mcut SDK / compositor/src / WebGPUBackendOptions

Interface: WebGPUBackendOptions

Defined in: packages/compositor/src/webgpu/webgpu-backend.ts:41

The WebGPU compositor backend. Image quads (video/image elements) composite as full-frame passes over a ping-pong texture pair — importExternalTexture keeps VideoFrames zero-copy — with effects as WGSL passes and every blend mode in one shader (mode uniform). Raster content (text, captions, multicam chrome, transitions, custom renderers) still paints through canvas2d in frame space and uploads as a texture layer in z-order, so output matches the reference Canvas2D backend.

Lifecycle: WebGPUBackend.create({ canvas, width, height }) once, then renderFrameWith(backend, project, timeMs, options) per frame, resize on project dimension changes, dispose when done. For export, pass the GPUTexture-backed canvas straight to Mediabunny's CanvasSource — no CPU readback.

Properties

canvas

canvas: HTMLCanvasElement | OffscreenCanvas;

Defined in: packages/compositor/src/webgpu/webgpu-backend.ts:43

Presentation canvas; the backend configures its 'webgpu' context.

device?

optional device?: GPUDevice;

Defined in: packages/compositor/src/webgpu/webgpu-backend.ts:48

Bring your own device (tests/sharing); otherwise adapter-requested.

height

height: number;

Defined in: packages/compositor/src/webgpu/webgpu-backend.ts:46

width

width: number;

Defined in: packages/compositor/src/webgpu/webgpu-backend.ts:45

Project (composition) size — passes render at this resolution.