Data-Video Generalist (US-based)

$13–15/hrrate as listed

Contribute video-based training data for AI models — recording screen or camera workflows to spec, annotating clips, and writing precise descriptions of what happens on screen. micro1 lists this as a US-based, remote, hourly generalist role in the $13–15/hr range as observed on the platform.

Summary, eligibility guidance and screening preparation are prepared by Sidequest from the platform's posting. Check the source listing before applying.

What the work involves

This is data production work rather than model research. Day to day you can expect some mix of: capturing video to a written spec (screen recordings of software tasks, camera footage of physical procedures, or staged multi-step demonstrations), segmenting and labeling clips, writing frame-accurate or timestamped descriptions of actions and state changes, and reviewing other contributors' submissions against a rubric. Specs change often — a task that asked for one continuous take last week may ask for narrated step boundaries this week — so the practical skill is reading guidelines carefully and following them exactly rather than doing what seems reasonable.

Because the listing is a generalist one, batches are assigned as they come in and the subject matter varies: consumer apps, office tooling, everyday household or hands-on tasks, and occasionally domain-specific workflows you already know. Volume is not guaranteed and tends to arrive in waves tied to client projects.

What the platform screens for

Instruction fidelity. Whether you can restate a multi-clause spec and identify where an edge case falls outside it.
Descriptive precision. Whether your written account of a video is unambiguous, ordered, and free of inference the footage doesn't support.
Basic capture competence. Usable audio, stable framing, correct resolution and frame rate, clean file naming, no accidental PII or on-screen credentials.
Consistency under volume. micro1's screen is AI-led and typically includes an async recorded interview plus a short sample task; expect follow-up questions that push on specifics rather than accepting general claims.
US-based eligibility, since the listing is scoped that way and some batches require US-context footage.

Logistics

Fully remote and largely asynchronous, with deadlines per batch rather than fixed shifts. Contributors commonly report 10–25 hours a week when work is available, self-scheduled. You'll need your own recording setup — a computer capable of clean screen capture, a phone camera adequate for handheld footage, reliable upload bandwidth, and a quiet space if narration is required. Pay is hourly at the band above as observed on the platform; rates and availability are set per project and are not guaranteed.

What they look for

Video Capturing
IMU (Inertial Measurement Unit)
Attention to Detail

Before you apply

Would you pass this screen?

Drawn from what this listing actually requires. Answered in your browser — nothing is sent to us, and we do not store your answers.

Are you currently based in and available to work from the United States?
US-based location · required
Do you have access to both a computer capable of clean screen capture and a camera (phone or dedicated) suitable for handheld footage?
Recording equipment · required
Do you have reliable, stable internet bandwidth adequate for uploading multiple video files (gigabytes per batch)?
Reliable upload bandwidth · required
Do you have access to a quiet physical space where you can record narration without background noise?
Quiet recording space · preferred

0 of 4 answered

Included with this listing

Interview prep — 11 questions

Matched to what this platform's screening actually measures. Answer each out loud in 60–90 seconds: claim → concrete example → implication.

What it tests: Whether you parse multi-clause instructions, identify constraint hierarchies, and spot the edge case (what counts as a state boundary vs. mid-action).
A strong answer: A strong answer names each constraint separately—narration required, state boundaries only, no jump cuts—then works through a concrete example (e.g., 'If I'm adding an item to cart, that's one state; clicking checkout is a separate one; I wouldn't cut while my hand is moving') to show you've thought about where the rule could break.
Where people lose it: Restating the spec in your own words without showing you've interrogated it: saying 'I'd capture the whole flow smoothly' misses that you've identified what 'state boundary' actually means in this context, and leaves you guessing during capture.