Image effects

Grading, filtering, lens distortion and frequency work, wired into the graph as nodes that take a picture in and give a picture out.

An image effect is an ordinary node. A picture goes into content_in, the result leaves on content_out, and effects chain in whatever order you wire them. Any point in the chain can be branched, inspected, keyframed or bypassed.

That is what separates them from Post effects, which are toggles on the Render Settings node and run over the finished frame in a fixed order. An image effect runs where you put it, on whatever is on the wire, including content that never went near the 3D renderer.

They sit in the Modify family in the node palette, under its Color and Filter tags.

Grading and colour

Five nodes cover the grade, from broad correction down to a single hue. Unlike the filters further down, they are point-wise, so the same maths applies whether the input is an image, a volume, an analytic shape's fill, a collection, or one flat colour.

• Colour Correct is the full grade in one node: lift_*, gamma_*, gain_* and offset_* per channel, then saturation and contrast around a contrast_pivot of 0.5. It also takes hand-drawn curves. Wire a Curve node into master_curve, or into red_curve, green_curve, blue_curve and alpha_curve for per-channel shaping. clamp_output is on, so results stay in range. See Colour Correct.

• Levels is the black-point, white-point and gamma remap. in_black and in_white set what becomes pure black and pure white, gamma moves the midtones, and out_black and out_white remap the result into a narrower range. channel_mode switches from Master to Per-Channel, which swaps in the r_*, g_*, b_* and a_* sets. invert flips the result, and pre_clamp and post_clamp are both on.

• HSV Grade is the selective grade. target picks whether a Hue, Saturation or Value band is isolated, target_hue with hue_width and hue_falloff says which band, and range_hue_shift, range_saturation and range_lightness move only what falls inside it. The three master_* controls move everything, so one node can lift the whole image and push a single colour at the same time. Its adjustment parameters also take a Curve, for per-hue control.

• Channel rewires R, G, B and A. mode chooses Mix, a matrix built from red_row, green_row, blue_row and offset; Shuffle, where out_red and its siblings each pick a source channel or a flat Zero or One; Extract, which pulls one channel or Luminance out as grey; and Premultiply or Unpremultiply for handling alpha. For hue, saturation and luminance as separate values, use the Split / Combine node instead.

• Color Space converts between sRGB, Linear sRGB and ACEScg through source_space and target_space. Those three are the whole set: it does not convert to HSV, HSL or LAB.

The first entry in both of Color Space's dropdowns, and the default on source_space, is Working Space. It is another name for Linear sRGB: it does not read the project's setting, so on a project whose Canvas color_space is ACEScg it converts the wrong way. Pick a space by name. See Colour management.

Sharpening, cleaning up and edges

• Sharpen is a variable-radius unsharp mask. amount is a percentage, 100 by default, radius sets how wide the halo is, and threshold holds back flat areas so grain is not sharpened along with detail. It also works on a volume, sharpening in three dimensions. See Sharpen.

• Denoise softens noise with a fixed 3×3 Gaussian, mixed back over the original by strength. Its method dropdown offers Median, Bilateral, Non-Local Means, FFT Gate and Wavelet, and radius and threshold sit beside it. None of those three controls is read: every combination gives the same soft blur, and strength is the only one that changes the result.

• Paint flattens detail into patches. Each pixel takes nine taps arranged as four overlapping quadrants and keeps the quadrant whose luminance varies least, so an even area smears into one flat patch while an edge stays where it is. size runs 1 to 4 and spreads those same nine taps further apart rather than adding more, so patches broaden at no extra cost, and strength mixes the result back over the original. The filter is isotropic: a patch does not curve to follow a contour. Changing size rebuilds the filter while strength is a live value, so strength is the one to drag or keyframe. See Paint.

• Slope finds edges. method picks Sobel, Prewitt or Laplacian, and output decides what leaves the node: Slope, Magnitude, Direction, or the raw X-Gradient and Y-Gradient. Fed 3D content rather than an image, source can read the rendered Depth, Normal or Object ID pass, which finds geometric edges no colour image shows. The other two entries in that list are not implemented: Color silently reads depth and Combined silently reads normals, so depth_weight, normal_weight and id_weight do nothing.

• Emboss shades an image as relief. angle places the light, depth scales the effect, blend mixes the relief back over the original, and method picks a Gradient or Convolution kernel. elevation applies to volumes only.

• Matte Refine cleans an alpha channel. erode, dilate and blur run in that order, tightening a matte then feathering it, and channel reads either Alpha or Luminance. Erode and dilate are square rather than round, so a corner tightens faster than an edge. Wiring a field into erode_field or dilate_field varies the amount across the frame, with the numeric control as the ceiling. All three radii at 0 is an exact passthrough, which also skips the Luminance conversion.

Lens and projection

• Lens Distort bends the image the way a lens does. mode is Barrel, Fisheye or Anamorphic; k1, k2 and k3 are the barrel coefficients, fov and fisheye_model shape the fisheye, and squeeze sets the anamorphic ratio. undistort runs the same maths backwards, which is how footage shot on a known lens is straightened before compositing.

• Panoramic remaps between 360 and panoramic layouts. projection covers Mirror Ball, Mercator, Stereographic, Sinusoidal, Mollweide, Polar, Cylindrical, Gnomonic, Panini, Octahedral, Dual Paraboloid and Cubemap Cross, rotation spins the sphere, and invert runs the projection the other way. Use it to turn an HDRI into a mirror ball, or a rendered equirectangular frame into a flat crop. It is a different node from Projection, which places content in 3D space rather than remapping pixels.

Both carry edge_mode, deciding what fills the area a warp pulls in from outside the frame: Transparent, Clamp, Repeat or Mirror. Both are image-only.

Frequency, pattern and erosion

• FFT filters in the frequency domain, transforming the picture, masking the frequencies and transforming back. mode is Low Pass, High Pass, Band Pass, Band Reject or Phase Only, with cutoff and cutoff_high setting the band and falloff choosing a Sharp, Gaussian or Butterworth rolloff. This is how a repeating pattern is removed when no spatial blur can separate it from the detail. spectrum_out carries the frequency picture after filtering, so the cutoffs are visible; the sixth mode, Analyze, puts that view on the main output and skips the filtering.

• Convolve applies one image to another as a filter kernel. Wire the filter into kernel_in, leave normalize_kernel on to preserve brightness, and set mode to Correlate rather than Convolve to match a pattern instead of blurring by it. The same node convolves audio with an impulse response, which is how a recorded space becomes a reverb. See Audio.

• Quantize posterises. The two halves have a switch each: quantize_value for the colour steps, on to start with, and quantize_position for the mosaic. value_levels, 8 by default, sets how many steps survive, and value_mode applies them Per Channel, on Luminance, or as a hard Threshold. Threshold ignores the levels and splits to black and white at value_threshold, a luminance pivot you can move, with value_threshold_softness feathering the split from crisp to soft. Turning on quantize_position snaps pixels into blocks instead of colours, which is the mosaic look, and that half needs a Cells node wired into cells to say what shape the blocks are. Picking a preset on that input adds the Cells node and wires it for you. Without the wire, quantize_position changes nothing. sample_mode then decides what colour a cell takes: Center, Top-Left and Average read the source once per cell, Average box-filtering it, while Cell UV fits the whole frame into every cell, sampled from a mip chain so a frame inside an 8-pixel cell holds still instead of shimmering. Fed shapes or points rather than an image, the node snaps their positions to a world grid instead, sized by block_size, or by block_size_x and block_size_y with separate_xy on; those controls show only on that arm.

• Erode is hydraulic and thermal terrain erosion, not a matte operation. It runs iterations passes of rainfall over the image read as a height field, carving channels and depositing sediment, and reports flow, sediment and erosion_delta alongside the result. Wire an image into mask to say where it may cut: white is full erosion, black is protected, and an unwired pin erodes everywhere. To shrink a matte, use Matte Refine.

Turning an image into something else

• Normal Map reads an image as a height map and writes surface normals, for lighting a flat picture or driving a displacement. strength scales the slope, and invert and flip_y match whichever convention the thing downstream expects.

• Image Trace converts a picture to vector paths. mode is Silhouette for one outline at threshold, or MultiThreshold for levels nested bands, and smoothness and min_area control how much noise survives the trace. It reads 3D content as well, taking the silhouette from the rendered frame. The outlines leave on shapes, ready to sweep, extrude or animate.

• Stack collects several images into one sequence, ready to hand an Iterator a different frame per copy. Wire the images into inputs. Its folder_path and the settings beside it list a directory onto the paths output as text and stop there, so a folder of stills has to be loaded and wired in.

How strongly an effect applies

Every effect carries strength, a 0 to 1 wet and dry mix. Most also carry a spatial_strength input that takes a field, so the effect can be full strength in one part of the frame and absent in another. Convolve, Erode, Color Space, Image Trace and Stack are the ones without it.

See also

• Post effects

• Rasters (images)

• Rasterizing

• Colour management