Interface ReadPixelsOpts<K>

Source render target. Typed as the base RenderTarget so a cube target from either backend (WebGLCubeRenderTarget, or three/webgpu's CubeRenderTarget, which is NOT a WebGLRenderTarget) is accepted; see faceIndex.

Pixel-format discriminator. Determines the returned typed-array kind.

Cube face to read when target is a cube render target (0..5 in three's px, nx, py, ny, pz, nz order). Threaded into WebGL's activeCubeFaceIndex argument and WebGPU's faceIndex argument — the two signatures put it in different slots (after the destination buffer vs after textureIndex). Ignored for a 2D target.

X offset in pixels (top-down convention). Defaults to 0. Full-target reads (the common case) leave this at 0; sub-region readers (e.g. the picking 5×5 voter) supply canvas-space top-down coords and the primitive converts to the backend's framebuffer convention internally.

Y offset in pixels (canonical top-down, row 0 = top of source). The primitive converts this to bottom-up coordinates for WebGL readback, including WebGPURenderer's WebGL2 fallback. Defaults to 0.

Region width in pixels. Defaults to target.width.

Region height in pixels. Defaults to target.height.

Default false. The primitive's canonical return convention is rows in top-down order (row 0 = top of the source target). Pass true to receive bottom-up rows instead — used by callers whose downstream consumers expect scene-space row order (e.g. EXR export).

Optional pre-allocated destination buffer for the raw readback. Must be exactly width * height * 4 elements of the right typed-array kind for kind. When provided, no fresh allocation is made for the raw readback; ideal for hot paths (e.g. picking) that fire at mouse-event rates.

Optional pre-allocated destination buffer for the row-flipped output (consulted when a row flip is needed). Must be the same size and kind as the readback, and a different buffer than out if one is supplied (the row-flip interleaves reads/writes across rows and cannot operate in place). When provided, the row-flip writes into this buffer instead of allocating one. The returned pixels is this buffer (not the raw out).