Physics based AI Animation tool
recently I came across this episode of Two Minute Papers discussing github.com/hr1juldey/DeepMimic youtu.be/eksOgX3vacs we have a few standard rigs in blender now, like riggify, blenrig,mblab, daz, auto rig pro, can someone implement this in blender. it seems that if we have a mediapipe mocap system and attachment of this system, we can easily get past the barrier to entry for animation.
10 Comments
last time someone showed the guts to make something from the papers it was flip fluids. and the rest is history, if the devs are willing they can make tools like the filmic, the cloth brush(industry pioneer) and this type of AI mocap tool for blender and give commercial bigshots a run for their money, I accept that it is only saving time in animation and not doing the animation itself, but at the end of the day it helps indie projects grow faster. and if marketed well it can make alot of money for blender development
i voted up on this , very interested in these technologies,
but letting computer do the work will not let you get past the barrier for understanding animation.. everyone has to learn that the hard way i guess ^^
edit: ohh this is not a good fit for blender , i thought i already saw the video. taking my vote back
anything you guys wanna say about this??
Yes: scientific papers don't make actual features.
most of us are tired of seeing every TMP video show up here as a feature request.
we should make a paper jar. Anyone posting a paper post must give a coin in the jar. Goes to Blender Funds.
please explain, if someone has already done the rnd and coding for you, what is stopping you from integration??
Because software is not magic. you can not just scan the paper into a machine and a feature pops out into your software- it takes a significant amount of work to go from "paper" to "feature", and that work is not free. If you have experience writing code, by all means jump in and help out! Then again if you had experience writing code, this proposal would probably not exist.
Think about it- there are dozens of papers written about Graphene. For almost 20 years we've been hearing about how it is going to change the world, and yet it's still vaporware. Same with CRISPR. And these are technologies with a huge profit potential to whoever can make it work in a non-lab setting. Software works the same way.
Papers are just controlled experiments, and they do not always work out in a practical sense.
well I use crisper every month, hehe I am a biology final year student, my question is that the code is available here, github.com/hr1juldey/DeepMimic , and I some how managed to run it once and second time it failed. I know it is hard but for blender to grow in this landscape of vr and virtual production era, we need both a solid foundation and ease of use features.
So you have just enough experience to be no help at all. Why do you think that would speed things up?