Skip to main content

Video & Audio filters

Video Filters

Some calling apps allow filters to be applied to the current user's video, such as blurring the background, adding AR elements (glasses, moustaches, etc) or applying image filters (such as sepia, bloom etc). StreamVideo's iOS SDK has support for injecting your custom filters into the calling experience.

How does this work? If you initialized the SDK with custom filters support and the user selected a filter, you will receive each frame of the user's local video as CIImage, allowing you to apply the filters. This way you have complete freedom over the processing pipeline.

You can find a working example of the filters (together with other great example projects) in our VideoWithChat sample project. Here is how the project you are about to build will look like in the end:

Adding a Video Filter

The VideoFilter class allows you to create your own filters. It contains the id and name of the filter, along with an async function that converts the original CIImage to an output CIImage. If no filter is selected, the same input image is returned.

For example, let's add a simple "Sepia" filter, from the default CIFilter options by Apple:

static let sepia: VideoFilter = {
let sepia = VideoFilter(id: "sepia", name: "Sepia") { image in
let sepiaFilter = CIFilter(name: "CISepiaTone")
sepiaFilter?.setValue(image, forKey: kCIInputImageKey)
return sepiaFilter?.outputImage ?? image
return sepia

You can now create a helper FilterService, that will keep track of the available filters, as well as hold state information about the selected filter and whether the filters picker is shown:

class FiltersService: ObservableObject {
@Published var filtersShown = false
@Published var selectedFilter: VideoFilter?

static let supportedFilters = [sepia]

Next, you need to pass the supported filters to the StreamVideo object, via its VideoConfig and connect the user:

let streamVideo = StreamVideo(
apiKey: apiKey,
user: userCredentials.user,
token: token,
videoConfig: VideoConfig(
videoFilters: FiltersService.supportedFilters
tokenProvider: { result in
// Unrelated code skipped. Check repository for complete code:

private func connectUser() {
Task {
try await streamVideo?.connect()

Now, let's enable the filter selection in the user interface. One option is to include the filters in the call controls shown at the bottom of the call view. For this, the first step is to override the makeCallControlsView function in your custom implementation of the ViewFactory:

class VideoViewFactory: ViewFactory {

/* ... Previous code skipped. */

func makeCallControlsView(viewModel: CallViewModel) -> some View {
ChatCallControls(viewModel: viewModel)

You will now create the ChatCallControls view that does two things. It will first place an icon to toggle the filters menu (via the filtersService.filtersShown property) and allows users to select the filter they want to apply.

Second, it will conditionally show a list of the filters with a button for each one to (de-)select it.

In this section, only the code to show the filters is added. You can see the full code here, but let's have a look at the simplified version:

public var body: some View {
VStack {
HStack {
/* Skip unrelated code */
// 1. Button to toggle filters view
Button {
withAnimation {
}, label: {
icon: Image(systemName: "camera.filters"),
size: size,
iconStyle: filtersService.filtersShown ? .primary : .transparent
/* Skip unrelated code */

if filtersService.filtersShown {
HStack(spacing: 16) {
// 2. Show a button for each filter
ForEach(FiltersService.supportedFilters) { filter in
Button {
withAnimation {
// 3. Select or de-select filter on tap
if filtersService.selectedFilter == filter {
filtersService.selectedFilter = nil
} else {
filtersService.selectedFilter = filter
} label: {
.background(filtersService.selectedFilter == filter ? : Color.gray)
/* more modifiers */

/* more modifiers */

Here are the three things this code does:

  1. Adding the icon for the filters, that will control the filtersShown state.
  2. Whenever the filtersShown is true, you're showing the list of the available filters.
  3. When a user taps on a filter, the CallViewModel's setVideoFilter method is called. This will enable or disable the video filter for the ongoing call.

That is everything that is needed for a basic video filter support.

Adding AI Filters

In some cases, you might also want to apply AI filters. That can be an addition to the user's face (glasses, moustaches, etc), or an ML filter. In this section this use-case will be covered. Specifically, you will show Stream's logo over the user's face. Whenever the user moves along, you will update the logo's location.


The code will be slightly simplified for the sake of this guide. If you want to see the entire example with the full code, you can see the sample on GitHub.

To do this, you will use the Vision framework and the VNDetectFaceRectanglesRequest. First, let's create the method that will detect the faces:

static func detectFaces(image: CIImage) async throws -> CGRect {
return try await withCheckedThrowingContinuation { continuation in
let detectFaceRequest = VNDetectFaceRectanglesRequest { (request, error) in
if let result = request.results?.first as? VNFaceObservation {
continuation.resume(returning: result.boundingBox)
} else {
continuation.resume(throwing: ClientError.Unknown())
let vnImage = VNImageRequestHandler(ciImage: image, orientation: .downMirrored)
try? vnImage.perform([detectFaceRequest])

The VNDetectFaceRectanglesRequest does not support the async/await syntax yet, so it is converted using the withCheckedThrowingContinuation mechanism (see Apple documentation.

Next, let's add some helper methods, that will allow conversion between CIImage and UIImage, as well as the possibility to draw over an image:

static func convert(cmage: CIImage) -> UIImage {
let context = CIContext(options: nil)
let cgImage = context.createCGImage(cmage, from: cmage.extent)!
let image = UIImage(cgImage: cgImage, scale: scale, orientation: .up)
return image

static func drawImageIn(_ image: UIImage, size: CGSize, _ logo: UIImage, inRect: CGRect) -> UIImage {
let format = UIGraphicsImageRendererFormat()
format.scale = 1
format.opaque = true
let renderer = UIGraphicsImageRenderer(size: size, format: format)
return renderer.image { context in
image.draw(in: CGRect(origin:, size: size))
logo.draw(in: inRect)

With those two helpers in place, you can now implement your custom AI filter. The same principle applies when creating a VideoFilter as in the first part of this guide.

static let stream: VideoFilter = {
let stream = VideoFilter(id: "stream", name: "Stream") { image in
// 1. detect, where the face is located (if there's any)
guard let faceRect = try? await detectFaces(image: image) else { return image }
let converted = convert(cmage: image)
let bounds = image.extent
let convertedRect = CGRect(
x: faceRect.minX * bounds.width - 80,
y: faceRect.minY * bounds.height,
width: faceRect.width * bounds.width,
height: faceRect.height * bounds.height
// 2. Overlay the rectangle onto the original image
let overlayed = await drawImageIn(converted, size: bounds.size, streamLogo, inRect: convertedRect)

// 3. convert the created image to a CIImage
let result = CIImage(cgImage: overlayed.cgImage!)
return result
return stream

Here's what this code does:

  1. It's detecting the face rectangle using the detectFaces method you defined earlier and it converts the CIImage to UIImage. The rectangle to the real screen dimensions (since it returns percentages).
  2. This information is passed to the drawImageIn method, that adds the logo at the convertedRect frame.
  3. The result is then converted back to a CIImage and returned.

The last remaining thing to do is to add the stream filter in the supported filters method. This is done in the FiltersService class:

static let supportedFilters = [sepia, stream]

The end result will look like this (supposedly with a different face):

Stream Filter

This guide shows you a concrete example how to add a specific AI filter. However, with the way this is structured we aim to give you full control over what you can build with this.

We would love to see the cool things you'll build with this functionality, feel free to tweet about them and tag us.

Audio Filters

The StreamVideo SDK also supports custom audio processing of the local track. This opens up possibilities for custom echo filtering, voice changing or other audio effects.

If you want to have custom audio processing, you need to provide your own implementation of the AudioProcessingModule protocol in our VideoConfig.

In the following example, we will build a simple audio filter that gives the user's voice a robotic touch.

let voiceProcessor = CustomVoiceProcessor()
let filter = RobotVoiceFilter(pitchShift: 0.8)
let audioProcessingModule = RTCDefaultAudioProcessingModule(
config: nil,
capturePostProcessingDelegate: voiceProcessor,
renderPreProcessingDelegate: nil

In this code, we are using several newly defined objects, so let's go through them. First, we create a CustomVoiceProcessor.

class CustomVoiceProcessor: NSObject, RTCAudioCustomProcessingDelegate {

private var audioFilter: AudioFilter?

func audioProcessingInitialize(sampleRate sampleRateHz: Int, channels: Int) {}

func audioProcessingProcess(audioBuffer: RTCAudioBuffer) {
var audioBuffer = audioBuffer
audioFilter?.applyEffect(to: &audioBuffer)

func audioProcessingRelease() {}

func setAudioFilter(_ audioFilter: AudioFilter?) {
self.audioFilter = audioFilter

protocol AudioFilter {

func applyEffect(to audioBuffer: inout RTCAudioBuffer)


extension RTCDefaultAudioProcessingModule: AudioProcessingModule, @unchecked Sendable {}

The CustomVoiceProcessor conforms to the RTCAudioCustomProcessingDelegate, which has 3 mandatory methods to implement. For our case, we only need to implement the audioProcessingProcess(audioBuffer: RTCAudioBuffer) method.

In this method, we are calling a method called applyEffect, which is available via our newly defined AudioFilter protocol. This protocol has one method, called applyEffect(to audioBuffer: inout RTCAudioBuffer).

By default, there is no filter (the audioFilter property is nil), but we expose a method called setAudioFilter, so we can set one from the UI.

Next, let's see our sample implementation of the AudioFilter protocol, namely the RobotVoiceFilter.

class RobotVoiceFilter: AudioFilter {

let pitchShift: Float

init(pitchShift: Float) {
self.pitchShift = pitchShift

func applyEffect(to audioBuffer: inout RTCAudioBuffer) {
let frameSize = 256
let hopSize = 128
let scaleFactor = Float(frameSize) / Float(hopSize)

let numFrames = (audioBuffer.frames - frameSize) / hopSize

for channel in 0..<audioBuffer.channels {
let channelBuffer = audioBuffer.rawBuffer(forChannel: channel)

for i in 0..<numFrames {
let inputOffset = i * hopSize
let outputOffset = Int(Float(i) * scaleFactor) * hopSize

var outputFrame = [Float](repeating: 0.0, count: frameSize)

// Apply pitch shift
for j in 0..<frameSize {
let shiftedIndex = Int(Float(j) * pitchShift)
let originalIndex = inputOffset + j
if shiftedIndex >= 0 && shiftedIndex < frameSize && originalIndex >= 0 && originalIndex < audioBuffer.frames {
outputFrame[shiftedIndex] = channelBuffer[originalIndex]

// Copy back to the input buffer
for j in 0..<frameSize {
let outputIndex = outputOffset + j
if outputIndex >= 0 && outputIndex < audioBuffer.frames {
channelBuffer[outputIndex] = outputFrame[j]

The filter is initalized with pitch shift, which configures how much the channel buffer's indexes should be shifted for the effect. We use a default value of 0.8, but you can configure it depending on how "robotic" the voice should be.

This is a simple algorithm that just does shifting of the indexes. For a more complex one, you can also use some voice processing library. The important part is that you update the channelBuffer with the filtered values.

Did you find this page helpful?