compute times for cross-fade between videos

I would use three instructions to apply the cross-fade:

1. A “pass-through” instruction that shows only the first video track, until one second before the end of the first asset.
2. A cross-fade instruction that simultaneously shows the last second of the first video track and the first second of the second video track, with opacity ramps.
3. A “pass-through” instruction that shows only the second video track, starting from one second into the second video track.

So, first, let's get the tracks:

import AVFoundation
import CoreVideo

func crossFade(asset0: AVAsset, asset1: AVAsset, crossFadeDuration: CMTime, to outputUrl: URL) throws {
    guard
        let asset0Track = asset0.tracks(withMediaType: .video).first,
        let asset1Track = asset1.tracks(withMediaType: .video).first,
        case let composition = AVMutableComposition(),
        case let compositionTrack0Id = composition.unusedTrackID(),
        let compositionTrack0 = composition.addMutableTrack(
            withMediaType: .video, preferredTrackID: compositionTrack0Id),
        case let compositionTrack1Id = composition.unusedTrackID(),
        let compositionTrack1 = composition.addMutableTrack(
            withMediaType: .video, preferredTrackID: compositionTrack1Id)
        else { return }

Now let's compute all of the times we need. First, the entire range of asset0Track in the composition, include both the pass-through and cross-fade periods:

    // When does asset0Track start, in the composition?
    let asset0TrackStartTime = CMTime.zero

    // When does asset0Track end, in the composition?
    let asset0TrackEndTime = asset0TrackStartTime + asset0Track.timeRange.duration

Next, the cross-fade's time range:

    // When does the cross-fade end, in the composition?
    // It should end exactly at the end of asset0's video track.
    let crossFadeEndTime = asset0TrackEndTime

    // When does the cross-fade start, in the composition?
    let crossFadeStartTime = crossFadeEndTime - crossFadeDuration

    // What is the entire time range of the cross-fade, in the composition?
    let crossFadeTimeRange = CMTimeRangeMake(
        start: crossFadeStartTime,
        duration: crossFadeDuration)

Next, the entire range of asset1Track in the composition, include both the cross-fade and pass-through periods:

    // When does asset1Track start, in the composition?
    // It should start exactly at the start of the cross-fade.
    let asset1TrackStartTime = crossFadeStartTime

    // When does asset1Track end, in the composition?
    let asset1TrackEndTime = asset1TrackStartTime + asset1Track.timeRange.duration

And finally, the two pass-through time ranges:

    // What is the time range during which only asset0 is visible, in the composition?
    let compositionTrack0PassThroughTimeRange = CMTimeRangeMake(
        start: asset0TrackStartTime,
        duration: crossFadeStartTime - asset0TrackStartTime)

    // What is the time range during which only asset1 is visible, in the composition?
    let compositionTrack1PassThroughTimeRange = CMTimeRangeMake(
        start: crossFadeEndTime,
        duration: asset1TrackEndTime - crossFadeEndTime)

Now we can insert the input tracks into the composition's tracks:

    // Put asset0Track into compositionTrack0.
    try compositionTrack0.insertTimeRange(
        asset0Track.timeRange,of: asset0Track, at: asset0TrackStartTime)

    // Put asset1Track into compositionTrack1.
    try compositionTrack1.insertTimeRange(
        asset1Track.timeRange, of: asset1Track, at: asset1TrackStartTime)

That is all we need to do for the AVMutableComposition. But we also need to make an AVMutableVideoComposition:

    let videoComposition = AVMutableVideoComposition()
    videoComposition.frameDuration =
        min(asset0Track.minFrameDuration, asset1Track.minFrameDuration)
    videoComposition.renderSize = CGSize(
        width: max(asset0Track.naturalSize.width, asset1Track.naturalSize.width),
        height: max(asset0Track.naturalSize.height, asset1Track.naturalSize.height))

We need to set the video composition's instructions. The first instruction is to pass through just compositionTrack0 for the appropriate time range:

    // I'm using a helper function defined below.
    let compositionTrack0PassThroughInstruction = AVMutableVideoCompositionInstruction.passThrough(
        trackId: compositionTrack0Id, timeRange: compositionTrack0PassThroughTimeRange)

The second instruction is for the cross-fade, so it's more complicated. It needs two child instructions, one for each layer in the cross-fade. Each layer instruction, and the overall cross-fade instruction, use the same time range:

    let crossFadeLayer0Instruction = AVMutableVideoCompositionLayerInstruction()
    crossFadeLayer0Instruction.trackID = compositionTrack0Id
    crossFadeLayer0Instruction.setOpacityRamp(fromStartOpacity: 1, toEndOpacity: 0, timeRange: crossFadeTimeRange)

    let crossFadeLayer1Instruction = AVMutableVideoCompositionLayerInstruction()
    crossFadeLayer1Instruction.trackID = compositionTrack1Id
    crossFadeLayer1Instruction.setOpacityRamp(fromStartOpacity: 0, toEndOpacity: 1, timeRange: crossFadeTimeRange)

    let crossFadeInstruction = AVMutableVideoCompositionInstruction()
    crossFadeInstruction.timeRange = crossFadeTimeRange
    crossFadeInstruction.layerInstructions = [crossFadeLayer0Instruction, crossFadeLayer1Instruction]

The third instruction is to pass through just compositionTrack1 for the appropriate time range:

    let compositionTrack1PassThroughInstruction = AVMutableVideoCompositionInstruction.passThrough(
        trackId: compositionTrack1Id, timeRange: compositionTrack1PassThroughTimeRange)

Now that we have all three instruction, we can give them to the video composition:

    videoComposition.instructions = [compositionTrack0PassThroughInstruction, crossFadeInstruction, compositionTrack1PassThroughInstruction]

And now we can use composition and videoComposition together, for example to export a new movie file:

    let export = AVAssetExportSession(asset: composition, presetName: AVAssetExportPresetMediumQuality)!
    export.outputURL = outputUrl
    export.videoComposition = videoComposition
    export.exportAsynchronously {
        exit(0)
    }
}

Here's the helper I used to create the pass-through instructions:

extension AVMutableVideoCompositionInstruction {
    static func passThrough(trackId: CMPersistentTrackID, timeRange: CMTimeRange) -> AVMutableVideoCompositionInstruction {
        let layerInstruction = AVMutableVideoCompositionLayerInstruction()
        layerInstruction.trackID = trackId

        let instruction = AVMutableVideoCompositionInstruction()
        instruction.timeRange = timeRange
        instruction.layerInstructions = [layerInstruction]

        return instruction
    }
}

And here's my test code. I used a macOS command-line app for testing:

let asset0 = AVURLAsset(url: URL(fileURLWithPath: "/tmp/asset0.mp4"))
let asset1 = AVURLAsset(url: URL(fileURLWithPath: "/tmp/asset1.mp4"))

let outputUrl = URL(fileURLWithPath: "/tmp/output.mp4")
try! crossFade(asset0: asset0, asset1: asset1, crossFadeDuration: CMTimeMake(value: 1, timescale: 1), to: outputUrl)

dispatchMain()

Result:

Note that I had to make the animation tiny and low color because of Stack Overflow's limit on image file size.

Input videos courtesy of Jeffrey Beach.