UI components

The SDK ships camera surfaces with permission gating, preview rendering, framing chrome, and retry messaging already built. You can use them as-is, restyle them with your own fonts, or ignore them entirely and drive the flows from your own UI.

The FaceCaptureDriver contract

Everything face-related routes through a driver, vended by whichever flow needs a capture:

let driver = loginFlow.faceCaptureDriver()        // or registrationFlow

val driver = loginFlow.faceCaptureDriver()        // or registrationFlow

The driver normalizes login's and registration's very different state enums onto one FaceCaptureUiState, which is why a single capture screen serves both. Its surface is small:

FaceCaptureUiState

attemptsMax == 0 means no face-mismatch budget is enforced — that's the signup branch, which mints a fresh token and so can only fail liveness.

iOS surfaces

SenseCryptFaceScanView

The flow-agnostic capture view. Drives a driver, submits passing frames, reports a CaptureOutcome.

SenseCryptFaceScanView(
    driver: driver,
    paused: false,
    titleFontName: "WixMadeforDisplay-Bold",
    bodyFontName: "Inter",
) { outcome in
    switch outcome {
    case .succeeded:            // done
    case .cancelled:            // user backed out
    case .failed(let reason):   // see Errors
    }
}

SenseCryptFaceLoginView

A thin login-specific wrapper over SenseCryptFaceScanView. It takes the LoginFlow directly, vends the driver itself, and delivers a richer result carrying the relying-party label.

SenseCryptFaceLoginView(loginFlow: flow) { result in
    switch result {
    case .succeeded(let rpLabel):     // signed in to rpLabel
    case .cancelled:                  break
    case .failed(let reason):         break
    case .givenUp(let lastReason):    // user abandoned after retries
    }
}

SenseCryptFaceLoginResult adds givenUp(lastReason: FaceCaptureRetryReason) on top of the outcomes CaptureOutcome carries — use it to distinguish "walked away after failing" from an explicit cancel.

SenseCryptScanQrView

The QR scanner: camera preview, branded status pill, animated brackets and scan line, and an in-screen permission-denied gate.

SenseCryptScanQrView(
    mode: .oneShot { result in
        switch result {
        case .decoded(let bytes, let format):
            flow.submitQrPayload(payload: bytes)
        case .cancelled:        break
        case .permissionDenied: break
        case .error(let message): break
        }
    },
    bodyFontName: "Inter"
)

Two modes:

Use .streaming when a decode might be rejected and you want scanning to continue — pair it with resetToScan() on the login flow.

SenseCryptFaceCaptureView

The lower-level capture view, for one-shot capture that isn't attached to a flow. It emits SenseCryptFaceCaptureResult.captured(faceCropJpeg:headPose:metrics:) rather than driving a driver, and only supports .streaming mode.

Prefer SenseCryptFaceScanView for authentication. Reach for this one only when you need raw capture output.

Notable parameters: onSessionReady (fires once the capture session exists, just after the permission gate — stash the handle if you need to feed it to a flow yourself), showSweepRing / showMeshDots (processing chrome the host toggles, since this view doesn't observe flow state), hideTopChrome, and capturingPillHints.

UIKit hosts

Both scan surfaces have UIHostingController bridges, so a UIKit app needn't drop into SwiftUI:

let vc = SenseCryptScanQrViewController(mode: .oneShot { result in /* … */ })
present(vc, animated: true)

let faceVC = SenseCryptFaceCaptureViewController(mode: .streaming(stream: continuation))

Android surfaces

Both are @Composable functions in ai.seventhsense.sensecrypt.authenticator.ui.

SenseCryptScanQrView

SenseCryptScanQrView(
    modifier = Modifier.fillMaxSize(),
    paused = alreadySubmitted,
    bodyFontFamily = BodyFont,
    onResult = { result ->
        when (result) {
            is SenseCryptScanQrResult.Decoded ->
                flow.submitQrPayload(result.loginQrPayloadBytes)
            SenseCryptScanQrResult.Cancelled,
            SenseCryptScanQrResult.PermissionDenied -> cancel()
            is SenseCryptScanQrResult.Error -> { /* let the flow surface it */ }
        }
    },
)

SenseCryptFaceCaptureView

SenseCryptFaceCaptureView(
    modifier = Modifier.fillMaxSize(),
    paused = sheetShowing,
    showSweepRing = processing,
    showMeshDots = processing,
    lastRetryReason = pillRetryReason,
    showPostingScrim = posting,
    titleFontFamily = HeadFont,
    bodyFontFamily = BodyFont,
    onSessionReady = { handle -> captureSessionHandle = handle },
    onFaceLost = { livenessRetryActive = false },
    onResult = { result -> /* Cancelled / PermissionDenied / Error */ },
)

Where the two platforms differ

The bundled surfaces are close but not identical. Three differences will bite if you port code between them:

The driver's stream accessor is named differently. iOS uses captureStateStream() on the driver — stateStream() exists only on the per-flow types. Android uses the state property throughout.

Decoded carries different payloads. iOS emits decoded(bytes: Data, format: BarcodeFormat), so you can tell a QR from a PDF417. Android emits Decoded(loginQrPayloadBytes: ByteArray) with no format field. Code that branches on barcode format won't port.

Android's face-capture result has no success case. iOS SenseCryptFaceCaptureResult includes captured(faceCropJpeg:headPose:metrics:); Android's has only Cancelled, PermissionDenied, and Error. On Android the captured frame reaches you through the onSessionReady / onPassingFrame callbacks as a FaceCaptureSessionHandle instead — there is no equivalent of iOS's raw crop-and-metrics result.

Building your own UI

Nothing requires the bundled surfaces. To replace them, you need to:

1. Request camera permission yourself.
2. Run a camera session and produce a FaceCaptureSessionHandle.
3. Observe the driver's FaceCaptureUiState and stop submitting frames outside awaitingCapture.
4. Call submitFaceCapture, retry, and cancel on the driver.
5. Render the retry reasons and the attemptsUsed/attemptsMax budget.

The state contract is the same either way — the bundled views are consumers of the same public driver API you'd be using.

Related

• Login — where the driver comes from.
• Registration — the signup capture branch.
• Reference — CaptureOutcome, FaceCaptureRetryReason, HeadPose, FaceMetrics.
• Install — camera permission and CameraX dependencies.