Dragon daydream

[Sadler]

Only the final few steps done in Magic.

[media]

[skarabee]

impressive! Kind of fractal Dragon.
If only last step done in Magic, what do you use before?

[Sadler]

The visual is using a PyTorch ML script called Illustrip and the music is composed by AIVA. There are better scripts but I don't have the hardware to run them.

Maybe obvious to most, Magic was used to react, pace and match the video to the music plus a little prost-processing.

[TKS]

The visual is using a PyTorch ML script called Illustrip and the music is composed by AIVA. There are better scripts but I don't have the hardware to run them

Being a lifetime fan of surrealistic art I like this a lot. I've never heard of Illustrip before. What hardware did you use and how long did it need to render this?
As a Linux user I hate to ask: does it run on Windows? Because of the software I use my multimedia computer is the only one in my house running WIndows and therefore has the juicy hardware needed for that.

BTW, I'm playing around with AIVA as well, when you need original music for your videos it's a great help and source for inspiration!

[Sadler]

Yes, it is running on Windows but I understand its an easier job to setup on Linux. This script runs on my laptop with an RTX 2080 8GB and this script took about 5 hours, if I remember correctly. You absolutely need NVidia because CUDA. The major limitation isn't the GPU (though it is required) it is the memory it has. 8GB is the very minimum, 12 would be better, 24 in a 4090 would be great. One day perhaps.

[TKS]

Did you use AdaBins for the 3D effect? I'm just testing Aphantasia, it's cool but doesn't looks like yours.

Edit: nevermind, I've used the (official) release v1.2b, which doesn't has illustrip.py.

Cloning the master branch did the job

Edit 2: care to share the text you've used? Is it in general better to have real sentences or just single words per line? Repeat important words to have one theme or does it remember what was in the first few lines?

[Sadler]

Sure, the texts were all two word phrases containing the word dragon, e.g. Chinese Dragon etc. As far as I can tell, the concepts are broadly mixed between scene changes and even persist longer. I haven't experimented enough to know the optimal phrases yet - working on it.

[Vegadeth]

The visual is using a PyTorch ML script called Illustrip and the music is composed by AIVA. There are better scripts but I don't have the hardware to run them.

Maybe obvious to most, Magic was used to react, pace and match the video to the music plus a little prost-processing.

These fractal style morphs are amazing! I did a quick search on Illustrip and did not find anything I felt safe to click on. Any insight on how to get Illustrip and if it is difficult to use?

[Sadler]

Check out Nerdy Rodent (and others) on YT who has some tutorials about how to set up and environment for running the python scripts - specifically for ML based art. The main steps are: install WSL, Anaconda and CUDA. Set up a conda (anaconda) environment and within that install (using pip or conda) the requirements (python, numpy etc) for your chosen repository.

Alternatively, find a colab. Here's the colab for Illustrip:
https://colab.research.google.com/githu ... ip3D.ipynb

The current daddy of the ML video art is PyTTI which you'll see a lot of on the various socials. It also has a colab but it is currently behind a patreon paywall.