissions and country limits. Recruitment of adult contributors, who see those requirements before they commit. Human review of every submission against your requirements, with rejected work re-recorded while the task is open.

And the rights: contributors grant a licence for training and evaluating machine learning models, including on behalf of GoKuna's customers, and accepted recordings are kept as part of the delivered dataset.

Consent you can stand behind

Every contributor is an adult who has confirmed they are 18 or older and agreed to only submit media they are authorized to share. Each task records the consent terms that applied to it. Capture rules keep identifiable bystanders, private information and other people's property out of frame without permission, and contributors preview each recording before upload.

Contributors are paid for accepted work at a reward shown before they reserve the task, through Stripe, with no fee taken from them.

Writing the specification

Describe the data the way you would check it. What must be in every item: the subject, the angle, the duration, the sound. What must never be: faces, voices, screens, text. How items should vary across the collection: rooms, regions, objects, light. The more of your acceptance criteria you write down, the more closely review can hold every submission to them.

Include how you will use the data. It shapes the quote, and it lets GoKuna check that the licence contributors grant, for training and evaluating machine learning models, covers your purpose before anything is collected.

How buying works

Send a brief from the homepage form: your email and a description of the data, how you will use it, and any budget or timing you have in mind. GoKuna replies by email with a quote. Scope and price are agreed before any collection starts, so there is no commitment until you have seen the terms.

Once agreed, the collection runs through the same task flow contributors already use, and accepted submissions make up your dataset.

Questions to settle before you buy any training data

Wherever you buy, ask: were the people who produced this data adults who agreed to this use? Does the licence cover training and evaluation for your purpose? Who checked that each item matches the specification? Could bystanders or private information appear in it?

A commissioned GoKuna collection answers each of these in writing before the first recording: age and consent at sign-up, a stated licence, human review against your requirements, and capture rules on bystanders and private information.

Can I buy LLM training data from GoKuna?

GoKuna collects real-world media: video, audio and images. It suits multimodal and perception models. Describe your model in a brief and the reply will say what GoKuna can collect for it.

How much does custom training data cost?

It depends on the scope. Send a brief and GoKuna replies with a quote; price is agreed before collection starts.

Is the data licensed for commercial model training?

Contributors grant a licence for training and evaluating machine learning models, including on behalf of GoKuna's customers.

What file formats can I ask for?

Each task states the file formats it accepts, its maximum file size and, for video, the minimum and maximum clip length. Put the formats you need in the brief.

Do contributors know who is buying?

No. Contributors see the task, its requirements and its reward. Your organisation is not named to them.

Can I limit collection to certain countries?

Yes. Any task can be restricted to specific countries.