These docs describe the main branch, including unreleased changes. Read the 0.6.0 docs.

FrameBuffer color matrices

This supported advanced API transforms stored foreground and background colors in an OptimizedBuffer. Use it when a FrameBuffer needs a cell-level color transform.

Buffer positions are terminal cells, not image pixels. Read Colors for normal color input and terminal palette behavior.

Methods#

import { TargetChannel } from "@opentui/core"

frameBuffer.colorMatrix(
  matrix: Float32Array,
  cellMask: Float32Array,
  strength = 1,
  target = TargetChannel.Both,
)

frameBuffer.colorMatrixUniform(
  matrix: Float32Array,
  strength = 1,
  target = TargetChannel.Both,
)

Both methods mutate the buffer and return void. The matrix must contain exactly 16 floats. A different length throws RangeError before the native call. A non-finite strength also throws RangeError.

Native code rejects a non-finite matrix coefficient, a target that is not a TargetChannel value, and a mask with more triplets than the Context’s renderCellsMax. A direct call then throws NativeError. In a paint hook, the frame fails when native code plays the recording.

colorMatrix() transforms only mask entries. colorMatrixUniform() transforms every buffer cell.

Matrix format#

The matrix is row-major. Each row computes one output channel:

[m00, m01, m02, m03]  red output
[m10, m11, m12, m13]  green output
[m20, m21, m22, m23]  blue output
[m30, m31, m32, m33]  alpha output

For normalized input C = [r, g, b, a], OpenTUI calculates:

transformed = M * C
result = C + (transformed - C) * effectiveStrength

strength = 0 keeps the original value. strength = 1 uses the complete matrix result. Strength is not clamped, so a negative value or a value above one extrapolates.

Intermediate calculations can be below 0 or above 1. The stored buffer is RGBA8. Native code clamps each final channel to 0..1, rounds it to 0..255, and writes it to storage. The native source comments that say no clamping describe the float calculation, not the RGBA8 write.

Native code stores a calculated channel that is not finite, for example after overflow, as zero.

Target channels#

Enum Value Cells changed
TargetChannel.FG 1 Foreground only
TargetChannel.BG 2 Background only
TargetChannel.Both 3 Foreground and background

Use the enum values. Core and native code reject other numbers.

Cell mask#

colorMatrix() reads packed triplets:

[x, y, perCellStrength, x, y, perCellStrength, ...]

The effective strength is methodStrength * perCellStrength.

Mask behavior is exact:

  • Native code reads Math.floor(cellMask.length / 3) triplets. It ignores one or two trailing floats.
  • Native code skips a coordinate that is negative, not finite, or above the native u32 range.
  • Positive fractional coordinates truncate toward zero.
  • Native code skips coordinates outside the buffer.
  • Native code skips a triplet whose effective strength is zero or not finite.
  • Duplicate coordinates apply the matrix more than once, in mask order.

The mask ignores the active scissor and opacity stacks. Limit coordinates in the mask when the transform must stay in a clipped region.

Uniform behavior#

colorMatrixUniform() uses a four-cell SIMD path and a scalar remainder. With strength === 0, native code returns without a change.

Uniform transforms ignore the active scissor and opacity stacks. They process width * height stored cells, including blank and continuation cells.

In a paint hook, native code plays the transform on the frame at that node’s place in paint order. A uniform transform then changes the whole frame as painted up to that point, not only the node’s area.

Cell color semantics#

Each cell color stores RGBA8 values plus terminal color-intent metadata. The matrix reads the resolved RGBA channels. It then writes a new explicit RGB color. A transformed indexed or default color loses its palette or default intent.

The methods do not change characters, continuation tags, image placements, text attributes, or hyperlink IDs. They can change foreground and background alpha if the fourth matrix row does so.

The transform has no retained allocation or cleanup method. Native code reads the supplied arrays only during the call. In a paint hook, the recording keeps a copy of them.

Exported matrices#

@opentui/core exports these Float32Array values. Each keeps alpha unchanged.

Export Transform
SEPIA_MATRIX Sepia
PROTANOPIA_SIM_MATRIX Protanopia simulation
DEUTERANOPIA_SIM_MATRIX Deuteranopia simulation
TRITANOPIA_SIM_MATRIX Tritanopia simulation
ACHROMATOPSIA_MATRIX Luminance grayscale simulation
PROTANOPIA_COMP_MATRIX Protanopia-oriented channel compensation
DEUTERANOPIA_COMP_MATRIX Deuteranopia-oriented channel compensation
TRITANOPIA_COMP_MATRIX Tritanopia-oriented channel compensation
TECHNICOLOR_MATRIX Increased and cross-reduced RGB channels
SOLARIZATION_MATRIX Partial channel inversion
SYNTHWAVE_MATRIX Magenta-biased channel mapping
GREENSCALE_MATRIX Luminance in the green channel only
GRAYSCALE_MATRIX Luminance copied to RGB
INVERT_MATRIX 1 - channel through the alpha column

These constants encode calculations. Their names do not promise accessibility outcomes for every display or viewer.

Apply a uniform matrix#

import { INVERT_MATRIX, TargetChannel } from "@opentui/core"

frameBuffer.colorMatrixUniform(INVERT_MATRIX, 1, TargetChannel.Both)

Apply a cell mask#

import { SEPIA_MATRIX, TargetChannel } from "@opentui/core"

const cells = new Float32Array([5, 2, 1, 6, 2, 0.5, 7, 2, 0.25])

frameBuffer.colorMatrix(SEPIA_MATRIX, cells, 0.8, TargetChannel.FG)

The effective strengths in this example are 0.8, 0.4, and 0.2.

Next#

Changes0.5.6
0.5.6
core: RenderLib.bufferColorMatrix, bufferColorMatrixUniform, bufferDrawGrayscaleBuffer, bufferDrawGrayscaleBufferSupersampled, bufferDrawPackedBuffer, and bufferDrawSuperSampleBuffer accept a typed array as well as a pointer. (#1394)