Test results for encoding experiments;
I'll be using Simple x264 Launcher for my experiments.
Using a fraps chunk from the final battle of God of War 3 that is around 3:30 minutes long 720p, I set out to create a baseline set of 8-bit encodes for GOW3 using three Constant Quality settings;
(Lower is higher quality/more size)
30 - What I used for GoW2 I think. Sometimes I used as low as 32.
25 - Very high quality given other settings not being altered.
20 - Stupidly high. I don't think I've ever used this high constant quality before.
45fps @ CRF 30
Baseline size - 4gb
30 size - 50mb
25 size - 94mb
20 size - 182mb
Image quality;Note that imgur converts the images just like a shitty image host would.Also note that these images may contain spoilers.Conclusion - 8bit
8bit always butchered my Vegas encodes. While the "8bit" we get out of these videos is much better than what I ever got out of Vegas, a CRF of 30 alone is not going to work very well for high-motion high-detail videos. If I push it lower I get more respectable returns but the size increase is dramatic. I can cut hairs and go for 28/26 but ultimately I want to deliver an experience as close to the source without breaking the host in the process.
The compression occurs the worst in any kind of dark colors. The rocks on the left, the rim of the very top metal bits, these all lose a shitton of quality on the lower settings. In a low-quality game like Conan this is of little consequence, but for GoW3 I want the best of the best.
The encode times are pretty quick for such a small video, but the 20 CRF is notably slower.
Let's try 10-bit.
33fps @ CRF 30
30fps @ CRF 25
30 size - 50mb
25 size - 94mbConclusion - 10bit
I expected massive changes out of 10-bit but can see virtually no changes at all upon first glance. The difference in CRF between 8bit and 10bit is practically non-existent, with only returning 300 kb out of a 90mb file. I didn't even bother uploading the screencaps because they are nearly identical - the blur is just in slightly different locations. There is no difference at all in the darks. However, as I study the images very closely, I notice that the major difference out of the minor differences is some edges with brights and darks are slightly less blurry. It's an improvement, a very subtle improvement.
While the encoding speeds are indeed slower they are not significantly slower enough to impact my pipeline. Something I had read was that 10bit's changes were so dramatic that you'd want to record with lower quality settings - I tried this and it's just as vomity and ugly as I'd expect. The impact I'm seeing here is nitpicky at best, absolutely nothing like what was described in various guides.
I decided to pursue tuning settings next, starting with Animation. This, too, had no impact. Grain, on the other hand, the setting that Nef uses for some of his older game encodes, produced some interesting returns.
The size increased from 50 to 76MB on a 30 CRF, but the image quality sky rocketed.Direct PNG screencaps via FTP
10-bit 30 CRF with Grain tuning: show
When paying attention to the flames on the box in the middle and the characters, there's little change to be seen in screenshots other than the obvious loss between baseline and compression. The goal of our efforts is to get rid of that loss. (It also seems the baseline is a different frame wtf) The Grain modifier definitely helps it in motion, and the size increase is acceptable. If I boosted the CRF to 26-28, the differences will be even less significant though you'd probably be looking at close to 100mb for 3 and a half minutes with no audio. Ouch. I was hoping to reduce the size demands, not increase them. Lowering the CRF to 32 brings us pretty close to the original size @ 59mb in exchange for some blur making it back in - I wouldn't use this for GoW3. Maybe CRF 31?
The size increase with Grain is due to the motion prediction. But Simple X264 Launcher offers no motion prediction settings. I slapped together a basic profile of CRF 30 and Grain in megui (using 8bit) but the encode times are literally 1/3 as fast and it's predicting an even larger file. Well then.Major tests
A test encode with GoW3 using CRF 30 and Grain in 10-bit brings out a 2gig file for 2:26 hours (with no audio). A perfectly acceptable return. The quality is pretty decent. Not perfect, but workable.Extras
mkvmerge is still giving me issues with videos regardless of the program version. The issues include the DXVA bsods from Alice and the "freeze" issues that Conan seg 6 have had and are much more numerous. As of yet I've been unable to find a replacement program or any mention of this problem elsewhere.God of War 3
I want to switch to 10-bit on behest of potential advantages, but I'm still hesitant.
The biggest issue with GoW3 I'm seeing in regards to the first segment right now actually has nothing to do with the quality
. Rather, the camera shaking. GoW3's camera shaking is the most stupidly violent camera shaking in all of the games I've LP'd thus far and, frankly, it makes me sick. I get motion sickness and the first 30 or so minutes of gow3 literally gives me a headache. It was like this during the gameplay but not nearly as bad as the encode. There could be a few reasons why.
1 - Fraps doesn't handle high motion 1:1. In fact, fraps is dropping/skipping/fucking up frames during the intro cinematics and you can see they are kind of "choppy". Possibly due to the fps getting halved and extra frames getting dropped randomly. I can possibly avoid this in the future by unlocking the FPS (same thing I plan to do when I test N64 emulation again).
2 - The grain tuning might be fighting motion prediction, and motion prediction hates
camera shaking. I have two encodes, one with grain and one without, but it's next to impossible for me to discern any real defining difference in the way the motion is handled, especially when I already feel queer.
Ultimately this problem has little bearing on 10-bit. I'm going to say, update your players if you haven't for some reason. I'll need to test this more carefully and more slowly. I know most people don't get motion sickness like I do (the flight points in Tera have the same effect), but I want to make something as comfortable for me
to watch as possible. How to tackle this unique problem I do not yet know. Whether it's related to the grain tuning I don't know. Worst case, I drop Simple X264 launcher, go to megui, and do some major fuckery with motion prediction to try to achieve something.
I can say with 69% certainty I will probably switch back to megui for GoW3 due to the new motion prediction settings I'm using since they alleviate my motion sickness a bit. So, there is unlikely to be an immediate switch to 10bit.
That said, segment 1 is just about verified, so this shouldn't take very long either way.
As an aside, I'm removed almost all non-public video material from the FTP to trim space.