A new rabbit hole (AI art & music)

Apologies if I am dragging anyone else down a black hole, but, I recently discovered a new app called Visions of Chaos

I am interested in this mainly because I was trying to get into generative art using GANs and Style Transfer etc

But having downloaded the app, I now see there is so much more potential in it, for a vast number of creative endeavours, including music

image

So I thought I’d share it here and see what amazing stuff you all can come up with.

Here are a few art pieces on the theme of modular synthesizers that I generated using the text-to-image AI scripts in the app:

11 Likes

I gotta say, Dan… I find those images impressive in their way, but also rather disturbing. And knowing that they were generated by AI, only moreso. One might be inclined to the interpretation that the AI equates synth modules with viscera - like they’re dismembered computers with all their inside bits on the outside. - and yet they still process! :scream:

Huh. Some of those look even more disturbing than some of the Google Deep Dream Generator images.

Hah!

Do not worry, our future machine overlords are not to blame, these are from a text-to-image script, which means that I input a text prompt in order to generate them. Their disturbing nature is due to the actual words I used in the text prompt. It is very possible to create a rainbow fairy land synthesizer art piece, if you use the right words.

For example, part of the text prompt for the 3rd piece was

bloodstained Moog synthesizer

1 Like

Trying to evolve in a less disturbing direction

1 Like

Wow, everyone that I generate just seems to get better, I love this, have to share these two new ones:

chrome audio interface with wires

oscilloscope wired up to a robot

2 Likes

Incredible - every one of them is an awesome album cover awaiting its music!

2 Likes

These look awesome and super inspiring. Pls don’t apologize! This is exactly the kind of rabbit hole that I‘d want to be dragged into :slight_smile:

If anyone wants some art generated (for an album cover or whatever), feel free to message me and I’ll see what I can create for you, I am really loving this process

4 Likes

I love it.

Sounds like the impetus for a really fascinating vlog!

Wonder what my logo icon (avatar) would look like lol

interesting images O.O

Gigeresque VCV logos anyone?

6 Likes

That is so strange but somewhat beautiful. Would you mind sharing how difficult it is to set this up? Installation, options, what to learn?

Fascinating. Like a very far forward version of Wombo Dream.

The main description page doesn’t seem to mention text->image scripts (well, it’s a looong list, so I wasn’t that thorough). What do they call it in the app?

Nice! getting some Giger vibes lol

1 Like

Wow! How do you express what you want, when it’s as specific as the VCV logo?

Sure, though fair warning, it is not simple or easy:

Note that text-to-image is just one of many modes available in Visions of Chaos, you can just download the app and play with everything that is not under the machine learning section and still have a great time

Now, specifically for the machine learning modes, there is a lot of pre requisites, you need an Nvidia RTX 20x or 30x GPU and then you must following these instructions for installing everything Softology - Visions of Chaos - Machine Learning Support

Once you have all that stuff installed, you need to activate one of the machine learning modes, then the app will prompt you to download all the models, I think currently there are 51 different models, each one can be multiple GB, these step was the hardest part, because the download can be a little bit flaky.

If you get all the models downloaded, then finally you can start generating images, my advice for that is to start really simple and at small resolutions, change one thing at a time and re-render, this way you will start to understand what each parameter affects

If you get stuck but want to persist, the author of the app has a discord and he is very active and responsive there

An alternative is to use something called Google CoLab, this is a cloud based virtual machine solution for running machine learning notebooks shared by other people, though this is just as complex, if not more so, and to really get the best benefit you would need a subscription to the pro version, but i’ll leave this tutorial in case thats of interest Quick & easy tutorial for Disco Diffusion V4.1 Google Colab. Make your own pictures! - YouTube

One of the options for some of the text-to-image scripts is to provide a seed image

What I did for those images was seed the VCV logo, set the Init scale, Skip timesteps & Guidance scale to a larger value (these control how alike to the seed image the generated image is) and then I wrote various text prompts that describe what I think the content of a Giger picture is.

For example, I think one of the prompts I wrote was Bleached biomech bones by H R Giger

In the end you just have to use trial and error until you arrive in a place where you have some idea of what each parameter does, and then tailor your text prompt to the result you are looking for. But even then there is still an element of randomness, so you will still sometimes end up running a workflow multiple times because you don’t like the output.

4 Likes