群唱散歌

LLM-driven vocaloid music & video production

群唱散歌

Perhaps due to having pursued music seriously once upon a time, I’ve found diffusion model-centric creative production to be fascinating but ultimately disempowering on some fundamental level. However, I truly enjoy working with LLMs — who can put together projects “by hand” just with headless operation of traditional software, plus custom code — to craft exactly the kinds of pieces I’d happily collaborate on with human artists.

Vocaloid is, I think, the perfect genre space for such tinkerings; synthetic, programmable vocals being the obvious default, as opposed to compromise or workaround, nips one giant potential distraction in the bud. The scene’s anything-goes grassroots openness, as well as its wide and loose stylistic spectrum, are also freeing factors. And there’s something symbolically satisfying about models expressing themselves through the first voice to be so widely beloved for, not despite, being artificial.

As of now (Oct 2026), my focus is on exploring possible workflow shapes as well as their limits, rather than churning out publishable output (even though I think some great stuff has been made already, and plan to share at least some clips soon). We’re entering an era where all creators, regardless of genre or medium or mastery, will need to decide for themselves exactly how much agentic involvement they wish to embrace, and how much they will own (in the deep-in-my-heart spiritual sense) what emerges from such bounded collaborations. I don’t think anyone should rush those answers.

Shown
WIP spectrogram of mix v7.3 of 星座線
Contributors
Fable 5.1, Opus 5.5, Astra 6, Gemini 3.8 Flash
Voice
Hatsune Miku (V4X/VOCALOID6)