Suno

A whole song with vocals in one pass, enough to judge direction before production work.

Model profile

Vendor

Suno

Form

网页服务

Input

文本 · 歌词 · 参考音频

Output

音频 · 分轨

Licence

商用范围、发行权限与训练数据来源需按当前订阅条款核对

Verified

2026-10-30

Related

TimbreStemsContemporary

What it lets you finish

Give it a style line and some lyrics and it returns a complete song with vocals, verses and a chorus. It is for getting a listenable version first: fixing direction with a client, temp-scoring a video, or hearing a reference for your own arrangement. You can also ask for instrumental only, or use one line to probe a style quickly. It handles zero-to-listenable, not listenable-to-releasable.

Inputs and outputs

Input is text: style and mood, lyrics, optional section tags for verse, chorus and bridge; some tiers accept reference audio to narrow timbre and groove. Output is usually stereo audio, with stems and vocals exportable on some tiers. One prompt returns several candidates and each run differs; what you can archive and reproduce is the prompt and seed, not the audio itself. The listening examples and instances below are references for this kind of output, not this model's generations.

Strengths

• Vocals and backing arrive together, saving the time of hiring a singer, arranging, and cutting a demo.
• Section tags do work: write verse and chorus and the structure usually follows, with the chorus lifting.
• Broad style range: folk, electronic, orchestral and rock all get a plausible surface.
• Changing one lyric line and regenerating lets you compare phrasing quickly, which suits lyric editing.

Weaknesses and failure modes

• Past three minutes the song tends to return to earlier harmony and orchestration. A reproducible test: generate twice from one prompt and compare the last thirty seconds; the sections are nearly identical.
• Diction blurs on fast lines, long words and Mandarin tones, worst in stacked chorus vocals. Check line by line against the lyric, not just the waveform.
• Two runs from the same prompt give different vocal timbres, so stitching across runs sounds like a change of singer.
• Stems are not clean: the vocal stem often carries reverb tails and cymbal bleed, and needs repair before remixing.
• Exact tempo, key and chord progression are not honoured; "125 BPM, D minor" comes back approximate, so anything precise belongs in MIDI.
• Reference audio can fix timbre broadly but rarely groove and instrumentation at once: the tone lands, the rhythm does not.

Terms, price and alternatives

Confirm: the commercial scope of your current tier, whether streaming release is allowed, what export rights you have for stems and MIDI, and the official statement on training data. Pricing is tiered; exact figures and regional availability belong on the official page, and this site does not fix them. Alternatives: Udio if you need a more recorded-sounding voice; Stable Audio if you only need instrumental beds and want to avoid vocal-related terms. Both may cost you whole-song structural coherence, so choose by the job at hand.

Sources

• 产品能力与导出形态，依官方公开说明与条款页核对，未抄原文
• 失败模式依公开用户报告与创作实践归纳，未引用具体评测数据

Used for fact-checking only; all prose is original.