IceDreamer;
Sadly, there are a few reasons why we won't see real 'high quality' smooth motion in our lifetimes (which I will quickly list below).
1) Temporal Aliasing and The Aperture Problem both place strict mathematical limits on the accuracy of interpolated frames, even with an infinitely fast computer and a mathematically 'perfect' algorithm.
2) Most consumers DO NOT want movies\videos to have smooth motion (look at the backlash against The Hobbit's use of 48-fps, which is not even close to actual 'smooth' motion) and cannot be made to understand that they can just trow away all the 'extra' frames to make the motion as jittery as they would like, while we cannot just make up new frames to get the motion as smooth as we want (so even though we have had the technical feasibility to do so for decades already, the free market simply cannot sustain such media creation).
3) Gains in practical, workstation-level performance have slowed to a crawl since the end of the 2000's. So, while we do have some very high quality motion interpolation algorithms at our disposal, it will take a very long time before we will see the requisite practical (i.e. Single-Threaded performance equivalent) computational power at 50% market penetration.
However, it is still possible to at least generate an interpolated video stream with an acceptable trade-off between artifacts and smoothness (mainly because 24-fps is so incredibly bad, that even a relatively large amount of artifacts can be generated without being as distracting as trying to resolve motion at 24-fps). Unfortunately, SVP uses literally the worst (quality wise; it is also the fastest / cheapest) of the practical motion interpolation algorithms (in any case, SVP's main goal has always been to try to bring motion interpolation to the attention of the masses and not to implement the highest quality quality motion interpolation algorithms; which would be pretty expensive to develop and too demanding for most people's systems to handle).
That being said, I have found that the following produces, on average, the best PSNR (which correlates very strongly with my own visual perception, but apparently not with that of most other people) on a variety of real-life content (NOT Anime!):
/***** SVSuper options *****/
levels.pel = 3;
levels.gpu = 0; //SUBJECTIVE: Compare performance and quality with gpu=1 for yourself
levels.scale.up = 2;
levels.scale.down = 4;
levels.full = true;
/***** SVAnalyse options *****/
analyse.vectors = 3;
analyse.block.w = 16;
analyse.block.h = 16;
analyse.block.overlap = 3;
analyse.main.levels = 0;
analyse.main.search.type = 4;
analyse.main.search.distance = -24;
analyse.main.search.sort = true;
analyse.main.search.satd = true; //VERY slow if 'true', but gives slightly higher quality
analyse.main.search.coarse.type = 4;
analyse.main.search.coarse.distance = -24;
analyse.main.search.coarse.satd = true;
analyse.main.search.coarse.trymany = true;
analyse.main.search.coarse.width = 1920; //This is for 1080p content. Use 4096 for 4k, however it may crash, run out of memory, etc.
analyse.main.search.coarse.bad.sad = 190;
analyse.main.search.coarse.bad.range = -38;
analyse.main.penalty.lambda = 11.2;
analyse.main.penalty.plevel = 1.65;
analyse.main.penalty.lsad = 4000;
analyse.main.penalty.pnew = 40;
analyse.main.penalty.pglobal = 50;
analyse.main.penalty.pzero = 80;
analyse.main.penalty.pnbour = 30;
analyse.main.penalty.prev = 20;
analyse.refine[0].thsad = 170;
analyse.refine[0].search.type = 4;
analyse.refine[0].search.distance = -38;
analyse.refine[0].search.satd = true;
analyse.refine[0].penalty.lambda = 9.8;
analyse.refine[0].penalty.lsad = 3900;
analyse.refine[0].penalty.pnew = 50;
/***** SVSmoothFps options *****/
smooth.rate.num = 3; //Set this to auto, or to your <TargetFrameRate>/<SourceFrameRate>
smooth.rate.den = 1;
smooth.algo = 23;
smooth.block = false;
smooth.cubic = 1;
smooth.linear = true; //Only works with GPU rendering, i.e. use gpu=1
smooth.mask.cover = 40; //Therse values are your main trade-off between smoothness and artifacts
smooth.mask.area = 15; //Making them larger decreases smoothness and obtrusive artifacts
smooth.mask.area_sharp = 1.65; // See https://www.svp-team.com/wiki/Plugins:_SVPflow for details
smooth.scene.mode = 0;
smooth.scene.force13 = false;
smooth.scene.luma = 1.66;
smooth.scene.blend = false;
smooth.scene.limits.m1 = 1600;
smooth.scene.limits.m2 = 2800;
smooth.scene.limits.scene = 3000;
smooth.scene.limits.zero = 180;
smooth.scene.limits.blocks = 33; //Adjust this value for scene-change detection. Higher values makes it less sensitive and might cause it to try
//interpolating scene-changes!
Unfortunately, your system is not even remotely powerful enough to get an acceptable real-time result, however you can still try these settings out by running an AviSynth conversion (through VirtualDub) on a test clip (i.e. Create an AviSynth script, open said script in VirtualDub, output the RAW avi, view in video player WITHOUT SVP).
Just use the above settings to create the strings as per this documentation page:
https://www.svp-team.com/wiki/Plugins:_SVPflow
It will take quite a long time, so I would recommend only parsing a hundred or so frames (use something like the AviSynth Trim or Select* commands) and letting it run overnight.
