Rotobot Next
Editable splines for AI rotoscoping — refine, don't redraw.
What it is
Rotobot Next takes an input video sequence and produces a set of animated splines that can be imported into the compositing and rotoscoping packages used in video post-production: Nuke, Silhouette, After Effects, and Fusion.
Most ML rotoscoping arrives as a dense pixel mask. When the edge is wrong, the artist has to redraw from scratch — there are no handles to grab. Rotobot Next instead outputs a small, sparse, visually editable data structure: Bezier and B-spline curves with keyframes placed at motion apexes, the way professional roto artists actually work.
How the pipeline works
- Send footage. Upload via SaaS, or run locally on-prem.
- ML compute. Asynchronous GPU compute generates the prediction.
- Splines exported. Animated Bezier curves with minimal control points and motion-aware keyframes.
- Open & refine. Import directly into Nuke, Silhouette, After Effects, or Fusion — no middleware, no conversion step.
Why this is hard to copy
- ML-informed shape layout. The only approach using ML to inform spline layout — addressing the shape-semantics problem pure segmentation cannot.
- Temporal splines. Keyframes placed at motion apexes, not every frame.
- DCC-native output. Splines that import directly into the host application.
- Deployment choice. Token-based SaaS, or local on-prem compute for studios with firewall requirements.
How you’ll buy it
Customers purchase token credits up front and consume them per job, with cost scaled to the complexity of the computation. An on-premise plugin is planned for studios bound by confidentiality agreements that prevent video leaving their network.
Who it’s for
Compositors and visual-effects supervisors working in Adobe After Effects, Blackmagic Fusion, Foundry Nuke, and BorisFX Silhouette — particularly mid-to-high-end VFX houses delivering final rotoscoping for feature film and high-end television.