GreenScreenStream provides a wide range of options for manipulating Mediastreams.
Generate a new MediaStream for <canvas> & <video> elements based on provided MediaStreamTrack and a background image/video just using JavaScript.
After processed and "green screened" you can capture the stream and use it in your WebRTC calls for instance.
All rendering is done in realtime using a WebGL2 pixel shader (glsl) and optionally machine-learning.
npm i @virtualscenery/greenscreenstream
Below you find a few different examples of greenscreenstream.
ML based Virtual Background - Image Background
ML based Virtual Background - Video Background
ML based Virtual Background - WebGL Background
ML based Virtual Background - Hologram Effect
https://virtualscenery.github.io/greenscreenstream-examples/example/WebGL
Look here for implementation.
Creates an instance of GreenScreenStream
constructor(greenScreenMethod: GreenScreenMethod, resolution: VideoResolution | Vector2, canvas?: HTMLCanvasElement)Initlializes the GreenScreenStream with the provided background (image or video) and settings.
initialize(backgroundUrl?: string, config?: GreenScreenConfig): Promise<boolean> Adds a MediaStreamTrack (i.e webcam)
addVideoTrack(track: MediaStreamTrack): Promise<void | any>;Starts rendering the greenscreen. You can optionally set a fps maximum here
start(maxFps?: number): voidStops the rendering process.
Optionally stop the media streams.
Stopping the streams works only if there are no references to them
outside of greenscreenstream.
stop(stopMediaStreams?:boolean): voidCapture the rendered result to a MediaStream that you apply to your <video> element.
captureStream(fps?: number): MediaStream; Sets the virtual background to a new image or video. Can be done while GreenScreenStream is running.
setBackground(src: string): Promise<HTMLImageElement | HTMLVideoElement | Error>Swaps out the currently used BodyPixModel used in ml mode (GreenScreenMethod.VirtualBackground) (See GreenScreenMethod (Enum))
setBodyPixModel(config: IGreenScreenConfig): Promise<void>Scales the passed in image to canvas size and returns a scaled copy of it. Gets called automatically everytime a new image background is set. The imageOptions defaults to the current size of the greenscreen canvas and high quality .
public async scaleImageToCanvas(image: HTMLImageElement, imageOptions?: ImageBitmapOptions): Promise<HTMLImageElement>Gets the most dominant color and a list (palette) of the colors most common in the provided MediaStreamTrack.
getColorsFromStream(): { palette: [number, number,number][], dominant: [number, number,number] } {Pass a mask (rgb), color to the shader , to use as a mask. Should be the dominant color, or on of the palette colors detected. See getColorsFromStream
setChromaKey(r: number, g: number, b: number, threshold?: number): void;Range is used to decide the amount of color to be used from either foreground or background.Playing with this variable will decide how much the foreground and background blend together.
setMaskRange(x:number,y:number): voidGet the most dominant color based on imageData and number of pixels
dominant(imageData: ImageData, pixelCount: number): [number, number,number] {Get an Array of the most significant colors in the MediaTrack
pallette(imageData: ImageData, pixelCount: number): [number, number,number][] | null {Describes the method GreenScreenStream should use for applying a virtual background.
GreenScreenMethod.VirtualBackground uses a machine learning model (Tensorflow BodyPix)
GreenScreenMethod.VirtualBackgroundUsingGreenScreen works without a machine learning model and thus consumes much less performance,
but requires the user to have a green screen.
enum GreenScreenMethod {
VirtualBackground,
VirtualBackgroundUsingGreenScreen
}Describes resolution presets GreenScreenStream should use.
enum VideoResolution {
SD,
HD,
FullHD,
WQHD,
UHD
}Describes a custom resolution that GreenScreenStream can use.
Both values default to zero.
constructor(x?: number, y?: number)Returns the current values as a string (" x : y")
toString(): string Returns the current value in a format that can be used as MediaTrackConstraints
toMediaConstraint(): MediaTrackConstraintsChecks if the provided input is a valid vector2. Returns true if so.
static isValidVector2(input: any): booleanProvides detailed configuration options for GreenScreenStream.
maskSettings can be uses to fine tune the virtual background appearance. (
bodyPixMode can be used to apply premade BodyPix configurations (see GreenScreenStreamBodyPixMode for more details),
while bodyPixConfig allows you to configure BodyPix as you see fit. If both are provided, bodyPixMode will be ignored.
IGreenScreenConfig {
maskSettings?: IMaskSettings,
bodyPixMode?: GreenScreenStreamBodyPixMode,
bodyPixConfig?: IBodyPixConfig
}Determines which BodyPix Preset GreenStream should use.
Presets Standard or Precise are recommended for most use cases.
Fast is meant for really weak clients, is unprecise and causes flickering.
Maximum uses a more complex ML Model and thus causes much more network traffic & gpu + cpu load.\
enum BodyPixMode {
Fast = 0,
Standard = 1,
Precise = 2,
Maximum = 3
}architecture: 'MobileNetV1',
outputStride: 16,
multiplier: 0.5,
quantBytes: 1architecture: 'MobileNetV1',
outputStride: 16,
multiplier: 0.75,
quantBytes: 2architecture: 'MobileNetV1',
outputStride: 16,
multiplier: 1,
quantBytes: 2architecture: 'ResNet50',
outputStride: 32,
quantBytes: 2Description TBA
interface IMaskSettings {
opacity?: number
flipHorizontal?: boolean
maskBlurAmount?: number
foregroundColor?: RGBA
backgroundColor?: RGBA
segmentPerson?: {
flipHorizontal?: boolean
internalResolution?: string
segmentationThreshold?: number
maxDetections?: number
quantBytes?: number
}
};export interface RGBA {
r: number, g: number, b: number, a: number
}