While helping Rolanmen1 put up this encode:
https://www.youtube.com/watch?v=kbjYUoamBNM&feature=youtu.be
I wrote a script that took what Dolphin outputted and fixed it to the original pixel art. Hope this helps someone in the future. It requires you to capture at 1x native (640x480). Here you go:
Language: AviSynth
function FixPixelArt(clip c, int n) {
goodPixels = c.crop(n * 5, 0, 4, -0)
return n == 0 ? goodPixels : StackHorizontal(FixPixelArt(c, n - 1), goodPixels)
}
ImageSource(file="hh74sgC.png", start=0, end=119, fps=60)
FixPixelArt(127)
PointResize(last.width / 2, last.height / 2)
In a game i'm TASing, the fastest form of movement causes the screen to blur for a few frames once every second. In Dolphin, disabling fog removes the blur and also makes the game look better by removing a few other filters. This movement technique wasn't known until recently so most of the levels in the TAS don't do it. Is it possible to request that the official TASvideos encode disables fog in the later levels for entertainment reasons? I'll make a comparison video between fog vs. no fog once I finish the level i'm working on.
Probably better to ask this when submitting, but sure if you make a compare video, we can see what you mean.
FixFPS
Fix the frames to a more consistent pattern with the help of Matroska Version 2 Timecodes. Mostly useful for fixing weird framerate from using a third party program to capture (ex. .kkapture with Mupen). Generally used with the plugin ExactDedup. It should be noted that FixFPS would try it's best not to lose any frames from the source. It does this by pushing frames to the next available spot, if possible, but it can cause side effects.
Usage
Language: avs
FixFPS(clip c, string times="times.txt", int frames=30000, int div=2, int mul=1)
Where:
Language: avs
c The clip to use as the source
times The path to Matroska timecodes (v2)
frames The output frame count
div The divider of the original frame rate to align to
mul The multiplier of the frame rate after the divide
Example
Let's say you capture Super Mario 64 with .kkapture and Mupen. You know the gameplay of Super Mario 64 is 30 FPS, but the intro Nintendo logo is 60 FPS. Normally the capture from .kkapture is 120 FPS. So the divider is 120 / 30 = 4 (to align to 30 FPS) and multiplier is 60 / 30 = 2 to output to 60 FPS. Here is a possible script:
Language: avs
LoadPlugin("ExactDedup.dll")
LoadPlugin("FixFPS.dll")
AVISource("Mario64.avi")
trim(0, 5999) # You might want to align to the divider, so in this case a multiple of 4
ExactDedup(firstpass=false, dupinfo="Mario64 dups.txt")
FixFPS("Mario64 times.txt", 1500, 4, 2)
Changelog
2017/06/13:
- Added some smoothing of frames so there is less likely to get jerky movement when the multiplier is greater than 1
2016/02/26:
- Initial release (real date was a few years ago)
Special thanks to Rolanmen1 for the 64-bit DLL!
Download: https://www.mediafire.com/?7bip52hu7v1pps2
Yes I'm sure. His is if you are over 30 FPS or 60 FPS. Mine is if you are exactly NTSC or double NTSC which is under 30/60 FPS.
Edit: Masterjun was on the journey with me to find out =).
I wanted to see if YouTube keeps frames or adds frames when sending NTSC (30000/1001) and double NTSC (60000/1001) FPS videos. The result is that YouTube does keep the frames unmodified!
Here is double NTSC:
https://www.youtube.com/watch?v=TABndwIFfoY
My sample for NTSC was somehow in violation with YouTube's terms and was removed =(
I noticed the PS1 SCPH1000 bootup logo looks really bad for around two seconds:
It's even worse when deinterlaced. (SEIZURE WARNING) (note: 30 FPS, original is 60 FPS)
This is a pretty big issue. However, if you want to keep your video interlaced (see first image), the solution is thankfully simple:
Language: Avisynth
vid = AVISource("MGSI_1.avi")
FreezeFrame(vid, 289, 421, 288)
In this example, 289 is the first frame the screw-up happens, while 421 is the last frame of that happening. 288 is the last "clean" frame just before the screw-up. This effectively replaces frames 289 to 421 with frame 288, while the audio remains unaltered.
If you deinterlaced your video beforehand, well...
Language: Avisynth
vid = AVISource("MGSI_1deint.avi", audio=false)
audio = WAVSource("MGSI_1deint.avi")
a = vid.Trim(0, 287) # pre-screwup
c = vid.Trim(422, 0) # post-screwup
d = vid.Trim(286, 287) # fix screwup
vidfixed = a + d + d + d + d + d + *insert 60 more d's* + d + c
AudioDub(vidfixed, audio)
As you can see, you actually need 2 frames to fix the screwup that happens between frames 288 and 421 to keep consistency with the rest of the video. There may be a better solution , but it works. The reason why it's frame 288 instead of frame 289 in this example is because the video would be one frame too long otherwise. After all, we want to keep the same frame count.
For both solutions, you can re-save them as AVI (I recommend that, actually) with a lossless codec of your choice (e.g. Lagarith).
For the record, you don't have to do this when dealing with SCPH1001 (probably not even SCPH1002) or with newer PS1 BIOSes (SCPH5500-5502, etc).
I should chime in about the real problem. The real problem is that the option "Weave" in deinterlacing for PSX core randomly swap fields. The real solution is to change the deinterlacing to Bob and use the following code in AviSynth:
Language: avisynth
SeparateFields()
SelectEven()
Weave()
Edit: On second read, this may be a different problem, oops, but I'll leave this here to note the problem with PSX core.
At one time I was a contributor to the PrBoom-Plus project and implemented an exact capture system in it. If that code is still there and still works, it would be superior to kkapture.
That would be great! Only reason I don't use the internal capture was due to the melt effect is missing. This is being captured by .kkapture though. Maybe your code didn't get to the stable releases =/.
- Get PrBoom-Plus
- Use .kkapture Version 1.02 and capture at 35.0 FPS
- While capturing, turn off anti-virus as it sometimes interferes with .kkapture
- Use glboom and make sure it is in window mode
- Use the follow settings in glboom-plus.cfg
Note: You need to manually change the screen_resolution since the options are not there in the game itself
- Get the ogg sound pack of SC-55 and modify glboom-plus.cfg to use the music
- Capture twice, once for YouTube and 512kb file (2880x2160), and once for regular file and optionally high resolution file (2560x1600)
- Use AreaResize to downscale to 512kb file or regular file. Get it here: http://www.mediafire.com/download.php?kn56wh7r81vk2rx
- The resolution for 512kb file is 320x240 while the resolution for regular file is 320x200
- For YouTube, make sure you add ChangeFPS(60) to change the frame rate to 60 FPS since it is choppy at 30 FPS
Here is a comparison of downscale methods where it shows AreaResize is the best:
http://imgur.com/a/NkFVl
Sorry for the delay, but finally got the encodes going. Should I even bother to make an high resolution download encode as this has been superseded already? Edit: I decided not to.
By the way, I think all current Doom encodes published on YT is wrong. It should be all upconvert to 60 FPS to get those extra missing 5 fps. Currently the YT encodes look quite jumpy.
IIRC, kkapture will heavily drop frames if one captures at HD at 70 (or 60?).
I'm not saying to capture at 60, I'm saying to add the frames with AviSynth with ChangeFPS().
Edit: The download encodes have the wrong resolution also for the current published run.
This is borderline AI, I think, as it uses neural networks to figure out what looks best for humans:
Link to video
It's using an upscaler caller nnedi3.
By the way, I think all current Doom encodes published on YT is wrong. It should be all upconvert to 60 FPS to get those extra missing 5 fps. Currently the YT encodes look quite jumpy.
It's even worse when deinterlaced. (SEIZURE WARNING) (note: 30 FPS, original is 60 FPS) This is a pretty big issue. However, if you want to keep your video interlaced (see first image), the solution is thankfully simple:
In this example, 289 is the first frame the screw-up happens, while 421 is the last frame of that happening. 288 is the last "clean" frame just before the screw-up. This effectively replaces frames 289 to 421 with frame 288, while the audio remains unaltered. If you deinterlaced your video beforehand, well...Language: Avisynth