Automatically detect and segment arbitrary objects in images based on simple text prompts — whether you specify a category name (“car,” “person,” “tree”) or use more descriptive phrases.

Masks are generated with Mask ML and seamlessly propagated using Matte Assist ML, giving you accurate and consistent results across frames. The workflow is fully flexible. Use it through the UI or integrate it into pipelines via the command line for large-scale automation.

This enables a true end-to-end automated workflow: prompt, segment, propagate, and render — dramatically reducing time spent on repetitive tasks.

Bring precision and flexibility to facial effects work.

The Face ML node automatically detects and segments key facial regions — such as eyes, lips, mouth, and surrounding skin — into individual mattes. This granular separation allows for targeted adjustments or creative effects to each feature independently, all accomplished without manual rotoscoping.

When performing subtle beauty retouching or building complex composites, Face ML provides high-quality, tracked mattes, ensuring the fine control needed for invisible beauty work.

The 3D Scene node now includes a fully integrated SynthEyes camera tracking and solving system, directly inside Silhouette.

Analyze 2D features in footage to accurately calculate the camera’s 3D position and orientation, generating both a point cloud of the scene and a reliable camera solve.

This powerful integration supports an Unproject/Reproject workflow, enabling paint or compositing directly into a locked-down sequence with precision and ease — without leaving the Silhouette environment.

• Cryptomatte: Available mattes are applied as overlays to the optional image input. The Node also supports selective matte activation directly within the viewer that once enabled are available to use in other Nodes and added to the Selection list.
• Mask ML: Adds the ability to output objects as Cryptomatte data via the data port, multiple object support in an Object List, and AI text-based Object Detection.
• Matte Assist ML: Compute cache provides seamless playback without recomputation. Additionally, when either the Mask Source > Cryptomatte or Roto options are selected and multiple objects are present in the Roto and Mask ML nodes, each object is output separately as Cryptomatte data via the data port.
• Tracker: Mark areas that interfere with tracking so they are automatically ignored using the Occlusion Matte input. Connect a 3D scene node to a Tracker-capable node’s data port, and the Point Tracker generates a tracked layer from the selected 3D track points.

Instantly replace imperfect frames.

Harness AI to quickly remove and restore photo flashes, artifacts, dropped frames, or even change a performance. Frame Fixer ML automates the process with intelligent machine learning. Set the first bad frame, input the number of frames affected, and AI does the rest.

• Use it on original footage or painted keyframes
• Create a morph between painted keyframes by connecting Paint to both the main and Data input ports