🚨To the people following me who never interact: I’m doing a big cleanup. Right now, the X algorithm keeps shoving my posts in front of people who don’t like, reply, repost, or even tap anything. That tells the system, “This content is boring,” and it stops showing my work to the people who actually do care. So I’m going to start removing a bunch of inactive or non-engaging followers. This isn’t personal. People’s interests change. Some of you followed me years ago for a totally different reason, or from one random post, or by accident. That’s fine. But I need my audience to be made up of people who actually want to see what I’m making now. If you’re reading this and you do want to stay connected, do me a favor: like, reply, or repost this or just interact with something I’ve posted recently That tells the algo you’re still alive on this feed. If you’re not into my stuff anymore, you’re absolutely free to unfollow. No hard feelings at all. I’d rather have a smaller group of people who are genuinely into what I share than a bloated number full of ghosts. To everyone who comments, DMs, shares, and supports: you’re who I’m here for. Let’s keep building around that.

Demucs separates the track into musical stems while we drive continuous latent motion across beat, bar, phrase, and section timescales; meanwhile the detected melody maps chromatic notes directly onto 24 landscape classes, cycling through environments as the melody moves.

He's the funky zero. Vocals now act as the class navigator. The scene holds one environment for three beats, then the vocal contour decides whether beat four smoothly moves to the neighboring class above or below, wrapping across 0↔23. Noise injection is different 4 both renders

New audio-reactive StyleGAN experiment: stereo channels independently drive latent motion through Demucs + MIDI/PCA, while a class-conditional landscape model continuously morphs between 24 environments every four beats

Audio-Reactive Stylegan V5B turns stereo music into a 3-panel StyleGAN performance: left and right channels independently drive latent motion via Demucs + MIDI/PCA, while a mirrored center visual holds one side for 3 beats, then morphs to the other on beat 4

Working on a landscape stylegan and some of the Krea2 rendered training images are incredible.

Built a new audio-reactive StyleGAN setup: stereo is split into separate left/right visual streams, Demucs separates instruments, MIDI transcription turns notes/chords into PCA steering, and rhythm still controls timing. Harmony now shapes where the visuals move, not just when

10/10 this is v4 of my stylegan audio visualizer. V4 turns music into StyleGAN motion using stems, beat/bar/phrase timing, section changes, multi-timescale envelopes, chroma/PCA controls, latent routing, noise modulation, and dynamic truncation.

9/10

8/10

7/10

6/10

5/10 includes audio produced by a custom trained model using my own music, includes video from a custom trained stylegan video using my own art. i pushed a button

4/10 first audio reactive stylegan video i've made in years

3/10

2/10

1/10