Interactive Generative Motion Editing via Scheduled Inpainting

Disney Research Studios has unveiled scheduled inpainting, a motion-editing approach that treats editing and generation as one workflow instead of separate tools. The core idea is to let artists keep the parts of an animation they want, then selectively regenerate the rest with a generative model so the result stays natural and consistent.

That matters because traditional motion editing tends to hold up for small tweaks, but breaks down when the change is structural: limbs drift, contacts warp, and the fix often becomes “author new motion anyway.” Scheduled inpainting is designed to give finer spatiotemporal control over that tradeoff, so you can decide how much of the original motion survives versus how much gets synthesized.

The system is built on top of generative models that already support direct manipulation, and it is meant to work interactively rather than as a one-shot offline bake. Disney says it supports practical editing tasks like extending a clip, stitching multiple clips together, and compositing motion from different sources.

The work was published on arXiv on July 30, 2026, and was authored by Dhruv Agrawal, Dominik Borer, Luca Vögeli, Robert W. Sumner, Martin Guay, and Jakob Buhmann. For game teams, the interesting part is less the paper title and more the workflow implication: if this class of method matures, motion cleanup could shift from manual repair toward guided regeneration, especially for animation-heavy pipelines.