image自动生成

本页由源码自动生成

数据来自 entelechy serve + {"op":"kernels"} / {"op":"kernel"} · packages/image.entpkg/kernels/*.ent。下方的详细描述保留源码中的英文原文。

分类：包 · 清单：use "kernels/";

简述

对图片的处理：色彩空间、混合与遮罩、对比、滤波、色调曲线、缩放锐化、抗锯齿、直方图、距离场、mipmap、像素画。

描述

What Entelechy does to a picture, knowing nothing of the world it shows: color spaces, blending and masks, two pictures compared, filters, tone and its curves, scaling and sharpening, antialiasing within a frame, histograms, distance fields, mipmaps, pixel art and sprite sheets. Each is a word a kernel calls on a pixel, its kernel a thin pass.

Layer: Graphics. Uses: random.

use "image.entpkg";

Kernels

• kernels/antialias.ent — Antialiasing within one frame (FXAA, fxaa). Kernels: image.fxaa@1.
• kernels/blend.ent — Blending, and any picture kept to part of another by a mask. Kernels: image.blend@1, image.masked@1.
• kernels/color.ent — Color as numbers. Kernels: image.encode@1, image.decode@1.
• kernels/distance.ent — Distance. Kernels: image.seeds@1, image.first_step@1, image.next_step@1, image.flood@1, image.distance@1.
• kernels/compare.ent — Two pictures compared: each pixel’s difference (color_apart), and how far apart they are as a whole, the most a pixel differs and how many differ by more than a step. Kernels: image.difference@1, image.apart_rows@1, image.apart@1.
• kernels/filter.ent — Filters. Kernels: image.blur@1, image.shrink@1, image.convolve@1, image.sharpen@1, image.edges@1, image.morph@1, image.resize@1.
• kernels/histogram.ent — Histograms. Kernels: image.histogram@1, image.cumulate@1, image.equalize@1.
• kernels/mipmap.ent — A mip strip. Kernels: image.mip_first@1, image.mip_next@1, image.mip_halve@1.
• kernels/pixel_art.ent — Pixel effects. Kernels: image.pixelate@1, image.palette@2, image.ramp@1, image.dither@1, image.outline@1, image.shadow@1.
• kernels/sprite_sheet.ent — Sheets.
• kernels/tone.ent — Tone and color grading. Kernels: image.adjust@1, image.levels@1, image.curve@1, image.neutral@1, image.lut@1.
• kernels/tone_map.ent — Tone curves, ACES’s fit and AgX, light of any brightness to 0..1 (tone_curve), sharpened against its neighbors (tone_sharpened), made vivid and warm (tone_warmth).
• kernels/upscale.ent — A picture scaled to another size, edge-aware (FSR 1’s EASU, upscaled) or nearest-neighbor (upscaled_near), and a sharpening that cannot overshoot (RCAS). Kernels: image.upscale@1, image.contrast_sharpen@1.
• kernels/wrap.ent — Wrap addressing for a linear read.

Graph presets

• graphs/blur.esn — A Gaussian blur, both ways, as a sub-graph.
• graphs/compare.esn — Two pictures compared: the difference to look at and how far apart, as a test holds a scene to another’s picture or a person sees where two renders part (two pictures read as inputs, then entelechy replay with show apart.apart[0]).
• graphs/distance.esn — The signed distance from every pixel to a shape, in pixels, negative inside.
• graphs/equalize.esn — Histogram equalization as a sub-graph.
• graphs/flood.esn — Jump flooding as a sub-graph.
• graphs/flood_pass.esn — One jump-flooding pass and the step after it.
• graphs/mipmap.esn — The mip strip of a square texture, as a sub- graph.
• graphs/mipmap_level.esn — One halving of a mip strip.
• graphs/sharpen.esn — Unsharp masking as a sub-graph.
• graphs/upscale.esn — Scaled to the window and sharpened.

Tests

tests/ holds scenes and their replays, run on every backend: antialias.replay, compare.replay, looks.replay, operations.replay, upscale.replay. Every Kernel of the package is reached by one of them.

Used by the darkroom, image, mipmap, nebula examples.

Kernel

Kernel	执行模式	简述
image.adjust@1	map	The picture adjusted: exposure stops, contrast (1 none), saturation (1 none, 0 gray), hue turned (a whole turn is 1).
image.apart@1	map	How far apart two pictures are as a whole (domain = 1): the most a pixel differs (x) and how many differ by more than the rows’ step (y), from image.apart_rows.
image.apart_rows@1	map	Each row of a difference (image.difference’s), one cell a row: the most a pixel differs (x) and how many differ by more than over (y). A node sets domain = size(difference) * vec2(0, 1) + vec2(1, 0).
image.blend@1	map	top composited over base, as the W3C’s compositing spec does it: where top covers (its alpha times opacity), its color is blended with base’s by mode, as much as base is there; then laid over base, the alphas joined as one sheet over another, the color weighed by both. space is where the arithmetic is done: on sRGB’s stored values (Encoded), as painting programs blend and the modes are defined, or on light itself (Linear). top is stretched over base whatever its size.
image.blur@1	map	A Gaussian blur along axis (one pass of the two a blur takes, graphs/blur.esn): sigma measured in the output’s pixels, taps one output pixel apart out to three sigmas.
image.contrast_sharpen@1	map	Sharpened once (amount 0 not at all, 1 the most): pulled from its four neighbors by an amount held between the darkest and brightest of the five, never past 0 to 1, with no white rims.
image.convolve@1	map	Any convolution: weights holds n by n numbers, row by row, centered on the pixel (n odd). The weights are used as given, not normalized.
image.cumulate@1	prefix with sum	The cumulative count: at each bin, the pixels in it and below.
image.curve@1	map	Curves: each channel looked up in curve, rows of the output for inputs spread evenly from 0 to 1 (x for red, y green, z blue), blended between rows.
image.decode@1	map
image.difference@1	map	The difference of a and b (domain = size(a); b read at the same place across the picture, whatever its size): each channel’s, and in w the most of them.
image.distance@1	map	The signed distance, in pixels, from each pixel’s center to the shape’s edge: from the nearest pixel of the shape when outside it (inside’s flood), less the distance to the nearest pixel outside it when within (outside’s flood); half a pixel taken off each, so the edge lies between the pixels either side. Where a flood found no seed, reach.
image.dither@1	map	Each pixel held to colors with ordered dithering (dithered_color): a gradient becomes a pattern of two colors.
image.edges@1	map	Edges: the strength of the Sobel gradient, gray, opaque.
image.encode@1	map	Linear light to sRGB’s stored values (for a window or a file), and back; alpha kept.
image.equalize@1	map	The picture equalized: each pixel’s lightness moved to the share of the pixels darker than it (cumulative, from image.cumulate), read between the counts below and through its bin by where in the bin it falls, so a smooth gradient stays smooth. The color is scaled to reach it: lightness grows as the cube root of a scale, so the cube of the ratio, which keeps the hue. amount mixes it with the source.
image.first_step@1	map	The first step: half the picture’s longer side, as a power of two.
image.flood@1	map	One pass: the nearest seed among this pixel’s and the eight a step away.
image.fxaa@1	map	The HDR picture image antialiased (the same size, fxaa); on 0 leaves it as it is.
image.histogram@1	scatter into input 1 with sum	How many pixels of the source fall in each bin.
image.levels@1	map	Levels: black to 0 and white to 1, clamped, then bent by gamma (above 1 brightens the middle).
image.lut@1	map	The picture through a 3D LUT (lut_at), its values taken 0 to 1.
image.masked@1	map	top over base as far as mask says: 0 base, 1 top, between them mixed (domain = size(base); top and mask any size, read smoothly across the picture). Any pass kept to part of a picture: a grade, an outline, a blur, laid only where a mask asks.
image.mip_first@1	map	The strip before any halving: the texture in the top square and nothing under it.
image.mip_halve@1	map	One halving: four texels of the level above averaged into one of the level named, everything else carried over.
image.mip_next@1	map
image.morph@1	map	Dilation (each channel the greatest within radius pixels, a disc) or erosion (the least): bright shapes grown or worn away.
image.neutral@1	map	The neutral LUT of side n (domain = vec2(n * n, n)): each color itself. Grade it as a picture and image.lut applies that grade.
image.next_step@1	map	The next step: half this one, never below one.
image.outline@1	map	A line of color round what is drawn: each pixel not drawn beside one that is, across a side, or across a corner too (corners).
image.palette@2	map	Each pixel the color of colors nearest it.
image.pixelate@1	map	Cells of cell pixels, each the color at its center.
image.ramp@1	map	Each pixel the color of colors, rows from dark to light, its lightness falls to (ramp_color).
image.resize@1	map	The source at the node’s domain, by bicubic (Catmull-Rom) interpolation: sharper than at_linear when enlarging. For halving and beyond, image.shrink.
image.seeds@1	map	Seeds: where the mask is on (its luminance times alpha past threshold, or not past it when inside is false), the pixel’s own center, else none. A seed is (x, y, 1, 0); none is all zeros.
image.shadow@1	map	A shadow of color cast by what is drawn, offset pixels away, under it: each pixel not drawn whose pixel offset back is.
image.sharpen@1	map	Unsharp masking: the source pushed away from its blurred copy by amount; alpha kept. graphs/sharpen.esn places it with its blur.
image.shrink@1	map	The source smaller: each output pixel the mean of the source pixels it covers (a box filter, exact at whole ratios).
image.upscale@1	map	Scaled to domain: edge-aware (upscaled), nearest-neighbor when near is 1 (upscaled_near).

方法

函数	简述
fxaa_luma(c: Vec3) -> Float	How bright it looks (0 to 1): brightness pressed into 0 to 1 and square-rooted, so a step in the dark counts as one in the light.
fxaa_at(image: Image, q: Vec2) -> Float
fxaa_stride(i: Int) -> Float	How far the ith step along an edge goes (pixels): a pixel at a time first, longer strides further on.
fxaa_walk(image: Image, q: Vec2, across: Vec2, dir: Vec2, edge: Float, gradient: Float) -> Vec2	From pixel q along dir, each step reading the mean brightness of this row and the row across (the brightness on the edge’s middle), until it differs from edge by a quarter of gradient: how far it went (x), and the difference in brightness there (y).
fxaa(image: Image, q: Vec2, threshold: Float, least: Float, subpixel: Float) -> Vec4	Pixel q of the HDR picture image antialiased: where the contrast is under threshold (of the brightest, at least least) it is left; subpixel is how much of what is finer than a pixel is smoothed (0 none).
blend_channel(mode: Blend, base: Float, top: Float) -> Float	One channel of top laid over base by mode.
blend(mode: Blend, base: Vec3, top: Vec3) -> Vec3	top laid over base by mode, channel by channel.
luminance(c: Vec3) -> Float	How bright a linear color looks (Rec. 709’s weights).
srgb_decode(v: Float) -> Float	One channel of sRGB’s transfer: a stored value to linear light, and back.
srgb_encode(v: Float) -> Float
srgb_to_linear(c: Vec3) -> Vec3
linear_to_srgb(c: Vec3) -> Vec3
rgb_to_hsv(c: Vec3) -> Vec3	Hue (0 up to 1, red at 0), saturation and value, and back.
hsv_to_rgb(c: Vec3) -> Vec3
cbrt(v: Float) -> Float	A cube root that keeps the sign: pow’s answer, which a device may give a few units off, made exact by one step of Newton’s method.
linear_to_oklab(c: Vec3) -> Vec3	Linear light to Oklab (Ottosson): lightness, then the green-red and blue-yellow axes; equal steps in it look equal. And back.
oklab_to_linear(c: Vec3) -> Vec3
blackbody(heat: Float) -> Vec3	Glowing matter by how hot it is, 0 to 1: deep red, orange, yellow, white.
hue(h: Float) -> Vec3	A vivid hue round the wheel, 0 to 1: red, yellow, green, cyan, blue, magenta, and red again.
color_apart(a: Vec4, b: Vec4) -> Float	How far apart two colors are: the most any of red, green and blue differs.
sobel(source: Image, c: Vec2) -> Vec2	The Sobel gradient of the source’s luminance at c: how fast it grows along x and along y.
cubic_weights(t: Float) -> Vec4	Catmull-Rom’s weights for the four taps round a fraction t.
lightness(c: Vec3) -> Float	How bright a linear color looks, 0 for black to 1 for white.
bin_of(c: Vec3, bins: Int) -> Int	The bin a color falls in, of bins.
mip_start(k: Int, side: Int) -> Float
mip_level(strip: Grid<Vec4>, uv: Vec2, k: Int, side: Int) -> Vec4
mip_wanted(du: Vec2, dv: Vec2, side: Int, levels: Int) -> Float
mip_at(strip: Grid<Vec4>, uv: Vec2, wanted: Float, side: Int, levels: Int) -> Vec4
nearest_color(c: Vec3, colors: Rows<Vec3>) -> Vec3	The color of colors nearest c in Oklab, where equal distances look equal; the first of two as near.
ramp_color(c: Vec3, colors: Rows<Vec3>) -> Vec3	The color of colors, rows from dark to light, that c’s Oklab lightness falls to: lightness 0 to 1 cut into as many equal steps as there are rows, so a picture keeps its light and dark in any tint.
bayer4(p: IVec2) -> Float	The threshold of pixel p in a 4 by 4 Bayer matrix, from 1/32 to 31/32: neighbors apart, so a share of pixels over it spreads evenly.
dithered_color(c: Vec3, colors: Rows<Vec3>, p: IVec2) -> Vec3	c held to colors with ordered dithering at pixel p: of the two colors nearest it in Oklab, the farther is taken at the pixels whose threshold the share of the way c lies toward it passes.
drawn(source: Image, p: IVec2) -> Bool	Whether pixel p of source is drawn: its alpha at least a half.
sheet_cell(k: Int, columns: Int) -> IVec2	The cell of frame k in a sheet columns cells across.
sheet_texel(k: Int, columns: Int, cell: IVec2, p: IVec2) -> IVec2	The texel of a sheet of cells of cell pixels that pixel p of frame k is.
sheet_place(t: IVec2, columns: Int, cell: IVec2) -> SheetPlace	Which frame texel t of a sheet is in, and where in it: what a Kernel drawing a whole sheet asks of each texel.
sheet_pixel(sheet: Image, k: Int, columns: Int, cell: IVec2, p: IVec2) -> Vec4	Pixel p of frame k of sheet.
frame_at(seconds: Float, fps: Float, frames: Int, loops: Bool) -> Int	The frame of an animation of frames frames at fps frames a second that shows seconds after it starts: round again (loops) or held at its last.
adjusted(c: Vec3, exposure: Float, contrast: Float, saturation: Float, hue: Float) -> Vec3	Exposure in stops; contrast round middle gray (0.18), as a power of the light, so it suits light brighter than white too; saturation and hue turned in Oklab, so a change looks even across colors.
lut_at(lut: Image, c: Vec3) -> Vec3	A 3D LUT’s color for c: the LUT is a strip of n squares side by side, n*n wide and n high, red across each square, green down it, blue square by square; blended between squares and within them.
aces(x: Vec3) -> Vec3	ACES tone mapping, the fitted curve: light of any brightness to 0..1, as film takes it.
agx(x: Vec3) -> Vec3	AgX tone mapping (Troy Sobotka’s, Blender’s own view), in its minimal fitted form: light of any brightness to 0..1 (linear), taken into AgX’s inset primaries, its log2 between -12.47 and 4.03 stops eased by the fitted sigmoid, and out again; bright colors desaturate toward white as film’s do, rather than turning hue.
tone_sharpened(hdr: Image, p: Vec2, c: Vec3, amount: Float) -> Vec3	Light c at p of hdr sharpened amount (0 not at all): pulled apart from the mean of its four neighbors, less where the contrast is already high (no white rims).
tone_curve(c: Vec3, curve: Int) -> Vec3	Light c through a tone curve: 0 ACES’s fit, 1 none (light past 1 clipped, as Blender’s Standard view), 2 AgX (Blender’s own).
tone_warmth(c: Vec3, vivid: Float, warmth: Float, shadow_tint: Vec3, light_tint: Vec3) -> Vec3	A toned color c made vivid (1 as it is) and tinted warmth of the way from shadow_tint in its shadows to light_tint in its lights.
upscaled_near(image: Image, q: Vec2, dst: Vec2) -> Vec4	What lies at pixel q when image is scaled to dst, nearest-neighbor: each pixel made a block, nothing interpolated (pixel art’s look). With a whole multiple (camera.extent’s whole) the blocks are one size; the window’s edge of less than a block repeats the last column and row.
upscale_light(c: Vec4) -> Float
upscale_tap(image: Image, p: Vec2) -> Vec4
upscale_edge(w: Float, la: Float, lb: Float, lc: Float, ld: Float, le: Float) -> Vec3
upscale_weight(off: Vec2, dir: Vec2, stretch: Vec2, lob: Float, clip: Float) -> Float
upscaled(image: Image, q: Vec2, dst: Vec2) -> Vec4	What lies at pixel q when image is scaled to dst: twelve points of the source (the nearest four and their cross neighbors), the edge’s direction and strength found by brightness, a Lanczos-shaped kernel stretched along the edge and narrowed across it; the result held between the darkest and brightest of the nearest four, so edges do not ring. At one size it is left as it is. A Kernel drawing at the window’s size and reading a picture drawn smaller reads it this way (image.upscale does only this). Where the picture is larger than dst (supersampled), the block of it falling in this pixel (at most four points a side) is averaged.
at_linear_wrapped(tile: Grid<Vec4>, q: Vec2) -> Vec4

类型与结构体

名称	写法
Blend	type Blend = enum { Normal, Multiply, Screen, Overlay, SoftLight, Add, Difference, Darken, Lighten };
Space	type Space = enum { Encoded, Linear };
Weighed	struct Weighed { sum: Vec4; weight: Float; }
Morph	type Morph = enum { Dilate, Erode };
DitherPair	struct DitherPair { near: Vec3; near_d: Float; next: Vec3; next_d: Float; }
SheetPlace	struct SheetPlace { frame: Int; at: IVec2; }

图预设

名称	接口	说明
blur	input source: Image param sigma: Float = 2.0 output result <- down.result	A Gaussian blur, both ways, as a sub-graph. sigma is in the output’s pixels; domain smaller than the source shrinks as it blurs. Place it with node soft: graph “image.entpkg/graphs/blur.esn” { source <- x.result; sigma = 3.0; }
compare	input a: Image input b: Image param over: Float = 0.0078125 output difference <- difference.result output apart <- apart.result	Two pictures compared: difference, each pixel’s (w the most a channel differs), and apart, the most any pixel differs (x) and how many differ by more than over (y). Placed: node apart: graph “image.entpkg/graphs/compare.esn” { a <- before; b <- after.result; }
distance	input mask: Image param threshold: Float = 0.5 param passes: Int = 11 output result <- signed.result	The signed distance from every pixel to a shape, in pixels, negative inside: the shape is where mask is on (see image.seeds). Place it with node field: graph “image.entpkg/graphs/distance.esn” { mask <- shape.result; }
equalize	input source: Image param bins: Int = 256 param amount: Float = 1.0 output result <- even.result output counts <- counts.result	Histogram equalization as a sub-graph: the pixels counted into bins by brightness, the counts accumulated, and each pixel moved to its rank. Place it with node even: graph “image.entpkg/graphs/equalize.esn” { source <- x.result; amount = 0.7; }
flood	input seeds: Image param passes: Int = 11 output result <- flooded.result	Jump flooding as a sub-graph: every pixel given the nearest of seeds (from image.seeds). passes covers a picture whose longer side is up to 2^(passes - 1) pixels, with one pass at a step of one to spare. Place it with node near: graph “image.entpkg/graphs/flood.esn” { seeds <- s.result; passes = 11; }
flood_pass	input seeds: Image input step: Rows<Int> output result <- pass.result output step <- next.result	One jump-flooding pass and the step after it. flood.esn places this passes times.
mipmap	input source: Image param side: Int = 256 param levels: Int = 7 output result <- halved.result	The mip strip of a square texture, as a sub-graph. Place it with node strip: graph “image.entpkg/graphs/mipmap.esn” { source <- texture.result; side = 256; levels = 7; } and read it with the functions in mipmap.ent. A placement does not know which repetition it is, so the level being written rides beside the strip as a row of one Int and is carried on with it.
mipmap_level	input strip: Image input step: Rows<Int> param side: Int = 256 output result <- result.result output step <- next.result	One halving of a mip strip: the level the counter names is written from the level above it, every other texel carried over unchanged. mipmap.esn places this levels times.
sharpen	input source: Image param sigma: Float = 1.5 param amount: Float = 0.6 output result <- crisp.result	Unsharp masking as a sub-graph: the source pushed away from its blur of sigma pixels by amount. Place it with node crisp: graph “image.entpkg/graphs/sharpen.esn” { source <- x.result; amount = 0.8; }
upscale	input picture: Image param amount: Float = 0.8 param near: Int = 0 output result <- sharp.result	A finished picture (tone-mapped, sRGB) scaled to the window’s size: an edge-aware upscale (image.upscale), sharpened once more without overshoot (image.contrast_sharpen), each frame stable on its own, nothing carried between frames. A picture already the window’s size passes through as it is. Placed: node shown: graph “image.entpkg/graphs/upscale.esn” { picture <- view.result; amount = 0.8; }

Kernel 说明

image.adjust@1

The picture adjusted: exposure stops, contrast (1 none), saturation (1 none, 0 gray), hue turned (a whole turn is 1).

执行模式	map
需要能力	field.read, field.write
输入	source: Image
参数	exposure: Float = 0.0, contrast: Float = 1.0, saturation: Float = 1.0, hue: Float = 0.0
输出	result: Image
源码	packages/image.entpkg/kernels/tone.ent

image.apart@1

How far apart two pictures are as a whole (domain = 1): the most a pixel differs (x) and how many differ by more than the rows’ step (y), from image.apart_rows.

执行模式	map
需要能力	field.read, field.write
输入	rows: Grid<Vec2>
参数	—
输出	result: Rows<Vec2>
源码	packages/image.entpkg/kernels/compare.ent

image.apart_rows@1

Each row of a difference (image.difference’s), one cell a row: the most a pixel differs (x) and how many differ by more than over (y). A node sets domain = size(difference) * vec2(0, 1) + vec2(1, 0).

执行模式	map
需要能力	field.read, field.write
输入	difference: Image
参数	over: Float = 0.0078125
输出	result: Grid<Vec2>
源码	packages/image.entpkg/kernels/compare.ent

image.blend@1

top composited over base, as the W3C’s compositing spec does it: where top covers (its alpha times opacity), its color is blended with base’s by mode, as much as base is there; then laid over base, the alphas joined as one sheet over another, the color weighed by both. space is where the arithmetic is done: on sRGB’s stored values (Encoded), as painting programs blend and the modes are defined, or on light itself (Linear). top is stretched over base whatever its size.

执行模式	map
需要能力	field.read, field.write
输入	base: Image, top: Image
参数	mode: Int = 0, opacity: Float = 1.0, space: Int = 0
输出	result: Image
源码	packages/image.entpkg/kernels/blend.ent

image.blur@1

A Gaussian blur along axis (one pass of the two a blur takes, graphs/blur.esn): sigma measured in the output’s pixels, taps one output pixel apart out to three sigmas.

执行模式	map
需要能力	field.read, field.write
输入	source: Image
参数	axis: Vec2 = [1.0, 0.0], sigma: Float = 2.0
输出	result: Image
源码	packages/image.entpkg/kernels/filter.ent

image.contrast_sharpen@1

Sharpened once (amount 0 not at all, 1 the most): pulled from its four neighbors by an amount held between the darkest and brightest of the five, never past 0 to 1, with no white rims.

执行模式	map
需要能力	field.read, field.write
输入	image: Image
参数	amount: Float = 0.8
输出	result: Image
源码	packages/image.entpkg/kernels/upscale.ent

image.convolve@1

Any convolution: weights holds n by n numbers, row by row, centered on the pixel (n odd). The weights are used as given, not normalized.

执行模式	map
需要能力	field.read, field.write
输入	source: Image, weights: Rows<Float>
参数	—
输出	result: Image
源码	packages/image.entpkg/kernels/filter.ent

image.cumulate@1

The cumulative count: at each bin, the pixels in it and below.

执行模式	prefix with sum
需要能力	field.read, field.write
输入	counts: Rows<Int>
参数	—
输出	result: Rows<Int>
源码	packages/image.entpkg/kernels/histogram.ent

image.curve@1

Curves: each channel looked up in curve, rows of the output for inputs spread evenly from 0 to 1 (x for red, y green, z blue), blended between rows.

执行模式	map
需要能力	field.read, field.write
输入	source: Image, curve: Rows<Vec4>
参数	—
输出	result: Image
源码	packages/image.entpkg/kernels/tone.ent

image.decode@1

执行模式	map
需要能力	field.read, field.write
输入	source: Image
参数	—
输出	result: Image
源码	packages/image.entpkg/kernels/color.ent

image.difference@1

The difference of a and b (domain = size(a); b read at the same place across the picture, whatever its size): each channel’s, and in w the most of them.

执行模式	map
需要能力	field.read, field.write
输入	a: Image, b: Image
参数	—
输出	result: Image
源码	packages/image.entpkg/kernels/compare.ent

image.distance@1

The signed distance, in pixels, from each pixel’s center to the shape’s edge: from the nearest pixel of the shape when outside it (inside’s flood), less the distance to the nearest pixel outside it when within (outside’s flood); half a pixel taken off each, so the edge lies between the pixels either side. Where a flood found no seed, reach.

执行模式	map
需要能力	field.read, field.write
输入	inside: Image, outside: Image
参数	reach: Float = 1000.0
输出	result: Grid<Float>

image.dither@1

Each pixel held to colors with ordered dithering (dithered_color): a gradient becomes a pattern of two colors.

执行模式	map
需要能力	field.read, field.write
输入	source: Image, colors: Rows<Vec3>
参数	—
输出	result: Image
源码	packages/image.entpkg/kernels/pixel_art.ent

image.edges@1

Edges: the strength of the Sobel gradient, gray, opaque.

执行模式	map
需要能力	field.read, field.write
输入	source: Image
参数	—
输出	result: Image
源码	packages/image.entpkg/kernels/filter.ent

image.encode@1

Linear light to sRGB’s stored values (for a window or a file), and back; alpha kept.

执行模式	map
需要能力	field.read, field.write
输入	source: Image
参数	—
输出	result: Image
源码	packages/image.entpkg/kernels/color.ent

image.equalize@1

The picture equalized: each pixel’s lightness moved to the share of the pixels darker than it (cumulative, from image.cumulate), read between the counts below and through its bin by where in the bin it falls, so a smooth gradient stays smooth. The color is scaled to reach it: lightness grows as the cube root of a scale, so the cube of the ratio, which keeps the hue. amount mixes it with the source.

执行模式	map
需要能力	field.read, field.write
输入	source: Image, cumulative: Rows<Int>
参数	amount: Float = 1.0
输出	result: Image
源码	packages/image.entpkg/kernels/histogram.ent

image.first_step@1

The first step: half the picture’s longer side, as a power of two.

执行模式	map
需要能力	field.read, field.write
输入	source: Image
参数	—
输出	result: Rows<Int>

image.flood@1

One pass: the nearest seed among this pixel’s and the eight a step away.

执行模式	map
需要能力	field.read, field.write
输入	seeds: Image, step: Rows<Int>
参数	—
输出	result: Image

image.fxaa@1

The HDR picture image antialiased (the same size, fxaa); on 0 leaves it as it is.

执行模式	map
需要能力	field.read, field.write
输入	image: Image
参数	threshold: Float = 0.125, least: Float = 0.04, subpixel: Float = 0.6, on: Int = 1
输出	result: Image
源码	packages/image.entpkg/kernels/antialias.ent

image.histogram@1

How many pixels of the source fall in each bin.

执行模式	scatter into input 1 with sum
需要能力	field.read, field.write
输入	source: Image, bins: Rows<Int>
参数	—
输出	result: Rows<Int>
源码	packages/image.entpkg/kernels/histogram.ent

image.levels@1

Levels: black to 0 and white to 1, clamped, then bent by gamma (above 1 brightens the middle).

执行模式	map
需要能力	field.read, field.write
输入	source: Image
参数	black: Float = 0.0, white: Float = 1.0, gamma: Float = 1.0
输出	result: Image
源码	packages/image.entpkg/kernels/tone.ent

image.lut@1

The picture through a 3D LUT (lut_at), its values taken 0 to 1.

执行模式	map
需要能力	field.read, field.write
输入	source: Image, lut: Image
参数	—
输出	result: Image
源码	packages/image.entpkg/kernels/tone.ent

image.masked@1

top over base as far as mask says: 0 base, 1 top, between them mixed (domain = size(base); top and mask any size, read smoothly across the picture). Any pass kept to part of a picture: a grade, an outline, a blur, laid only where a mask asks.

执行模式	map
需要能力	field.read, field.write
输入	base: Image, top: Image, mask: Grid<Float>
参数	—
输出	result: Image
源码	packages/image.entpkg/kernels/blend.ent

image.mip_first@1

The strip before any halving: the texture in the top square and nothing under it.

执行模式	map
需要能力	field.read, field.write
输入	source: Image
参数	side: Int = 256
输出	result: Image
源码	packages/image.entpkg/kernels/mipmap.ent

image.mip_halve@1

One halving: four texels of the level above averaged into one of the level named, everything else carried over.

执行模式	map
需要能力	field.read, field.write
输入	strip: Image, step: Rows<Int>
参数	side: Int = 256
输出	result: Image
源码	packages/image.entpkg/kernels/mipmap.ent

image.mip_next@1

执行模式	map
需要能力	field.read, field.write
输入	previous: Rows<Int>
参数	—
输出	result: Rows<Int>
源码	packages/image.entpkg/kernels/mipmap.ent

image.morph@1

Dilation (each channel the greatest within radius pixels, a disc) or erosion (the least): bright shapes grown or worn away.

执行模式	map
需要能力	field.read, field.write
输入	source: Image
参数	radius: Int = 1, mode: Int = 0
输出	result: Image
源码	packages/image.entpkg/kernels/filter.ent

image.neutral@1

The neutral LUT of side n (domain = vec2(n * n, n)): each color itself. Grade it as a picture and image.lut applies that grade.

执行模式	map
需要能力	field.write
输入	—
参数	n: Int = 33
输出	result: Image
源码	packages/image.entpkg/kernels/tone.ent

image.next_step@1

The next step: half this one, never below one.

执行模式	map
需要能力	field.read, field.write
输入	step: Rows<Int>
参数	—
输出	result: Rows<Int>

image.outline@1

A line of color round what is drawn: each pixel not drawn beside one that is, across a side, or across a corner too (corners).

执行模式	map
需要能力	field.read, field.write
输入	source: Image
参数	color: Vec4 = [0.0, 0.0, 0.0, 1.0], corners: Bool = False
输出	result: Image
源码	packages/image.entpkg/kernels/pixel_art.ent

image.palette@2

Each pixel the color of colors nearest it.

执行模式	map
需要能力	field.read, field.write
输入	source: Image, colors: Rows<Vec3>
参数	—
输出	result: Image
源码	packages/image.entpkg/kernels/pixel_art.ent

image.pixelate@1

Cells of cell pixels, each the color at its center.

执行模式	map
需要能力	field.read, field.write
输入	source: Image
参数	cell: Float = 8.0
输出	result: Image
源码	packages/image.entpkg/kernels/pixel_art.ent

image.ramp@1

Each pixel the color of colors, rows from dark to light, its lightness falls to (ramp_color).

执行模式	map
需要能力	field.read, field.write
输入	source: Image, colors: Rows<Vec3>
参数	—
输出	result: Image
源码	packages/image.entpkg/kernels/pixel_art.ent

image.resize@1

The source at the node’s domain, by bicubic (Catmull-Rom) interpolation: sharper than at_linear when enlarging. For halving and beyond, image.shrink.

执行模式	map
需要能力	field.read, field.write
输入	source: Image
参数	—
输出	result: Image
源码	packages/image.entpkg/kernels/filter.ent

image.seeds@1

Seeds: where the mask is on (its luminance times alpha past threshold, or not past it when inside is false), the pixel’s own center, else none. A seed is (x, y, 1, 0); none is all zeros.

执行模式	map
需要能力	field.read, field.write
输入	mask: Image
参数	threshold: Float = 0.5, inside: Bool = True
输出	result: Image

image.shadow@1

A shadow of color cast by what is drawn, offset pixels away, under it: each pixel not drawn whose pixel offset back is.

执行模式	map
需要能力	field.read, field.write
输入	source: Image
参数	offset: Vec2 = [1.0, 1.0], color: Vec4 = [0.0, 0.0, 0.0, 0.5]
输出	result: Image
源码	packages/image.entpkg/kernels/pixel_art.ent

image.sharpen@1

Unsharp masking: the source pushed away from its blurred copy by amount; alpha kept. graphs/sharpen.esn places it with its blur.

执行模式	map
需要能力	field.read, field.write
输入	source: Image, blurred: Image
参数	amount: Float = 0.6
输出	result: Image
源码	packages/image.entpkg/kernels/filter.ent

image.shrink@1

The source smaller: each output pixel the mean of the source pixels it covers (a box filter, exact at whole ratios).

执行模式	map
需要能力	field.read, field.write
输入	source: Image
参数	—
输出	result: Image
源码	packages/image.entpkg/kernels/filter.ent

image.upscale@1

Scaled to domain: edge-aware (upscaled), nearest-neighbor when near is 1 (upscaled_near).

执行模式	map
需要能力	field.read, field.write
输入	image: Image
参数	near: Int = 0
输出	result: Image
源码	packages/image.entpkg/kernels/upscale.ent

测试

antialias.replay, compare.replay, looks.replay, operations.replay, upscale.replay