
Song map.
LocalAIAudioToSongMap maps a whole song: its beats, bars, tempo changes and sections like intro, build and drop. Drivers can read it.
The song shows as a lane on the timeline, and loop ends, markers, keys and clip edges snap to it.
Draw light that follows a dancer, and light real footage as if it were a 3D set. Keys from the whole graph now line up in one Dope sheet, and a sculpt can be masked from its own surface.
ImageLightLines draws glowing lines that bead, dash, reveal and hide behind depth. Wire a body pose in and the lines follow a dancer.
ImageSpreadReveal lights a picture up from a point or a line, with a flare on the bright parts. ImageScope shows a histogram, waveform, parade, vectorscope and false color, with levels as numbers.
The time bar now shows the selected node’s keys. The new Dope sheet lines up keys from many nodes, so you can move or scale a whole show at once.
Tempo ramps and meter changes get their own lane. Select a span on the ruler to cut, copy, paste or clear it across the graph.



Draw on a work plane, cut with a box, circle, polygon or freehand shape, and keep shaping from there.
Mask a sculpt from its own shape: its hollows and ridges, how hidden each spot is, its color, a picture or any point value. Paint through that mask and the dirt settles into the creases.
Add or cut with a shape and keep both parts editable until you merge them. Decimate and quad remesh run from the Mesh menu and keep the mask, color, face sets and UVs.
3DGeoFootagePlate turns a picture and its depth into a 3D plate seen from the shot’s camera. Scene lights, shadows, depth of field and motion blur then work on real footage.
Any component can now run on its own thread or in its own process, beside the show. From the outside nothing changes: it keeps its node, its knobs and its wires.
The Profiler has a new Engines tab, and a component's inspector shows where it runs and its live frame rate. GPU chains, 3D renders and AI models already run on their own workers, so read the Profiler first.


The rest of the release, in short.

LocalAIAudioToSongMap maps a whole song: its beats, bars, tempo changes and sections like intro, build and drop. Drivers can read it.
The song shows as a lane on the timeline, and loop ends, markers, keys and clip edges snap to it.
LocalAIVideoErase paints a person or object out of a video, along with the shadows and reflections it casts. Give it the video and a mask.
LocalAIVideoRestore cleans up old or compressed video and rebuilds lost detail like hair, leaves and the edges of text. It can also make the picture up to 4 times larger.

OutputVideoFile and the timeline render have presets for YouTube, Instagram, TikTok, Vimeo and other platforms, with loudness leveling to a target LUFS.
Safe area and social app guides show in the viewport.

Fisheye lenses work on all 3D cameras and render paths, CPU and GPU, including splats, TAA and motion vectors.
ImageShader gets an output size, edge modes for each input and linear reading of 8-bit inputs. ImageEdgeDetect draws line art from id or normal layers.
ImageHeatDistort can take its shimmer from a picture, ImageTriplanar outputs its UVs, and ImageAutoCrop outputs its crop box.
Squash, flip guard and auto clavicle on two-bone IK. Balance over the feet and a center of mass goal on full-body IK. Ragdolls get back up and blend into animation.
Blend shapes and surface deform run on the GPU. Match FK to IK and IK to FK are in the viewport Rig menu.

Relight, delight, shadows, deformers, cleanup, color and LUTs, dissolves, decimate and mesh conversion now run on the GPU.
Splats show correctly in screen space reflections and AO, and 3DGeoSplatNormals gets screen space and hybrid modes.
Depth pictures carry tags that say what their values mean, so depth nodes read them right with no setup. Depth cameras can pair each depth frame with its color frame.
ImageMarkerTrack gives a full 3D position and rotation for ArUco markers and QR codes.
iFacialMocap and Facemotion3D over TCP, BVH export from character export, and hand and head mesh skeletons that plug straight into the rig nodes.
AVIF pictures read and write, and a folder of numbered frames works as one clip in an arrangement.
$ kinematic loudness sample_audio.wav
loudness -16.2 LUFS
true peak -2.9 dBTP
sample peak -3.0 dBFS
range 0.0 LU
length 3.20 s, 44100 Hz, stereoMany more settings can be driven, and each node says which. Node search understands plain phrases like “slow motion” or “camera move”.
New kinematic drivers, kinematic scopes and kinematic loudness commands, and render --preset.
A new name, a timeline on every graph, and modeling, sculpting, painting, photo and sound editing in the viewport.
The smaller changes, in short. The changelog has all of them.
No account, no license key. macOS, Windows, Linux and Raspberry Pi.