Indicates a feature suggested and voted up by users on the Flame Feedback portal.

---

Action / GMask Tracer / Image

Improvements have been made to the Node Bins:

• Some Matchbox shaders included with Flame like A2Beauty are now available from the All Nodes bins.
• Matchbox and Lightbox nodes can be added to a custom bin.

---

Auto Matte

The legacy Auto Matte tool has been renamed to Luma Matte to avoid confusion with the new AutoMatte machine learning isolation model available in the Selective object.

---

GMask Tracer

Default Output

By default, GMask Tracer has Result and OutMatte outputs to make it more compatible with Action when constantly navigating between the two. The goal is to allow to only have to use the F4 keyboard shortcut to navigate between outputs by default.

Selective

The Selective node is available in GMask Tracer in Matte Pipeline mode. This mode allows the use of the Selective node to extract mattes using machine learning models, for example, and supports the same pipeline and Matchbox processing capabilities as Action and Image.

---

Inference

It is no longer necessary to use the PreBRN and PostBRN Matchbox to apply the pre-processing and post-processing required by the BiRefNet machine learning model. These two processes will be applied automatically when the following tokens are added to the JSON ONNX model sidecar.

Normalization

Add the following input token to the JSON sidecar to automatically apply ImageNet dataset normalization to any incoming image, allowing models requiring this normalization to perform optimally.

"Normalization" : {"Mean": [0.485, 0.456, 0.406], "StandardDeviation":[0.229, 0.224, 0.225]}

Matte Post-Processing

Add the following output token to the JSON sidecar to automatically apply a sigmoid to the output so the matte gerenated by models like BiRefNet can be used for compositing.

"PostProcessing": "Sigmoid"

---

Match Grain

The Grain output has been renamed OutGrain to avoid conflicts in the Python API.

---

Resize

On macOS Sequoia 15.4 or later, when running on workstations equipped with Apple silicon, Machine Learning Upscale is available for use in the Resolution option to double, triple, or quadruple the input resolution.

---

Temporal Smoothing

This Batch node allows for temporal smoothing of Matte that exhibits temporal stability issues.

In addition to the Matte input, Forward and Backward Motion Vectors inputs need to be provided, since motion vectors are required for the processing. The window size value defines how many frames are going to be used to process the current frame. For example, 5 frames means that 2 frames on each side of the current frame are used for the selected filter to do the processing. The filters define how the temporal smoothing is going to be done, and their usage depends on the type of temporal instabilities encountered.

---

Timewarp And Morph

On macOS Sequoia 15.4 or later, when running on workstations equipped with Apple silicon, the frame interpolation machine learning models can run on the GPU and neural engine to offer better performance.

---

Type

Attributes

The Layer Opacity channel can be modified from the user interface using the numeric field in the Layer section of the Attributes menu, which has been moved from the right side to the left side, below the Characters section. In previous versions, the channel was only available in the Animation editor.

Select Font Widget

Favourited fonts appear in the Favourites tab rather than at the top of other tabs.

The names of the fonts within the Select Font widget can be displayed using the default font used by Flame by disabling the Use Font to Display Names and Style option in the gear menu located at the bottom-left of the widget.

Workflow

Animation channels for all characters are selected after exiting the editing mode using the Esc key.