I spent a little time the other day running several different Compressor recipes so I could compare different compression options. I thought I’d share the results here, in case they’re of use to other folks on the interwebs. And of course, if you’re reading this and you spot anything that looks wrong, please shout out – corrections and additional input encouraged…!
For my test I was working with a 4:05 movie edited in Final Cut Express and sourced in AVCHD on my Canon HF100. I started by exporting it to a 2.4 Gig QuickTime file for archive purposes. The main attributes of my first test, which turned out to be a pretty good recipe for YouTube, particularly when in a time crunch, were:
- bitrate max of 10,000
- full resolution
- left FPS as is
- single-pass compression
- deinterlace under the Frame Controls rather than Filter.
(Full specs on this later in the post, along with all the tests and a few notes on the results.)
A few top-line conclusions:
Deinterlace: Using the deinterlace filter in Compressor chewed up my text; using deinterlace under Frame controls worked like a charm.
Resolution: Dropping resolution by 50% hurt the image and softened text but didn’t really speed things up or shrink the file much, at least going from 1280 x 720 down to 640 x 360. Of course, there will be situations where you have to drop the resolution, but both YouTube and Vimeo suggest leaving HD resolutions at 1280 for HD, so in this case there was no reason to downsample.
Bitrate: Changing the bitrate had a direct impact on file size (2500 = half the file size of 5000) and a noticeable impact on the image, though going down to 2500, the image still looked pretty good for web video. YouTube currently suggests not capping bitrate, Vimeo requests that you set a max of 5000, which is what I did for example #6, below.
Multi-pass: Multi-pass adds a lot to encoding time (4X in this case) but did give me a higher contrast image with richer colors. So if time allows, it looks like multipass is better, but in a time crunch, single pass can work.