Can you share your configuration settings?
Standart)
这个app能否自制多轴视频呢？
trying this program
Look good.
The only problems I encounter is that the program don’t seem to like japanese caracter.
And it can have a huge load on you CPU, my R9 7950X3D can go up to 100%. Going between 20% and 100%
Don’t knwo for smaller CPU
1 Like
I pushed a new release to the Github. It adds Unicode support (so Japanese should no longer break everything), and lets you turn off motion balancing, which can improve accuracy on videos with static cameras.
@Ijuuin @y89451r @fakemeso @MotionOfDeeOcean @sleepy4567 etc.
Also, for transparency, Gumroad paid out what was owed, so better late than never on that front. It sucks that payment processors are so Christian Dominionist that most storefronts won’t touch sex tech for fear of being arbitrarily permabanned from capitalism, but at least it just cost me the business, not all the back revenue from it.
6 Likes
I did say it before and i’ll say it again, this is why you always go open source. Yeah you may not earn some revenue for your hobby work, but at the end of the day one person projects don’t take off like that unless you’ve already spent many years developing them. The absolute best path forward for this industry is democratization and transparency. When access to sex tech (be it hardware, software, knowledge, etc.) is free, open, decentralized, it cannot be controlled or silenced because it is not owned by any one, but by everyone. It is the ultimate form of freedom of information. And a nice side effect is you will always find people who are willing to help you or build upon your work if it is worthwhile.
The altruist in me would go as far as to say that open sourcing what you do in your spare time is the only way to avoid having your work be a complete waste of time from a collective standpoint. The backbone of IT infrastructure is built on the shoulders of those who were altruistic enough to understand this, and they were proven right.
3 Likes
Ideologically, I agree with you. FOSS is indisputably better for the world.
Unfortunately, I have a pressing personal need for an income that isn’t tied to the labor market. Losing this avenue (especially to censorship) was a setback, and I’m a little salty about it.
5 Likes
Until now, file names containing Japanese characters or platform-dependent characters (for example, ‘-’ instead of ‘-’) would sometimes cause the program to crash.
It may appear to be working, but the program’s progress was at 0% and it was slacking off.
I understand that it’s not perfect, but I appreciate your efforts to improve it.
1 Like
I remember trying this a while ago, but I had to invert all the scripts. (never came back to reply because it wasn’t a big deal)
Came back now because I came across the “3OVA” thread…
I’m wondering, what is funscript flow “expecting” as an input or how does the motion tracking work?
maybe there’s a setting i missed? not a big deal… just wondering about the inversion…
TL;DR: You might have better luck with POV mode.
The algorithm is pretty simple:
First, it uses something called optical flow, which is a way to take two frames of a video, and measure the speed and direction of motion is at each pixel (The math behind this is some differential equation that I don’t understand).
Then, for each point, it computes how much the rest of them are moving “towards” or “away” from it (the term for this is Divergence), and picks the one that has the highest value. When two people are moving relative to each-other, this is the center point between them. When one of them is stationary, it doesn’t know which way is forward or backward, and usually ends up picking an edge arbitrarily (there’s no machine learning here, just math, so it has no understanding of what’s actually going on in the image).
After that, it measures how quickly everything is moving towards or away from the detected center point, and just aggregates it over time to get a plot. A few quick cleanup passes (removing noise and drift) gets from there to a funscript.
POV mode fixes the center point to the bottom center of the screen, which makes the motion consistently register up for up and down for down. This won’t work as well for videos with inconsistent framing, but it’s a tradeoff.
2 Likes