About This Role
This project-based role involves recording paired real-camera videos to train video-editing AI models. You will contribute to the "Filmmaking: Video-to-Video Edits" project on Outlier, a platform connecting subject matter experts with leading AI companies. All footage must be original camera recordings, without digital alteration.
Responsibilities
- Plan and record continuous 10–30 second actions with 3+ beats, such as dance phrases, sports drills, cooking sequences, workouts, craft activities, or two-person interactions
- For Open sets: record one performance simultaneously on 2+ cameras (3–4 preferred), at least ~30° apart or at a clearly different height or distance
- For Precise pairs: record input and output takes on a locked tripod, changing only the action (and any props the new action requires)
- Align all videos frame-for-frame in post by trimming only, without warping, time-stretching, or frame blending
- Deliver every video at exactly 1920×1080 or 1080×1920, H.264, 30 Mbps, 30 fps
- Create an audit video for every deliverable: a sync proof for Open sets, a 50% opacity overlay for Precise pairs
- Write a short 1–2 sentence prompt describing exactly what changed, plus one metadata row per deliverable
Requirements
- Ability to record one performance simultaneously on 2+ cameras, with exposure, white balance, and frame rate locked
- A tripod that can be locked for zero frame shift between takes
- Ability to align videos frame-for-frame in post, with the same frame count, same start frame, and no drift
- Access to controllable locations: kitchens, living rooms, home offices, balconies, gardens, or studio sets
- Performers who will sign appearance and ML-training releases (no minors)
- Original footage only, no stock, no web video, and nothing recorded for another client
Important Details
- Project categories: Open, Multiple Angles (one performance recorded simultaneously on 2+ cameras from clearly different angles); Precise Edits, Action Change (the same scene recorded twice on a locked tripod, with the person performing a different full action in each take)
