Udio: Generate music in your favorite styles with a text prompt
An AI music generator called Udio is drawing attention for producing high-quality, stylistically flexible songs from text prompts, often outperforming existing tools like Suno in vocal realism and sound quality. Early users praise its creative potential and entertainment value but flag issues with slow performance under load, limited account options, and a one-shot, text-only workflow that offers little fine-grained control. The launch also intensifies ethical and economic concerns around training data, voice likeness, and the impact on working composers and other audio professionals, prompting calls for clearer copyright rules and possible watermarking of AI-generated music.
Link, access, and onboarding
- Several dislike that the HN submission and launch were via Twitter, preferring a direct link to the site.
- Many are frustrated that signup requires Google, Discord, or X; several explicitly request plain email registration.
- Some speculate this is for spam/abuse prevention and cheaper “real user” verification than phone numbers.
Quality, capabilities, and comparisons
- Many describe the output as strikingly good, often “better than Suno v3,” especially in sound quality and vocal expressiveness.
- Example styles include gospel, barbershop, opera, Broadway musical, punk, jazz, tech death metal, comedy songs, and children’s songs.
- Long-form structure (e.g., solo classical piano) is still seen as weak: pieces can wander without coherent themes.
UX, features, and performance
- 1,200 free songs/month is viewed as generous.
- During launch, the service is heavily overloaded: very slow generations, frequent errors, and mismatched songs.
- Requested features:
- Upload a beat and have vocals/rap added.
- Stem export, instrumentals-only, more mix control, better “extend” loudness normalization.
- Better remix controls, clearer UI, favorites, separating “good” from “meh” outputs.
- API access and a “reverse prompt” / interrogation feature.
- Some praise the interface; others find tags warped and prompt box too small.
Training data, legality, and watermarking
- Multiple commenters ask what music was used for training and whether it was licensed.
- One reports obviously mimicked voices of famous singers (including a living artist) in another language and questions permissions.
- Concerns are raised about TOS clauses like mandatory arbitration and class-action waiver.
- Some want mandatory watermarking of AI-generated music; others question the need.
Impact on musicians and creative work
- Strong anxiety and anger from composers and media musicians who fear displacement and devaluation of human music, especially for cheaper, “good enough” commercial work.
- Others argue AI will mainly hit low-end, stock/utility music, likening it to past tech shifts (sampling, digital tools) that lowered entry barriers but didn’t kill high-end art.
- One view: listeners mostly care whether music sounds good, not how it was made; another stresses that flooding markets with competent but uninspired AI work could bury genuinely original art.
Text-to-music UX and future tools
- Debate over whether text prompts are a good long-term UX or mostly a novelty/tech demo.
- Critics say the process feels one-shot and non-collaborative; they want musician-centric, iterative interfaces.
- Others envision future tools where users hum/tap a beat, then get orchestrated arrangements with more fine-grained control.
Miscellaneous
- Some regions (e.g., Australia) report the service blocked.
- A few note the pace of AI audio progress, citing Suno Chirp and ex–DeepMind founders as context.