I just saw a Youtube video on AI in the 3D industry, and I was curious to know if any of those features are planned for Blender, either 2.8, or maybe later…?
Like what? Can you post the youtube video for context?
1 Like
there’s an AI denoiser plugin somewhere in the forums.
I think AI retopo and “just do what I want” topology editing could be a thing very soon.
The key to AI is having a lot of data to train your deep neural networks on, so the easiest stuff to do are things where you can synthesize an unlimited amount of training examples. So for example, de-noising is really easy as you can grab a million images off the internet, add various types of random noise, and feed the noisy and original images in as training data.
Retopo and other modeling operations are a bit trickier to produce synthetic training examples for, but that’s just about the only problem you need to solve to make it work. All the pieces are there to do this otherwise. I think there are people who have been able to load Tensorflow and similar AI libraries into Blender’s Python environment for example, and the GPUs we use for rendering are exactly the tools you want for training neural nets too.
1 Like
This was the AI denoise addon for blender
and not too sure if it is procedural yet, but scene city looks very like what was int he video
I imagine all the improvements will only come to blender first as addons. It would be a very impressive addon to have it folded in to the main release. I would love to see the live video mo-cap in though
I am working on a denoiser which I designed around Cycles. Hopefully it gets production ready soon.
Blender 2.9 will have just one button. It renders out what it thinks its best to satisfy you. Dont need to do anything. No more learning. Just stay in bed.
2 Likes
Yes, impatiently waiting for the first alpha release.
1 Like
Note that the D-Noise plug-in is just a wrapper around the NVIDIA® OptiX™ AI-Accelerated Denoiser, it’s not a denoiser implementation itself.