Pricing
Cost drivers
A single RGB clip costs less to capture and QA than an episode syncing ten streams — 4K video, depth, IMU, 3D hand and body pose, segmentation, and action labels — under 2 ms. More streams means more equipment, more QA passes, and more storage per hour.
You are billed for hours that clear the accept criteria in your spec, not hours submitted. A tighter spec with a higher reject rate on unfamiliar tasks means more raw capture per validated hour — which is a timeline and recruiting cost we absorb, not one passed to you as a line item.
Environments, tasks, and geographies we already have active contributor coverage for move faster and cheaper. A spec that asks for a condition, region, or task combination we have little standing coverage for takes longer to recruit and scope, which shows up in timeline more than unit price.
A median of 48 hours from signed spec to first validated batch where coverage is deep; three to four weeks where it is thin. Compressing that timeline where coverage is already thin usually means recruiting in parallel across more regions, which is a scope decision made explicit before you sign.
Why no price list
A price list implies a fixed unit of work, and custom training data rarely is one. Ten synchronized streams at 4K with hand and body pose is a materially different job from a single RGB clip; a task with deep existing coverage is a different job from one that needs recruitment in a new region. Publishing one number would either overcharge the simple case or undercharge the hard one.
Instead, every project starts from a written spec with explicit accept and reject criteria. The spec is what gets quoted — reach, stream count, validated hours, and timeline — before any capture begins, so the price maps to the actual work rather than a rate card.
Billing model
Validated hours: footage that has been captured, reviewed, and passed the accept criteria written into your spec. A take that fails QA is not billed — it is rejected and recaptured, at our cost, until it clears the bar or the spec is revisited.
That is the deliberate trade: you carry the risk of defining the spec correctly up front, and we carry the risk of executing it. It also means the reject rate you see on the program velocity chart is our cost to manage, not yours to absorb.
Scoping a budget
Start with a Spec Pilot: a small, fast first batch that proves the spec and accept criteria at low volume before you commit to scale. It surfaces where the reject rate is high, where coverage is thin, and whether the stream count you asked for is actually necessary for the model you are training.
From there, the levers to pull are the ones above: drop a stream you do not strictly need, widen the tolerance on an edge case, or phase geography instead of launching every region at once. Each of those moves cost and timeline together, and you see the trade-off before you approve it.
What moves cost
| Driver | What it changes |
|---|---|
| Stream count per episode | Equipment, QA passes, and storage cost per hour |
| Validated hours needed | Total billed hours; raw capture volume required to reach them |
| Coverage depth for your spec | Recruiting time and cost, more than unit price |
| Turnaround required | Parallel recruiting across regions when coverage is thin |
| Reject rate on the spec | Our recapture cost, not yours — billed hours stay validated-only |
Every collection runs through the same seven-stage end-to-end custom collection pipeline.
FAQ
Sources
Tell us the streams, volume, and timeline your model needs. We will scope a written spec and quote reach, turnaround, and price before any capture begins — starting with a Spec Pilot if you want proof first.