FXTD®Radiance for ComfyUI
RADIANCE / DOCUMENTATION3.5.0 · Release candidate

Node reference / Radiance for ComfyUI

VFX nodes

Plate prep, masks, inpaint, depth, optics, motion, multipass, and scene cuts.

31 nodes. All sections

Aberration

RadianceChromaticAberration

Aberration node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Simulate lateral chromatic aberration (fringe) as a lens artefact.

Inputs

Input Type Default Range or choices What it does
image IMAGE Image whose red, green and blue channels are scaled radially by separate amounts. Alpha is not shifted.
shift_r float 0.005 -0.1 to 0.1, step 0.001 Radial scale for Red channel. Positive = pushed outward.
shift_g float 0 -0.1 to 0.1, step 0.001 Radial scale for Green channel. Typically 0 (reference channel).
shift_b float -0.005 -0.1 to 0.1, step 0.001 Radial scale for Blue channel. Negative = pulled inward.
center_x float 0.5 0 to 1, step 0.01 Horizontal center of the distortion/effect (0.0 = left, 1.0 = right).
center_y float 0.5 0 to 1, step 0.01 Vertical center of the effect (0.0 = top, 1.0 = bottom).
invert boolean off Invert shift for CA removal / undistortion pass.

Outputs

Output Type
image IMAGE

Anamorphic Streaks

RadianceAnamorphicStreaks

Anamorphic Streaks node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Add anamorphic lens streak flares to bright highlights.

Inputs

Input Type Default Range or choices What it does
image IMAGE Image to add streaks to, ideally scene-linear so that highlights above 1.0 drive the flare.
threshold float 1 0 to 10, step 0.01 Per-channel value above which highlights generate streaks; only the excess over it is streaked. 1.0 = SDR white.
streak_length int 64 1 to 512, step 1 Maximum streak tail length in pixels.
streak_color_r float 0 0 to 5, step 0.01 Red component of the anamorphic streak colour. Values > 1.0 produce HDR-energy streaks.
streak_color_g float 0.5 0 to 5, step 0.01 Green component of the anamorphic streak colour.
streak_color_b float 1 0 to 5, step 0.01 Streak color tint. Default is the cool blue-cyan typical of anamorphic lenses.
intensity float 1 0 to 10, step 0.01 Vignette or streak intensity. Values > 1.0 add HDR energy.
streak_direction choice Horizontal Horizontal, Vertical, Diagonal +45, Diagonal -45 Direction of streak propagation.
streak_falloff float 3 0.5 to 20, step 0.1 Exponential decay rate of the streak kernel. Higher = tighter falloff (shorter effective tail). Lower = longer, softer streak.

Outputs

Output Type What it is
image IMAGE Original image with streaks composited (add blend).
streak_pass IMAGE Isolated streak pass for manual compositing (pre-add).

Bit-Depth Degrade

RadianceBitDepthDegrade

Bit-Depth Degrade node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Quantise an image to a lower bit depth, with optional dither, and output the error and a threshold mask. Use it to preview banding in 8/10-bit deliveries; feed the display-encoded image as it will be delivered.

Inputs

Input Type Default Range or choices What it does
image IMAGE Display-encoded 0..1 image, quantised as is (no transfer conversion); values outside 0..1 are clipped. Alpha passes through unchanged.
bit_depth int 8 4 to 16, step 1 Target bits per channel; the 0..1 range is split into 2^bits - 1 steps.
dither_mode choice triangular none, triangular, floyd-steinberg none: plain rounding. triangular: random TPDF noise of +/-1 step before rounding. floyd-steinberg: error diffusion (the classic scan-line result, computed a diagonal at a time).
delta_gain (optional) float 10 1 to 100, step 0.5 Multiplier on the absolute error |original - quantised| for the delta_amplified output, clipped to 1.
banding_threshold (optional) float 0.004 0.0005 to 0.05, step 0.0005 Per-pixel error (0..1 units, largest channel) above which banding_mask is white. 0.004 is about one 8-bit code value. This flags quantisation error, not detected bands.
restore_from_quantized (optional) boolean off Currently has no effect: the node ignores this setting.

Outputs

Output Type
quantized IMAGE
delta_amplified IMAGE
banding_mask IMAGE
metrics STRING

Blend Composite

RadianceBlendComposite

Blend Composite node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Composite two images using Normal, Add, Screen, Multiply, Overlay, Soft Light, Difference, or Divide blend modes.

Inputs

Input Type Default Range or choices What it does
base IMAGE Bottom layer (background).
blend IMAGE Top layer (foreground).
mode choice Normal Normal, Add, Screen, Multiply, Overlay, Soft Light, Difference, Divide Blend operation of blend over base, in fp32. Add, Multiply and Divide keep HDR values above 1.0; Screen clamps both layers to 0-1 first; Overlay and Soft Light assume 0-1 display-range values.
opacity float 1 0 to 1, step 0.01 Overall strength of the blend layer.
mask (optional) MASK Optional per-pixel mask (grayscale). White = full blend, Black = base only.

Outputs

Output Type
image IMAGE

Camera Sync

RadianceCameraSync

Camera Sync node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Read a JSON camera (focal length, f-stop, shutter, transform) for the optical nodes.

Inputs

Input Type Default Range or choices What it does
camera_data string {} multi-line text JSON camera data: {"focal_length": 35.0, "f_stop": 2.8, "shutter_angle": 180.0, "transform": [...]} or {"frames": [ ... ]}
frame_offset int 0 -10000 to 10000, step 1 Index into a "frames" list (animated camera). Ignored for a single camera.
camera_file (optional) string Path to a .json camera file. Takes precedence over camera_data.

Outputs

Output Type
camera RADIANCE_CAMERA
focal_length FLOAT
f_stop FLOAT
shutter_angle INT

Compression Artifacts

RadianceCompressionArtifacts

Compression Artifacts node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Add compression artifacts (JPEG blocking, color banding).

Inputs

Input Type Default Range or choices What it does
image IMAGE Display-encoded image. It is quantised to 8 bits and clipped to 0-1 in every mode, so HDR values are lost; alpha is kept.
artifact_type choice JPEG JPEG, Banding, Both JPEG: real JPEG round trip plus block averaging (block_size). Banding: posterise to banding_levels. Both: JPEG then banding.
quality int 50 1 to 100, step 1 JPEG encoder quality, 1 = worst, 100 = best. Ignored in Banding mode.
block_size (optional) int 8 4 to 32, step 4 Size in pixels of the square blocks each averaged to one flat colour after the JPEG pass (a mosaic, not a DCT setting). Applies in JPEG and Both. Blocks cut off at the right or bottom edge are averaged over the pixels they have.
color_subsampling (optional) boolean on Apply chroma subsampling (4:2:0) to simulate video codec color compression.
banding_levels (optional) int 32 4 to 256, step 4 Quantisation steps per channel in Banding and Both modes (rounded down, so it darkens by up to one step). Fewer = stronger banding.
noise_amount (optional) float 0 0 to 0.1, step 0.005 Standard deviation of Gaussian noise added at the end, in 0-1 code values (0.01 = about 2.5 of 255). 0 = off.
seed (optional) int 0 0 to 18446744073709551615 Random seed for reproducible noise patterns.

Outputs

Output Type
image IMAGE

Defocus

RadianceDepthOfField

Defocus node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Apply cinematic depth of field blur with optional depth map input.

Inputs

Input Type Default Range or choices What it does
image IMAGE Image or frame batch to defocus. Works on linear or display-encoded values; super-whites are kept.
blur_amount float 5 0 to 50, step 0.5 Maximum defocus in pixels: Gaussian sigma for Circle (kernel capped at 31 px), kernel radius for the other shapes. Below 0.1 the image is returned unchanged.
depth_map (optional) IMAGE Depth in 0-1 from the first channel, 0 = near, 1 = far; resized to the image. Without it, a radial map is used (sharp centre, blurred corners).
focus_distance (optional) float 0.5 0 to 1, step 0.01 Depth value that is in focus, on the depth_map scale (0 = near, 1 = far). With no depth map, 0 is the frame centre.
focus_range (optional) float 0.1 0.01 to 0.5, step 0.01 Depth distance either side of focus_distance that stays sharp; blur then ramps up to full over the rest of the depth range.
bokeh_shape (optional) choice Circle Circle, Hexagon, Octagon, Anamorphic Oval Aperture shape. Non-circular shapes use a convolution kernel for physically accurate bokeh.
highlight_boost (optional) float 1 1 to 3, step 0.1 Brightens defocused highlights (luma above 0.8) by up to this factor, scaled by the amount of blur. 1.0 = off.
foreground_blur (optional) boolean on Blur areas nearer than focus_distance (depth below it). Off keeps the foreground sharp and blurs only the background.
use_gpu (optional) boolean on Run the effect on GPU via CUDA/MPS. Falls back to CPU if unavailable.

Outputs

Output Type
image IMAGE

Depth

RadianceDepthMapGenerator

Depth node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Depth Anything V2 monocular depth estimation. Video-safe — standardizes each frame with spatial-temporal alignment preventing flickering. Outputs 3-channel grayscale depth map. Connect to Depth of Field node for realistic defocus blur.

Inputs

Input Type Default Range or choices What it does
image IMAGE Display-encoded image or frame batch. Frames with values above 1.05 are Reinhard tone-mapped to 0..1 first; alpha is ignored.
model_size choice Large (335M - Best) Small (25M - Fast), Base (98M - Balanced), Large (335M - Best) Depth Anything V2 model size. Small = fast previews, Large = best quality.
normalize (optional) boolean on Normalize depth to 0-1 range. For video, frames are standardized for temporal consistency.
invert (optional) boolean off Invert depth (white=far, black=near).
blur_edges (optional) float 0 0 to 5, step 0.5 Gaussian blur to smooth depth discontinuities.
use_gpu (optional) boolean on Run depth estimation on GPU. Requires a CUDA-capable device. Falls back to CPU if unavailable.

Outputs

Output Type
depth_map IMAGE

Grain

RadianceFilmGrain

Grain node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Add simulated film grain: Gaussian noise blurred to a per-channel grain size and added to the image, optionally reduced in highlights (hdr_aware) so it stays usable on HDR plates.

Inputs

Input Type Default Range or choices What it does
image IMAGE Image to grain. Grain is added in the image's own encoding; alpha is left untouched.
grain_size float 1 0.1 to 8, step 0.05 Base grain size in pixels (Gaussian sigma). 1.0 ≈ ISO 400 medium format.
grain_strength float 0.04 0 to 1, step 0.001 Grain amplitude. 0.015=fine, 0.04=medium, 0.12=heavy.
grain_size_r_offset float 0.1 -1 to 1, step 0.05 R channel grain size offset from base (pixels). Red grain is typically coarser.
grain_size_g_offset float 0 -1 to 1, step 0.05 G channel grain size offset. Green has finest grain (most photo-sites).
grain_size_b_offset float -0.1 -1 to 1, step 0.05 B channel grain size offset. Blue grain is typically finest.
hdr_aware boolean on Scale grain amplitude by 1/sqrt(luma + 1): full strength at black, about 0.7x at 1.0 and 0.45x at 4.0. Grain size is unchanged. Disable for uniform grain.
seed int 0 0 to 2147483647, step 1 Random seed for a repeatable grain pattern. 0 = no reseed, so the pattern depends on the global random state when the node runs.

Outputs

Output Type
image IMAGE

HDR Grain Matcher

RadianceHDRGrainMatcher

HDR Grain Matcher node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Transfer grain from a reference plate onto a clean image: the reference's fine detail is high-pass filtered in log2 space and added to the target in log2, so it scales with exposure and does not burn out highlights.

Inputs

Input Type Default Range or choices What it does
target IMAGE Clean RGB image or sequence to receive grain, ideally scene-linear. Negative values are clamped to 0.
reference IMAGE Grainy plate to take grain from. Must match the target's width and height; frames are reused in a cycle if it is shorter than the target. All high-frequency detail is transferred, so use a flat, defocused area if possible.
intensity float 1 0 to 5, step 0.05 Multiplier on the extracted grain. 0 = no grain, 1 = matched to the reference.
kernel_size int 3 1 to 15, step 2 Box-blur size in pixels used to split grain from the image. Larger values capture coarser grain (and more image detail); 1 extracts nothing.
r_gain float 1 0 to 2, step 0.05 Extra grain multiplier for the red channel. 1.0 = unchanged.
g_gain float 1 0 to 2, step 0.05 Extra grain multiplier for the green channel. 1.0 = unchanged.
b_gain float 1 0 to 2, step 0.05 Extra grain multiplier for the blue channel. 1.0 = unchanged.

Outputs

Output Type
grained_image IMAGE

HDR Inpaint Crop

RadianceHDRCrop

HDR Inpaint Crop node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Crop an image batch to the region around a mask, with extra context, for inpainting at full resolution. Pair with Radiance HDR Stitch to put the result back.

Inputs

Input Type Default Range or choices What it does
image IMAGE Plate or frame batch to crop. Values pass through unchanged, so HDR and scene-linear data are kept.
mask MASK Region to inpaint. Pixels above 0.05 on any frame define one union box used for every frame, so the crop does not move over time. An empty mask crops the full frame.
context_padding float 1.5 1 to 4, step 0.05 Crop size as a multiple of the mask's bounding box, centred on it. 1.0 = tight to the mask, 1.5 = 50% larger for surrounding context.
force_multiple int 16 1 to 256, step 1 Round the crop width and height up to a multiple of this (16 suits Wan and LTX latents). 1 = no rounding. A crop that reaches the frame edge is limited to the frame size and may not be a multiple.

Outputs

Output Type
cropped_image IMAGE
cropped_mask MASK
stitcher_data STITCHER_DATA

HDR Inpaint Stitch

RadianceHDRStitch

HDR Inpaint Stitch node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Composite an inpainted crop from Radiance HDR Crop back into the original plate through its mask, with a feathered or multi-band blend. Output is not clamped above 1.0.

Inputs

Input Type Default Range or choices What it does
original_image IMAGE The full-size plate that was passed to HDR Crop.
cropped_image IMAGE The processed crop. It must keep the crop's exact width and height, as it is pasted back at the stored coordinates.
cropped_mask MASK Mask for the crop (normally cropped_mask from HDR Crop). Only this area of the crop replaces the plate.
stitcher_data STITCHER_DATA Crop coordinates from HDR Crop.
blend_mode choice Linear_Laplacian Linear_Laplacian, Linear_Gaussian, Standard Linear_Laplacian: 3-level multi-band blend with the feathered mask, negatives clamped to 0. Linear_Gaussian and Standard: plain mix through the feathered mask. No transfer conversion is applied.
feather_radius int 16 0 to 128, step 1 Mask softening radius in pixels (a box blur of 2 x radius + 1), used by every blend mode. 0 preserves the input mask without softening.

Outputs

Output Type
stitched_image IMAGE
stitch_blend_mask MASK

LensDistortion

RadianceLensDistortion

LensDistortion node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Apply or remove lens distortion using Brown-Conrady coefficients.

Inputs

Input Type Default Range or choices What it does
image IMAGE Image or sequence to distort or undistort. Pixels are only resampled (no colour change), so any encoding works.
k1 float 0 -1 to 1, step 0.005 Primary radial distortion. k1<0 = barrel, k1>0 = pincushion.
k2 float 0 -1 to 1, step 0.005 Secondary radial distortion. Affects extreme corners. Use sparingly.
scale float 1 0.1 to 2, step 0.01 Uniform scale applied after distortion. Use to crop black borders.
center_x float 0.5 0 to 1, step 0.01 Optical center X (0.5 = image center).
center_y float 0.5 0 to 1, step 0.01 Optical center Y (0.5 = image center).
padding_mode choice zeros zeros, reflection, border Edge fill: zeros=black, reflection=mirrored, border=edge-clamped.
invert boolean off Invert warp for plate undistortion. Uses exact closed-form inverse — valid for all k1/k2 combinations except the degenerate singularity at 1+k1·r²+k2·r⁴=0 (clamped automatically).

Outputs

Output Type What it is
image IMAGE Distorted image.
st_map IMAGE ST-Map: R=U, G=V normalized [0,1] sample coordinates. Wire to Nuke STMap / Fusion DisplaceImage for downstream use.

Matte

RadianceLinearMatting

Matte node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Refine a rough mask into a soft alpha matte with a guided filter that follows the image's edges (no learned matting model). Also outputs the image premultiplied by the matte.

Inputs

Input Type Default Range or choices What it does
image IMAGE Guide image whose edges the matte will follow. Used as given (scene-linear or display-encoded); eps behaves differently for each.
mask MASK Rough matte to refine, 0 to 1.
method choice GuidedFilter GuidedFilter Guided filter on the scene-linear image. The only method implemented.
trimap_dilation int 12 0 to 128, step 1 Radius of the unknown trimap band and guided-filter window. Regions whose entire neighbourhood is at least 0.99 stay foreground; those at most 0.01 stay background. Only the unknown band is refined. 0 preserves the mask.
eps float 0.0001 1e-06 to 0.1, step 1e-06 Guided filter regularisation. Small values make the matte follow fine image edges closely; large values give a smoother matte closer to a blurred mask.

Outputs

Output Type
alpha_matte MASK
foreground_image IMAGE

MotionBlur

RadianceMotionBlur

MotionBlur node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Apply physically-based motion blur using optical flow vectors.

Inputs

Input Type Default Range or choices What it does
image IMAGE Frames to blur, same size and batch as motion_vectors. Averaged as given, so scene-linear input gives physically correct highlight streaks.
motion_vectors IMAGE 32-bit UV vectors from Radiance Optical Flow.
shutter_angle float 180 0 to 720, step 1 Standard cinema is 180°. Higher = more blur. 360° = full frame motion blur.
samples int 8 2 to 32, step 1 Number of sub-frame integration samples. Higher = smoother streaks.
energy_conservation boolean on On: plain average, total light conserved. Off: legacy look, the whole frame is scaled so its brightest value matches the source peak (brightens everything, not only streaks).

Outputs

Output Type
image IMAGE

MotionVectors

RadianceOpticalFlow

MotionVectors node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Estimate dense optical flow between adjacent frames.

Inputs

Input Type Default Range or choices What it does
images IMAGE Batch of frames to analyze.
preset choice Medium Fast, Medium, Ultra Solver effort. DIS: ultrafast, fast or medium preset (mostly a speed trade, similar accuracy). Lucas-Kanade: window radius 3, 5 or 9 px.
flow_scale float 1 0.1 to 10, step 0.1 Scale factor for output vectors. 1.0 = pixel units.
visualize boolean off Outputs a color-coded visualization of the motion field.
solver (optional) choice Auto Auto, DIS, Lucas-Kanade Auto uses DIS, and falls back to Lucas-Kanade only if OpenCV is missing. DIS holds a dense field out to about 20 px of motion and is roughly 11x faster. Lucas-Kanade is the previous solver: accurate to about 8 px, after which the field thins out even though the median stays close.

Outputs

Output Type
motion_vectors IMAGE
visualization IMAGE
stats STRING

Multipass AOV Reader

RadianceMultipassAOVReader

Multipass AOV Reader node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Read a real multilayer/AOV OpenEXR and split its named layers into the same passes as the Master extractor. Ground-truth renderer passes — not estimates. Missing layers come through black.

Inputs

Input Type Default Range or choices What it does
exr_path string Path to a multilayer/AOV EXR (Arnold/Redshift/Karma/Cycles/V-Ray).
beauty_layer (optional) string auto Override the beauty layer name, or 'auto' to detect (RGBA / 'beauty').
albedo_layer (optional) string auto Override the albedo/diffuse layer name, or 'auto'.
normal_layer (optional) string auto Override the normal layer name, or 'auto'.
depth_layer (optional) string auto Override the depth (Z) layer name, or 'auto'.

Outputs

Output Type
passes RADIANCE_PASSES
beauty IMAGE
albedo IMAGE
normal_map IMAGE
depth IMAGE
roughness IMAGE
specular IMAGE
metallic IMAGE
ao IMAGE
emission IMAGE
transmission IMAGE
highpass IMAGE
world_position IMAGE
curvature IMAGE
shadow_mask IMAGE
midtone_mask IMAGE
highlight_mask IMAGE
reflection_mask IMAGE
motion_vector IMAGE
segmentation_id IMAGE
motion_visualization IMAGE
alpha IMAGE

Multipass Composite

RadianceMultipassComposite

Multipass Composite node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Alpha-over composite of a foreground onto a background with optional depth holdout, background shadow and light wrap. Use it to put a (relit) CG element into a plate; feed scene-linear images for correct edges.

Inputs

Input Type Default Range or choices What it does
foreground IMAGE Foreground colour. Sets the output size and batch; its pixels are ignored when relit_foreground is connected.
alpha IMAGE Foreground matte, first channel, 0..1 (1 = foreground fully covers the background).
background (optional) IMAGE Background plate, resized to the foreground. Black when not connected.
relit_foreground (optional) IMAGE Replaces the foreground pixels (for example the relit output of Multipass Relight). Same premultiplication as set by premultiplied_input.
foreground_depth (optional) IMAGE Foreground depth pass, first channel. Used for the depth holdout only when background_depth is also connected.
background_depth (optional) IMAGE Background depth pass in the same units and polarity as foreground_depth. Background pixels nearer than the foreground hold it out.
shadow_mask (optional) IMAGE Shadow cast onto the background, first channel, 1 = full shadow. Darkens the background by shadow_strength before the over.
alpha_invert (optional) boolean off Use 1 - alpha as the matte. Applied after unpremultiplying, which still uses the original alpha.
premultiplied_input (optional) boolean off Set on when the foreground is already multiplied by alpha; it is divided by alpha first to avoid a double premultiply.
depth_near_is_white (optional) boolean on Depth polarity of both depth passes. On: larger values are nearer. Off: smaller values are nearer (distance style Z).
depth_bias (optional) float 0.01 -0.25 to 0.25, step 0.001 Depth tolerance in depth-pass units. Positive values keep the foreground in front when depths are nearly equal, negative favours the background.
shadow_strength (optional) float 0.35 0 to 1, step 0.01 How much shadow_mask darkens the background: 0 = none, 1 = black where the mask is 1.
light_wrap (optional) float 0 0 to 1, step 0.01 Amount of blurred background added to the foreground's soft edges. 0 = off; needs background connected.
light_wrap_radius (optional) int 8 0 to 64, step 1 Box blur radius in pixels for the wrapped background. The edge band uses half this radius.

Outputs

Output Type
composite IMAGE
premultiplied_foreground IMAGE
holdout_mask IMAGE
depth_matte IMAGE
comp_info STRING

Multipass Estimate

RadianceMultipassEstimate

Multipass Estimate node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Estimate render passes from a plate with trained models: MoGe-2 for metric geometry (depth, position, normals, AO, curvature) and Marigold IID for materials (albedo, roughness, metallic) and lighting (diffuse and specular light). No image-filter guesses.

Inputs

Input Type Default Range or choices What it does
beauty IMAGE The plate. One image or a batch of frames.
beauty_encoding choice sRGB (display) sRGB (display), Linear (scene) How the plate is encoded. Ordinary images and video frames are sRGB. Choose Linear for scene-linear EXR plates; values above 1.0 are clipped for the models and left out of the lighting fit.
geometry boolean on Depth, position, normals, AO and curvature (MoGe-2, 662 MB).
materials boolean on Albedo, roughness, metallic (Marigold IID Appearance, 2.5 GB).
lighting boolean on Diffuse and specular lighting (Marigold IID Lighting, 1.7 GB; shares files with Appearance).
download_missing_models boolean on Fetch missing weights from Hugging Face on first run (pinned versions). RADIANCE_ALLOW_DOWNLOADS=0 or HF_HUB_OFFLINE=1 always blocks this.
prev_frame (optional) IMAGE Previous frame for motion vectors of a single image.
moge_model (optional) MOGE_MODEL Optional: a model from ComfyUI's Load MoGe Model. Must be MoGe-2 for normals.
geometry_detail (optional) int 9 0 to 9 MoGe resolution level: 0 fastest, 9 most detail.
fov_x_degrees (optional) float 0 0 to 170, step 0.1 Known horizontal field of view. 0 = estimate it.
normal_convention (optional) choice OpenGL (Y-Up) OpenGL (Y-Up), DirectX (Y-Down) Encoding of the camera-space normal pass (0..1, +Z toward the camera). DirectX (Y-Down) flips the green channel; match the renderer or tool reading it.
ao_radius_m (optional) float 0.5 0.01 to 50, step 0.01 Occlusion search radius in metres of scene space.
ao_quality (optional) int 8 2 to 32 AO slices and steps per slice.
material_steps (optional) int 4 1 to 50 Marigold denoising steps. 4 is the trained default.
material_resolution (optional) int 768 0 to 2048, step 64 Marigold processing size (long edge). 768 is the trained size; 0 = native.
ensemble_size (optional) int 1 1 to 10 Average several Marigold predictions. Slower, steadier.
seed (optional) int 0 0 to 4294967295 Marigold noise seed, reused on every frame to limit flicker.
batch_is_sequence (optional) boolean on Treat the batch as consecutive frames for motion vectors.

Outputs

Output Type What it is
passes RADIANCE_PASSES Every estimated pass plus the linear beauty, for Write EXR Passes and Relight.
beauty IMAGE The plate, scene-linear.
albedo IMAGE Base colour, scene-linear (Marigold IID Appearance).
roughness IMAGE Roughness 0..1 (Marigold IID Appearance).
metallic IMAGE Metallic 0..1 (Marigold IID Appearance).
diffuse_lighting IMAGE Diffuse light contribution, scene-linear, scaled to the plate (Marigold IID Lighting).
specular_lighting IMAGE Non-diffuse light (reflections, highlights), scene-linear, scaled to the plate.
normal_map IMAGE Camera-space normals, 0..1 encoded (MoGe-2).
depth IMAGE Metric z depth in metres; 0 where there is no surface (MoGe-2).
world_position IMAGE Camera-space position in metres, OpenGL axes (MoGe-2).
ao IMAGE Ambient occlusion from the metric geometry, 1 = open.
curvature IMAGE Mean curvature in 1/metre, convex positive.
motion_vector IMAGE Backward optical flow to the previous frame in pixels, +y up.
geometry_mask IMAGE 1 where MoGe found a surface, 0 for sky and invalid pixels.
info STRING JSON: models, estimated field of view, lighting scale factors and fit error.

Multipass Relight

RadianceMultipassRelight

Multipass Relight node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Relights a CG or estimated plate from its albedo and normal passes with one directional or point light (GGX specular, Lambert diffuse), using roughness, metallic, AO and shadow passes when supplied. Use it to change key light direction or colour in comp without re-rendering; outputs are scene-linear.

Inputs

Input Type Default Range or choices What it does
albedo IMAGE Diffuse albedo (base colour) pass, scene-linear with no lighting baked in. Also used as the specular colour where metallic is 1.
normal_map IMAGE Normal pass: 0..1 encoded values are decoded to -1..1 automatically, signed passes are used as is. View direction is +Z toward the camera.
beauty (optional) IMAGE Original beauty render, scene-linear. Only used when mix_with_beauty is above 0.
roughness (optional) IMAGE Roughness pass, first channel, 0 = mirror, 1 = fully rough (clamped to 0.045 minimum). Defaults to 0.5 when not connected.
metallic (optional) IMAGE Metalness pass, first channel, 0 = dielectric, 1 = metal (no diffuse, albedo tints the reflection). Defaults to 0 when not connected.
specular (optional) IMAGE Dielectric specular level pass, 0..1, scaling the default 4% reflectance at normal incidence (1 = 4%). Defaults to 1 when not connected.
ao (optional) IMAGE Ambient occlusion as renderers write it: 1 = open, 0 = fully occluded.
alpha (optional) IMAGE Coverage matte, first channel, 0..1. Passed through to the alpha output and used for output_premultiplied; defaults to 1.
shadow_mask (optional) IMAGE Shadow pass, first channel: 1 = fully shadowed, 0 = lit. Blocks direct diffuse and specular light but not ambient.
depth_map (optional) IMAGE Normalised 0..1 depth pass. Only used by a Point light when world_position is not connected, to give pixels a synthetic Z.
world_position (optional) IMAGE Position pass in camera space metres (OpenGL axes: X right, Y up, +Z toward the camera). Only used by a Point light, and overrides depth_map.
normal_convention (optional) choice OpenGL (Y-Up) OpenGL (Y-Up), DirectX (Y-Down) Green channel convention of the normal pass. DirectX (Y-Down) flips the Y component before lighting.
light_type (optional) choice Directional Directional, Point Directional: light_x/y/z is a direction, no falloff. Point: light_x/y/z is a position, with soft distance falloff 1 / (1 + 0.08 d^2).
light_x (optional) float -0.35 -10 to 10, step 0.01 Light X (+ = screen right). Directional: component of the direction toward the light. Point: position in world_position units, or in screen units (X spans +/- aspect ratio) without it.
light_y (optional) float 0.45 -10 to 10, step 0.01 Light Y (+ = up). Directional: component of the direction toward the light. Point: position in world_position units, or in screen units (Y spans -1..1) without it.
light_z (optional) float 1 -10 to 10, step 0.01 Light Z (+ = toward the camera). Directional: component of the direction toward the light, so positive values light the front. Point: position along the camera axis.
light_r (optional) float 1 0 to 8, step 0.01 Linear red multiplier of the light colour. Also tints the ambient term.
light_g (optional) float 1 0 to 8, step 0.01 Linear green multiplier of the light colour. Also tints the ambient term.
light_b (optional) float 1 0 to 8, step 0.01 Linear blue multiplier of the light colour. Also tints the ambient term.
intensity (optional) float 1 0 to 20, step 0.05 Linear multiplier on the direct light (diffuse and specular). Does not affect ambient.
ambient (optional) float 0.03 0 to 4, step 0.01 Flat fill light level, tinted by the light colour and multiplied by the AO pass. Applied to diffuse only.
specular_intensity (optional) float 1 0 to 8, step 0.05 Linear multiplier on the specular term only. 0 gives a purely diffuse relight.
depth_scale (optional) float 10 0.01 to 1000, step 0.1 Z range, in screen units (image height = 2), that the 0..1 depth pass spans around mid grey. Only used for a Point light with depth_map and no world_position.
depth_near_is_white (optional) boolean on Set on when near pixels are white in depth_map. Only used for a Point light with depth_map and no world_position.
mix_with_beauty (optional) float 0 0 to 1, step 0.01 Linear blend of the relit result toward the beauty input: 0 = relight only, 1 = original beauty. Needs beauty connected.
output_premultiplied (optional) boolean off Multiply the relit output (and the beauty mix) by alpha. Off leaves it unpremultiplied.

Outputs

Output Type
relit IMAGE
diffuse_light IMAGE
specular_light IMAGE
lighting IMAGE
alpha IMAGE
relight_info STRING

Read Mask

RadianceLoadImageMask

Read Mask node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Load an image with optional non-destructive mask override. If a companion '_radmask.png' file exists alongside the source image, its alpha channel is used as the mask instead of the image's own alpha. The Radiance mask editor frontend saves masks in this companion format.

Inputs

Input Type Default Range or choices What it does
image choice files found in the matching models or input folder Image file from the ComfyUI input folder, loaded as 8-bit sRGB (0-1). If '_radmask.png' sits beside it, that file's alpha is the mask as-is; otherwise the image's own alpha is inverted, as in Load Image.

Outputs

Output Type
image IMAGE
mask MASK

Relight Engine

RadianceRelightEngine

Relight Engine node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Add a directional light (Lambert + Blinn-Phong) to an image from its normal map.

Inputs

Input Type Default Range or choices What it does
image IMAGE Image to light. The lighting pass is added on top of it, not multiplied by an albedo.
normal_map IMAGE Normal map encoded 0 to 1 (RGB = XYZ x 0.5 + 0.5). Resized to the image if the sizes differ.
light_dir_x float 1 -2 to 2, step 0.01 Light direction X component. Normalized internally — sets the horizontal angle of the synthetic light.
light_dir_y float 1 -2 to 2, step 0.01 Light direction Y component. Positive = light from above.
light_dir_z float 1 -2 to 2, step 0.01 Light direction Z component. Positive = light in front of surface.
light_color_r float 1 0 to 2, step 0.01 Red component of the synthetic light color. Values > 1.0 produce HDR emission.
light_color_g float 1 0 to 2, step 0.01 Green component of the synthetic light color.
light_color_b float 1 0 to 2, step 0.01 Blue component of the synthetic light color.
diffuse_intensity float 1 0 to 10, step 0.01 Lambertian diffuse reflection strength. Controls broad, soft illumination.
specular_intensity float 0.5 0 to 10, step 0.01 Specular highlight strength. Higher values produce brighter, more visible glints.
specular_roughness float 0.1 0.01 to 1, step 0.01 Surface roughness for Blinn-Phong specular. Low = sharp glints (metallic), high = soft broad highlights (matte).
camera (optional) RADIANCE_CAMERA Optional camera. Only the position in its transform is used, as the viewpoint for the specular term relative to a -1 to 1 image plane. Without it a viewer at z = 2 is assumed.

Outputs

Output Type
image IMAGE
lighting_pass_only IMAGE

Rolling Shutter

RadianceRollingShutter

Rolling Shutter node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Simulate rolling shutter artifacts (skew, wobble, flash banding).

Inputs

Input Type Default Range or choices What it does
image IMAGE Image or frame batch. Every frame gets the same static distortion.
skew_amount float 0 -50 to 50, step 1 Total shear in pixels between the first and last scanline; the sign sets the lean. 0 = no skew.
shutter_direction (optional) choice Vertical Horizontal, Vertical, Both Vertical: scan runs top to bottom, rows shift sideways. Horizontal: columns shift up/down. Both: half of each.
wobble_frequency (optional) float 0 0 to 20, step 0.5 Jello wobble cycles per half frame along the scan axis (2x this over the full frame). Needs wobble_amplitude above 0.
wobble_amplitude (optional) float 0 0 to 20, step 0.5 Peak-to-peak wobble displacement in pixels (half that in Both mode). 0 = off.
flash_band_position (optional) float -1 -1 to 1, step 0.05 Centre of a partial-exposure flash band along the scan axis, -1 = top/left edge, 1 = bottom/right. Any value below -0.5 turns the band off.
flash_band_width (optional) float 0.1 0.01 to 0.5, step 0.01 Gaussian width of the flash band as a fraction of half the frame. The band adds a flat +0.3 to all channels.
use_gpu (optional) boolean on Run the effect on GPU via CUDA/MPS. Falls back to CPU if unavailable.

Outputs

Output Type
image IMAGE

SAM Multi-Mask Picker

RadianceMultiMaskVisualPicker

SAM Multi-Mask Picker node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Pick one mask from a stack of candidate masks (for example the several outputs of a segmenter) by index.

Inputs

Input Type Default Range or choices What it does
masks MASK Candidate masks. A standard MASK batch (N, H, W) is treated as N candidates for one frame; a 4-D (B, N, H, W) stack picks the same candidate on every frame.
picker_index int 0 0 to 5, step 1 Zero-based index of the candidate to output. Wraps around if it is past the number of candidates.

Outputs

Output Type
selected_mask MASK

Scene Cut Detect

RadianceSceneCutDetect

Scene Cut Detect node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Automatically detect scene cuts in a video using visual change metrics.

Inputs

Input Type Default Range or choices What it does
images IMAGE Full video sequence as IMAGE batch.
cut_confidence float 0.5 0.01 to 0.99, step 0.01 How different two frames must be to count as a cut, on a 0–1 scale that means the same thing for every method. 0.5 is a textbook hard cut; lower is more sensitive. Fixed, not relative to the clip, so cut-free footage reports no cuts.
min_shot_frames int 12 1 to 500 Minimum frames between detected cuts.
method choice combined histogram, edge, combined histogram: colour distribution diff (fast). edge: Sobel edge map diff (catches content cuts). combined: weighted blend of both (recommended).

Outputs

Output Type
cut_data STRING
shot_count INT
score_plot IMAGE

Scene Cut Split

RadianceSceneCutSplit

Scene Cut Split node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Split a video at detected scene cut points into discrete segments.

Inputs

Input Type Default Range or choices What it does
images IMAGE The same frame batch that was analysed by RadianceSceneCutDetect. Returned whole if cut_data has no shots.
cut_data string JSON from RadianceSceneCutDetect.
shot_index int 0 0 to 9999 Which shot to extract (0-based). Connect shot_count output to know the range.

Outputs

Output Type
frames IMAGE
shot_index INT
start_frame INT
end_frame INT
shot_info STRING

Shot Grade Router

RadianceShotGradeRouter

Shot Grade Router node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Route frames to per-shot grade nodes based on scene cut metadata.

Inputs

Input Type Default Range or choices What it does
shot_index int 0 0 to 9999 Current shot index from SceneCutSplit.
grade_table string [ { "exposure": 0.0, "temperature": 0, "saturation": 1.0, "contrast": 1.0 }, { "exposure": 0.3, "temperature": -150, "saturation": 0.9, "contrast": 1.1 } ] multi-line text JSON array of per-shot grade parameters.

Outputs

Output Type
exposure FLOAT
temperature FLOAT
saturation FLOAT
contrast FLOAT
shot_index INT

Stabilize

RadianceSubpixelStabilizer

Stabilize node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Stabilise an image sequence to one anchor frame using sub-pixel FFT phase correlation. Corrects translation only (no rotation or scale); also outputs the per-frame shift as an image (red = x, green = y, in pixels).

Inputs

Input Type Default Range or choices What it does
image IMAGE Image sequence to stabilise. Shifts are measured on the channel average of each frame.
anchor_frame int 0 0 to 1000, step 1 Zero-based index of the frame every other frame is aligned to. Values past the end use the last frame.
max_shift int 64 4 to 512, step 4 Largest correction applied, in pixels per axis. A larger measured shift is clamped to this rather than rejected.

Outputs

Output Type
stabilized_sequence IMAGE
displacements_xy IMAGE

Temporal Stitch Stabilizer

RadianceTemporalStitchStabilizer

Temporal Stitch Stabilizer node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Smooth a mask sequence over time with a per-pixel Gaussian filter across frames, to remove edge jitter and popping before stitching.

Inputs

Input Type Default Range or choices What it does
masks MASK Mask sequence, one mask per frame in timeline order. A single mask is returned unchanged.
temporal_sigma float 2 0 to 16, step 0.5 Gaussian width in frames; the filter reaches 3 x sigma frames each way. Values of 0.1 or less bypass the node.

Outputs

Output Type
stabilized_masks MASK

Vignette

RadianceVignette

Vignette node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Apply a natural or stylised vignette to the image edges.

Inputs

Input Type Default Range or choices What it does
image IMAGE Image to vignette. RGB is multiplied by the falloff, so linear and display-encoded images both work; alpha is untouched.
strength float 0.5 0 to 1, step 0.01 Vignette intensity at corners. 0=no effect, 1=full black.
power float 2 0.5 to 8, step 0.1 Falloff exponent. 2.0 = natural cos⁴ approximation. Higher = harder edge.
center_x float 0.5 0 to 1, step 0.01 Horizontal center of the vignette (0 = left, 1 = right).
center_y float 0.5 0 to 1, step 0.01 Vertical center of the vignette (0 = top, 1 = bottom).
feather float 1 0.1 to 4, step 0.05 Radial feather: scales the normalized radius. >1 = pushes falloff inward.
tint_r (optional) float 1 0 to 2, step 0.01 Red multiplier for vignetted (dark) areas. 1.0=neutral.
tint_g (optional) float 1 0 to 2, step 0.01 Green channel multiplier for vignette tint. 1.0 = neutral.
tint_b (optional) float 1 0 to 2, step 0.01 Blue multiplier for vignetted areas. >1 = cool shadow edges.

Outputs

Output Type What it is
image IMAGE Vignetted image.
vignette_mask IMAGE Grayscale vignette mask [0,1] — wire to downstream grades or multiply nodes.

Write EXR Passes

RadianceEXRPassesWriter

Write EXR Passes node in Radiance 3.5.0, default interface
Captured from Radiance 3.5.0 · Default interface · Select to open the full-resolution screenshot. Controls can change with presets and connections.

Write all passes inside the RADIANCE_PASSES bundle into a single-part multilayer or true multi-part OpenEXR file.

Inputs

Input Type Default Range or choices What it does
passes RADIANCE_PASSES Pass bundle to write; must contain a beauty pass, and every pass must match its width and height. Pixel values are written as is, with no colour conversion.
filename_prefix string radiance_vfx_passes File name stem, not a path. Files are named ..exr with a 4-digit padded frame number.
bit_depth choice 16-bit Half Float 16-bit Half Float, 32-bit Float EXR pixel type. Files containing depth, world_position, motion_vector or object_id passes are always written as 32-bit float.
compression choice ZIP ZIP, ZIPS, PIZ, RLE, Uncompressed, PXR24, B44, B44A, DWAA, DWAB EXR compression. ZIP, ZIPS, PIZ, RLE and Uncompressed are lossless; lossy B44/B44A/DWAA/DWAB raise an error when data passes (depth, position, motion, object ID) are present.
output_path (optional) string Output folder. Empty = ComfyUI output folder; a relative path is a subfolder of it; an absolute path is used as is. Created if missing.
remote_path (optional) string Optional absolute folder (for example a NAS share) that each written file is also copied to. A failed copy only logs a warning.
frame_index (optional) int 1001 0 to 999999 Frame number of the first image in the batch; later images count up from it.
exr_layout (optional) choice Single-part multilayer Single-part multilayer, Multi-part Single-part multilayer: all passes as named channel layers in one part (widest compatibility). Multi-part: one EXR 2.0 part per pass.
custom_metadata (optional) string multi-line text Extra EXR header attributes, one key=value per line, stored as strings. Lines without = are ignored.

Outputs

Output Type
output_path STRING