Oink
AIDepthcam AIDepth

AIDepth

Analyze runs Video Depth Anything Small (Apache-2.0, temporally stable) over the frames and aligns it to MoGe-2 (MIT) metric depth on a few keyframes.

raw .md

Analyze runs Video Depth Anything Small (Apache-2.0, temporally stable) over the frames and aligns it to MoGe-2 (MIT) metric depth on a few keyframes. Output: depth layer (1/Z like ScanlineRender, or Z), optional depth_metric and N (camera-space normals), upsampled to the plate with a plate-guided filter. Create Camera makes a Camera with the estimated focal length. Model: Auto uses MoGe-2 ViT-B (measured as good as ViT-L and 4x faster); ViT-L is an optional download. Field of view, first that applies: Known FOV (when Field of View is set to known), a Camera on the cam input, the plate's header metadata, then the MoGe-2 estimate. Analyze on 1 to 3 frames keeps the models loaded, so the next frame is fast.

NodeAIDepth
MenuAI › AIDepth
Names in each format
.oink comp (JSON)"type": "OC_N_aidepth"
.nk scriptnone: written as an Oink node
Pythonnuke.createNode("OC_N_aidepth")
Assistant (MCP)create_node(type="AIDepth")
Tab searchaidepth, or its initials aid

AIDepth properties

Inputs and outputs

PortNameType
input 0Imageimage
input 1camscene
output 0Imageimage

Knobs

Panel tab Aidepth, in panel order.

KnobLabelTypeDefaultRange / values.nk knobDescription
depth_conventiondepthmenuinvinv (1/Z (Nuke)), z (Z)
output_metricmetric Z layerboolfalseAlso output depth_metric (.r = Z metres)
output_normalsnormals layerboolfalseAlso run MoGe-2 on every frame for an N layer
keyframesmetric keyframesfloat51 to 64 (integer)Frames MoGe-2 measures for the metric scale / shift fit
encodeinputmenusrgbsrgb (scene linear), none (display)
analysis_widthanalysis sizefloat1036256 to 2048 (integer)Long side of the frames sent to the model (px)
fov_modefield of viewmenuestimateestimate, known
known_fovknown FOVfloat501 to 179
haperturehoriz aperturefloat24.5760.1 to 100Film back width (mm) for the focal readout / Create Camera
spatial_sigmaupsample spatialfloat10.1 to 4Joint bilateral spatial sigma (model pixels)
range_sigmaupsample edgefloat0.080.005 to 2Joint bilateral range sigma (plate colour, lower = sharper edges)
keep_other_frameskeep other framesboolfalseAnalyze keeps results of frames outside the new range
keep_loadedkeep model loadedboolfalseLeave the models in VRAM after Analyze (faster re-runs)

Other knobs (Python, .nk and the assistant; not on the panel)

KnobLabelTypeDefaultRange / values.nk knobDescription
gl_colorViewer Controls Colorfloat ×30, 0, 00 to 1
postage_stamp_frameStatic Frameint1-2147483648 to 2147483647
lifetimeStartLifetime Startint0-2147483648 to 2147483647
lifetimeEndLifetime Endint100-2147483648 to 2147483647
useLifetimeUse Lifetimebooleanfalse
cachedCachedbooleanfalse
bookmarkBookmarkbooleanfalse
dope_sheetDope Sheetbooleanfalse
clone_groupClone Groupstring
link_sourceLink Sourcestring
link_overridesLink Overridesstring
split_viewsSplit Viewsstring
moge_modelModelenumAUTOAUTO, VITL, VITBGeometry model (metric depth, field of view, normals); licence tier: optional / bundled
results_dataResultsstring{}
fov_xFOV X (deg)float0-3.4028234663852886e+38 to 3.4028234663852886e+38
fov_yFOV Y (deg)float0-3.4028234663852886e+38 to 3.4028234663852886e+38
focalFocal (mm)float0-3.4028234663852886e+38 to 3.4028234663852886e+38
analysed_firstAnalysed Firstint0-2147483648 to 2147483647
analysed_lastAnalysed Lastint-1-2147483648 to 2147483647
fit_scaleFit Scalefloat0-3.4028234663852886e+38 to 3.4028234663852886e+38
fit_shiftFit Shiftfloat0-3.4028234663852886e+38 to 3.4028234663852886e+38
fit_errorFit Errorfloat0-3.4028234663852886e+38 to 3.4028234663852886e+38Median relative depth error of the fit on the keyframes

On this page