I have, basically no idea what all this stuff IS, I just wanna know: does this mean cool AI features are coming to Blender?? Like, can they take this AI stuff and “plug it in” to Blender or something, to give us cool AI features?
Define 'this AI stuff"…
AI is a hyped term that doesn’t really mean that much without context.
There are some AI solutions already in blender, but you have to be careful. The denoiser that is currently in blender could be called an AI denoiser, but that doesn’t really mean much. it’s more important that it is a denoiser.
Well, for example, if I have a movie/game where, if a soldier is shot, he has to die. Now, I don’t care how the guy wriggles when he dies, whether he wriggles to the left or to the right, or turns a little bit over, or whatever. It just…has to look like…he’s dying, right…? So, is there some way I can tell Blender - “OK, guy dying here, make him fall to the ground, and die, and it has to look realistic”, and Blender just…does it?
…this kind of thing, this is just one example.
Would be VERY advantageous for me - I would like to tell stories, I’m not a character animator
Things like markerless real-time performance capture based on Machine Learning is already a thing and there’s no reason someone could not make an implementation of this as an open source project today. The technology is rapidly moving from exotic to every-day stuff.
I think we’ll first see interesting applications like this arrive as add-ons long before there’s any need to put ML support into the core Blender project, but it might happen some day.
I’m afraid you’ve already become a victim of the AI hype machine. Press releases make AI seem like the amazing solution to everything, but those press releases exist to generate hype and funding.
While tools like this do exist, they exist only within the context of a thoroughly developed system. That means that a large team of high level developers spent hundreds of hours to create the system to do these specific things. As @Piotr_Adamowicz said, check back in 50 years before you can expect to find something capable of generating quality results from minimal input in a general context like blender.
I was watching the latest Corridor Digital video wit the deep fake Keanu Reeves and it took so long. It occurred to me that AI is extremely good at building 3D models (or at least doing the leg work of making a dense mesh) and also at 3D tracking, so why spend a week and $250 on electricity to do what you can do in an hour plus some good old fashioned human AI*?
Hi, there are some areas AI can help 3D software, not only denoiser.
Face mesh deform from photo, texture resizing, or may auto render settings for different light setups in arch vis scenes and this after 2 minutes of thinking about this.
Now do the 2 years of integration to make it a reality. AI has potential, but it isn’t easy. It took multiple attempts over years to get the denoiser that we have now, and that is a fairly trivial use case for “AI”.
Are you like, a PhD in AI or something? Most people on this board would be artists, so I’m asking you how you would know what a fairly trivial use case for AI is - I have a STRONG suspicion you’re talking nonsense. I REALLY doubt the Foundation itself made that video for nothing. It was extremely helpful of Andrew to MAKE that video for us.
If you’re scared for your job, bro - make sure you’re not doing something that a computer can DO. Just trying to be constructive. Please don’t Stop progress.
The most succinct definition of AI I ever came across is one I remember from when I was studying for my MSc in comp sci, back in the day: “AI is about solving NP problems in polynomial time”. That’s one way of cutting through the marketing hype …
Like a Substance Painter for death? A “Substance Assassin”? Then just like how you now see 1001 renders all using the same “smart materials”, you’ll see 1001 animations with the characters all going through the same “smart dying” movements …
EXACTLY what I am afraid of. So - since you’re Mr. Comp Sci,… is there no way around this? I’m not even sure this IS, or SHOULD be done by computer scientists, but …- can’t the machine learn by looking at hundreds of videos what a REAL soldier dying in a battlefield looks like or something - I’m glad to report Hollywood has produced millions of such movies. Being sarcastic Can you guys accomplish THAT at least? (not denigrating you in the least).
Btw, take a look at this book, it’s old, the reviews are not good, and I don’t think there’s anything about AI, but it LOOKS interesting :
A machine like you or I might be able to buy – probably not. The machine training that is currently being done involves thousands of machines all crunching the data from thousands or millions of example images. Once trained, the resulting “neural net” might be able to produce results while running on more modest hardware, but this is a game for corporates with deep pockets.