I eventually found a solution by digging through a lot of half complete samples and poor AVFoundation documentation from Apple.
The biggest confusion is that while at a high level, AVFoundation is "reasonably" consistent between iOS and OSX, the lower level items behave differently, have different methods and different techniques. This solution is for OSX.
Setting up your AssetWriter
The first thing is to make sure that when you set up the asset writer, you add an adaptor to read in from a CVPixelBuffer. This buffer will contain the modified frames.
// Create the asset writer input and add it to the asset writer.
AVAssetWriterInput *assetWriterVideoInput = [AVAssetWriterInput assetWriterInputWithMediaType:[[videoTracks objectAtIndex:0] mediaType] outputSettings:videoSettings];
// Now create an adaptor that writes pixels too!
AVAssetWriterInputPixelBufferAdaptor *adaptor = [AVAssetWriterInputPixelBufferAdaptor
assetWriterInputPixelBufferAdaptorWithAssetWriterInput:assetWriterVideoInput
sourcePixelBufferAttributes:nil];
assetWriterVideoInput.expectsMediaDataInRealTime = NO;
[assetWriter addInput:assetWriterVideoInput];
Reading and Writing
The challenge here is that I couldn't find directly comparable methods between iOS and OSX - iOS has the ability to render a context directly to a PixelBuffer, where OSX does NOT support that option. The context is also configured differently between iOS and OSX.
Note that you should include the QuartzCore.Framework into your XCode Project as well.
Creating the context on OSX.
CIContext *context = [CIContext contextWithCGContext:
[[NSGraphicsContext currentContext] graphicsPort]
options: nil]; // We don't want to always create a context so we put it outside the loop
Now you want want to loop through, reading off the AssetReader and writing to the AssetWriter... but note that you are writing via the adaptor created previously, not with the SampleBuffer.
while ([adaptor.assetWriterInput isReadyForMoreMediaData] && !done)
{
CMSampleBufferRef sampleBuffer = [videoCompositionOutput copyNextSampleBuffer];
if (sampleBuffer)
{
CMTime currentTime = CMSampleBufferGetPresentationTimeStamp(sampleBuffer);
// GRAB AN IMAGE FROM THE SAMPLE BUFFER
CVImageBufferRef imageBuffer = CMSampleBufferGetImageBuffer(sampleBuffer);
NSDictionary *options = [NSDictionary dictionaryWithObjectsAndKeys: [NSNumber numberWithInt:kCVPixelFormatType_32BGRA], kCVPixelBufferPixelFormatTypeKey,
[NSNumber numberWithInt:640.0], kCVPixelBufferWidthKey,
[NSNumber numberWithInt:360.0], kCVPixelBufferHeightKey,
nil];
CIImage *inputImage = [CIImage imageWithCVImageBuffer:imageBuffer options:options];
//-----------------
// FILTER IMAGE - APPLY ANY FILTERS IN HERE
CIFilter *filter = [CIFilter filterWithName:@"CISepiaTone"];
[filter setDefaults];
[filter setValue: inputImage forKey: kCIInputImageKey];
[filter setValue: @1.0f forKey: kCIInputIntensityKey];
CIImage *outputImage = [filter valueForKey: kCIOutputImageKey];
//-----------------
// RENDER OUTPUT IMAGE BACK TO PIXEL BUFFER
// 1. Firstly render the image
CGImageRef finalImage = [context createCGImage:outputImage fromRect:[outputImage extent]];
// 2. Grab the size
CGSize size = CGSizeMake(CGImageGetWidth(finalImage), CGImageGetHeight(finalImage));
// 3. Convert the CGImage to a PixelBuffer
CVPixelBufferRef pxBuffer = NULL;
// pixelBufferFromCGImage is documented below.
pxBuffer = [self pixelBufferFromCGImage: finalImage andSize: size];
// 4. Write things back out.
// Calculate the frame time
CMTime frameTime = CMTimeMake(1, 30); // Represents 1 frame at 30 FPS
CMTime presentTime=CMTimeAdd(currentTime, frameTime); // Note that if you actually had a sequence of images (an animation or transition perhaps), your frameTime would represent the number of images / frames, not just 1 as I've done here.
// Finally write out using the adaptor.
[adaptor appendPixelBuffer:pxBuffer withPresentationTime:presentTime];
CFRelease(sampleBuffer);
sampleBuffer = NULL;
}
else
{
// Find out why we couldn't get another sample buffer....
if (assetReader.status == AVAssetReaderStatusFailed)
{
NSError *failureError = assetReader.error;
// Do something with this error.
}
else
{
// Some kind of success....
done = YES;
[assetWriter finishWriting];
}
}
}
}
Creating the PixelBuffer
There MUST be an easier way, however for now, this works and is the only way I found to get directly from a CIImage to a PixelBuffer (via a CGImage) on OSX. The following code is cut and paste from AVFoundation + AssetWriter: Generate Movie With Images and Audio
- (CVPixelBufferRef) pixelBufferFromCGImage: (CGImageRef) image andSize:(CGSize) size
{
NSDictionary *options = [NSDictionary dictionaryWithObjectsAndKeys:
[NSNumber numberWithBool:YES], kCVPixelBufferCGImageCompatibilityKey,
[NSNumber numberWithBool:YES], kCVPixelBufferCGBitmapContextCompatibilityKey,
nil];
CVPixelBufferRef pxbuffer = NULL;
CVReturn status = CVPixelBufferCreate(kCFAllocatorDefault, size.width,
size.height, kCVPixelFormatType_32ARGB, (__bridge CFDictionaryRef) options,
&pxbuffer);
NSParameterAssert(status == kCVReturnSuccess && pxbuffer != NULL);
CVPixelBufferLockBaseAddress(pxbuffer, 0);
void *pxdata = CVPixelBufferGetBaseAddress(pxbuffer);
NSParameterAssert(pxdata != NULL);
CGColorSpaceRef rgbColorSpace = CGColorSpaceCreateDeviceRGB();
CGContextRef context = CGBitmapContextCreate(pxdata, size.width,
size.height, 8, 4*size.width, rgbColorSpace,
kCGImageAlphaNoneSkipFirst);
NSParameterAssert(context);
CGContextConcatCTM(context, CGAffineTransformMakeRotation(0));
CGContextDrawImage(context, CGRectMake(0, 0, CGImageGetWidth(image),
CGImageGetHeight(image)), image);
CGColorSpaceRelease(rgbColorSpace);
CGContextRelease(context);
CVPixelBufferUnlockBaseAddress(pxbuffer, 0);
return pxbuffer;
}