While being an interesting ML challenge that would need to generate valid code, it would also be useful for performers and VJ's.
A few possible approaches:
- Generate new Milkdrop presets
- Find similarity between Milkdrop presets
- Walk between groups of Milkdrop presets
- Use the mash-up technique (of the Milkdrop editor) to combine only select sections of code
- Automatically categorize Milkdrop presets
- A preset editor which suggests lines of code
The average Milkdrop preset is about 15 kilobytes and can be opened in a text editor. So the presets are human readable. Also the code of a single preset is organized into groups: initialization code, per_frame equations, per_vertex equations, warp shader, composite shader.Open-source Milkdrop projects to build on: https://github.com/projectM-visualizer/projectm --- https://github.com/mvsoft74/BeatDrop
=====
Dataset curated by hand
I've spent the last 7 months curating a best-of collection of Milkdrop presets. I went through 52,000 presets and selected only the best 9,795 Milkdrop presets. Includes screenshots of each preset.
https://thefulldomeblog.com/2020/02/21/nestdrop-presets-collection-cream-of-the-crop/
The presets have been organized into 11 category folders: Dancer, Drawing, Fractal, Geometric, Hypnotic, Particles, Reaction, Sparkle, Supernova, Waveform, Transition. Within each of these category folders the presets have been further organized into 183 various subcategory folders.
Full transparency, I'm part of the NestDrop team. I'd like to breath some life into preset creation. http://www.nestimmersion.ca/nestdrop.html