DeepMind releases Lyria 2 music generation model

Original Article ↗ Hacker News Discussion ↗

Access, rollout, and geography

  • Many object that “release” is misleading: DeepMind is offering a waitlist and limited “trusted tester” access, not an open product.
  • US‑only availability frustrates non‑US users and recalls previous Google experiments that were geo‑ and age‑locked.
  • Some see this as emblematic of Google: hypey research demos, opaque access, then quiet discontinuation.

Capabilities, quality, and missing features

  • Several commenters say current models (including Lyria, Suno, etc.) mostly generate bland imitations of mainstream pop, with weak musical “identity.”
  • There’s strong demand for:
    • Editing and remixing existing tracks
    • Proper stem separation and multitrack output (per‑instrument control)
    • MIDI / parameter‑level generation instead of muddy full‑mix WAVs
  • Open tools (e.g. stem separation models) are cited as partial workarounds; people want these integrated directly into AI music UIs.

Creative workflows vs “slop” generation

  • Some musicians report AI tools (e.g. Udio, Suno) are genuinely fun and productive, likening themselves to producers rather than performers.
  • Others find prompt‑based generation “painting with a shotgun” and missing the joy, nuance and skill‑building of playing instruments or working in a DAW.
  • A recurring view: AI music is fine for background/functional use (lo‑fi, hold music, game ambience, DnD sessions), but not compelling for active, attentive listening.

Impact on musicians, art, and meaning

  • Anxiety that AI will flood platforms with low‑effort tracks, making it harder for human musicians—especially emerging ones—to be discovered or paid.
  • Counter‑arguments:
    • There has always been vast amounts of low‑quality music; good human work still stands out.
    • New tools (from multitrack recording to autotune) were also decried but ultimately expanded who could create.
  • Many stress that audiences care about the persona, story, and human connection behind music; that’s hard to automate even if the sound is similar.
  • Others argue that for a large segment of listeners, music is just pleasant “organized sound” and AI is acceptable or indistinguishable.

AI art vs AI chores and Moravec’s paradox

  • A long subthread laments that AI is focused on creative work rather than physical chores (laundry, dishes, cleaning).
  • Multiple replies point out why: perception‑and‑manipulation robotics is far harder to scale than cloud‑based content generation, echoing Moravec’s paradox.
  • Debate extends into existential questions: if AI can do all valuable work (including creative), what meaning, careers, and social structures remain for humans?

Perceptions of DeepMind and future directions

  • Some are pleased Lyria 2 is framed as “tools for musicians” rather than direct replacement, but others see this as a stepping stone to automating more of the pipeline.
  • A number of commenters express fatigue with AI‑generated media and wish for ways to filter or block it; others are excited for more interactive, AI‑assisted creation embedded deeply into pro music software.