Suno V5 AI song generator

You type a description and press one button. What comes back is a finished track: lyrics written, sung, arranged, mixed, with its own cover art, playable in the browser and downloadable as an MP3. Not a stem pack. Not a MIDI sketch. A song.

Suno renders two variants for every request. The studio scores them on whether each came back with audio, a title and a prompt, keeps the more complete one at the front, and that is the one you see. You do not get to choose between them, which is worth knowing before you go looking for the second take.

Finished tracks stay in your library. Publishing one to Explore, the public feed here, is also what makes it playable for anyone who is not signed in as you, so an unpublished track is private in the strong sense: no other browser can stream it at all.

The studio's tabs differ by how much of the writing you want to do. Describe a theme and the model writes the words. Ask for lyrics first, read them, then hand them over. Or paste lyrics you already have, in which case the model sings those exact lines instead of its own reading of a theme.

Writing lyrics is free on every one of those routes, including the standalone AI lyrics generator, which needs no account at all. Signed in, that is 20 sets a day; signed out, the tool draws on a shared daily pool. The 30 credits are charged when a track is generated, and only then.

Batch mode is the fourth route and it is the same model underneath. One brief becomes up to 200 items, each expanded into its own prompt, generated and charged as its own track at 30 credits. A batch is not cheaper per song. What it saves is the writing.

You cannot set how long a track runs, and that is not a missing control. The music request that leaves this server carries a prompt, a model id, the custom-mode flag, the instrumental flag, and in custom mode a title, a style line and a vocal-gender code. There is no duration field anywhere in it, so there is nothing for a slider to pass through.

This has cost us before. A batch template on this site used to ask for tracks of four to five minutes to fill a fifty-minute class. Three came back at 3:23, 3:27 and 2:00. Re-rolling does not close a gap like that, it just spends another 30 credits an attempt, so the template now asks for what the provider can actually honour: tracks long enough to carry a block, with clean in and out points, loopable when a block overruns.

The flip side is that length cannot inflate the bill either. Video here is billed per second, so a longer clip costs strictly more. A song is one number whatever comes back. The credit plans decide how many songs a month buys; the per-song figure itself never moves.

The instrumental flag is a hard boolean on the request and it outranks your prompt. An instrumental submission whose prompt describes a singer comes back sounding like it lost its vocal; a vocal submission whose prompt says instrumental gets invented lyrics. Set the switch to match what you wrote, and keep the words song, chorus and anthem out of an instrumental prompt, because they pull vocals back in even with the flag set.

Asking for a male or female lead does more than it looks like it does. Suno's vocal-gender parameter only takes effect in custom mode, and custom mode needs exact lyrics rather than a theme, so choosing a voice on the description tab makes the system write lyrics for you first and submit those. That extra call is absorbed here rather than billed. Worth knowing anyway, because the words you end up with are the ones that step produced. Duet is not a real parameter at all and goes in as a style instruction.

Language behaves the same way: ten of them plus auto, steering what gets written rather than putting an accent on a finished vocal. All of it sits on one form in the song studio, with auto as the default that changes nothing.

The shape that works is two parts: one line saying what the track is and what it is for, a blank line, then a single Style line of comma-separated descriptors ending in a BPM number. The four below are real prompts from this site rather than illustrations of a format.

What separates a usable track from mush is specificity in a few named places. Name a subgenre rather than an umbrella: neo-soul, lo-fi boom bap, deep house, not pop or electronic. Hold one dominant mood, because stacking melancholic, euphoric and aggressive averages them into nothing. Name three to five instruments with an adjective each, felt piano and brushed snare rather than piano and drums. Give BPM as a number and never contradict it in words elsewhere. Past roughly eight descriptors the model stops weighting the extra ones.

Two hard limits are worth knowing before you write a paragraph. A description-mode prompt over about 500 characters is rejected by the provider before anything generates, and exclusions belong at the end with at most three of them, because the positives are processed first. The studio tour walks through the rest of the form.

People search for a Suno API, so it is worth saying plainly: this site does not have one. No endpoint takes your key, no keys are issued, and nothing here resells access to somebody else's. It is a web app. If an API is genuinely what you need, this is the wrong page, rather than a page that is going to talk you round.

What is wired up instead is most of what an API leaves you to build: a prompt box that already knows the model's shape, a free lyrics writer, billing at a flat 30 credits a song, an automatic refund when a generation fails, a library that keeps every finished track, and a public feed you can publish to. The model directory lists everything else running on the same balance, from video to stills.