Foundry has released SmartRoto as a separately licensed plug-in for Nuke 16.1 and later. The documented process starts with an artist drawing an initial shape, selecting smart keyframes, locking accepted results and propagating across a chosen range. Additional keyframes or local repairs remain available. Foundry also says processing stays local and footage is not transmitted, while recommending sub-2K material for less capable GPUs. The page lists an annual price and a 30-day trial.
Speed comes from reducing repetition, not removing judgment
The useful design for short-drama post is that familiar shot responsibility remains visible. Hair, transparency, motion blur, occlusion and off-screen entrances still require artistic judgment; the model carries a defined shape through more frames. Foundry claims up to four-times speed on some work, but this is vendor evidence and should not enter a schedule untested. Teams should select the three hardest shot types in the project and record manual setup, propagation time, added keyframes, edge repair and final approval.
Procurement must combine the plug-in license with the current Nuke version, GPU memory and footage resolution. A license price is visible; host upgrades, GPU queues, training and repair after failed propagation are easier to miss. Measure cost per approved second or approved shot against a conventional roto baseline. If hair and rapid occlusion still demand extensive manual work, the clearest gain may sit in medium-complexity subject separation rather than every shot.
The official page establishes availability, workflow, version requirements and vendor performance claims. It does not provide independent samples for serialized short drama, varied hair, low-bitrate sources or highly compressed vertical footage. Before production, lock plug-in and model behavior, retain original splines, keyframes and renders, and rerun the same test set after updates. Automated propagation becomes controlled craft only when failures are locatable and versions reversible.