Making a Suno AI Song: Prompts, Rights and API Code (2026)

A Suno AI song is a complete track, usually with sung vocals, lyrics and a full arrangement, that Suno’s model generates from a short text description or from lyrics you write yourself. You describe the genre, mood and instruments, and in under a minute you get two finished takes. This guide explains how the model works, how to write prompts that give you usable results, what you’re allowed to do with the output, why Suno has been in court, and how to generate tracks from your own code.

What is a Suno AI song?

Suno is a generative AI music platform made by Suno Inc, a company based in Cambridge, Massachusetts. It launched publicly in December 2023. “Generative” means the model creates new audio instead of putting together pre-recorded loops. You give it text, and it returns a waveform that contains vocals, instruments, mixing and song structure all at once.

A typical Suno AI song has recognisable sections: intro, verse, chorus, bridge and outro. It sounds like one finished recording, not a bare melody sketch. The model has gone through several versions. The ones you’ll see most often are V4, V4.5 and V5, and newer versions generally sound cleaner and follow instructions more closely.

You can use it in a few ways:

• The web app and mobile apps on iOS and Android. This is where most people make their first song.
• Suno Studio, an editing environment for paid users. You can rearrange sections, regenerate parts and work with stems. Stems are separate audio tracks for each element, such as vocals, drums and bass, so you can change one without touching the others.
• An API, for developers who want to generate music inside their own product. That’s covered later in this article.

Every generation gives you two takes of the same idea. Expect the two to differ a lot in melody, vocal delivery and arrangement. That’s on purpose: generating is cheap, so the workflow is to make several takes, pick the best, and keep refining it.

How to create a Suno AI song step by step

The Suno AI music generator has two input styles. Simple mode takes a single sentence like “a sad country song about a dog who waits by the door”, and the model writes the lyrics and chooses the style. Custom mode gives you separate fields for lyrics, style and title. Use custom mode as soon as you care about the result.

1. Write the style line. This short field covers genre, era, tempo, instruments and vocal type. For example: “indie folk, 90 bpm, fingerpicked acoustic guitar, brushed drums, warm female vocal, intimate”. BPM means beats per minute, which is the tempo.
2. Add lyrics with section tags. Put tags in square brackets, such as [Verse], [Chorus] and [Bridge], so the model knows where the song changes. Without tags, it guesses, and the chorus often lands in the wrong place.
3. Give it a title. The title is mostly for your library, but it can nudge the mood a little.
4. Generate and compare both takes. Listen to the first 30 seconds of each. If neither one gets the vocal right, change the style line before you change the lyrics.
5. Extend, edit or remix the keeper. You can lengthen a track, replace a section or make a cover in a different style. Paid users can do finer edits in Suno Studio.
6. Download it. Suno offers audio downloads and a lyric video format. Which formats you can download, such as MP3 versus higher-quality WAV, depends on your plan.

Most people take three to five rounds to get a Suno AI song they want to keep. The first round tells you what the model heard in your prompt. The later rounds fix what it got wrong.

Writing prompts and lyrics that work

Most bad results come from vague prompts. The model fills any gaps with the most common choices for that genre, which is why “pop song about summer” sounds like a thousand other songs. The fix is to be specific in the places that matter.

The style line

Think of the style line as a producer’s brief. It works best as short, comma-separated descriptors, not full sentences.

Element	Weak	Stronger
Genre	rock	garage rock, early 2000s
Tempo	fast	140 bpm, driving
Instruments	guitars	fuzz guitar, live drums, Hammond organ
Vocal	singer	raspy male vocal, shouted gang chorus
Mood	good vibes	defiant, sweaty, joyful

Don’t name real artists. Suno filters artist names in prompts, and even when a name gets through, describing the sound you want gives you more control than a reference.

The lyrics field

Structure matters more than polish. Here’s a lyric block that the model will follow reliably:

[Intro]

[Verse 1]
Porch light humming, half past nine
Your old coat still on the line

[Chorus]
Leave the door open, leave the door open
I'll be home before the morning

[Verse 2]
Kettle cold and the radio low
Counting trains that never slow

[Chorus]
Leave the door open, leave the door open
I'll be home before the morning

[Outro]

A few rules make a big difference. Keep verse lines about the same length so the melody can repeat. Write the chorus the same way each time it appears, because repetition tells the model where the hook is. Leave an [Intro] or [Outro] tag empty when you want an instrumental part. If you’d rather not write lyrics at all, generate a draft with a lyrics tool and then edit it by hand. A Suno AI song with lyrics you’ve revised almost always sounds less generic than one where the model wrote every word.

Is Suno free, and can you sell the music?

Suno has a free tier with daily credits. Suno advertises it as up to 10 songs a day with no credit card. Paid subscriptions add more credits, newer model versions, Suno Studio and, most importantly, commercial rights.

For most people, the rights question matters more than the price:

• Free tier: songs are for personal, non-commercial use. You can listen to them and share them for fun, but you can’t monetise them on streaming services or use them in paid client work.
• Paid plans: Suno’s terms grant commercial use for songs made while you’re subscribed. Upgrading later doesn’t apply this to songs you made on the free tier.

Copyright is a separate question from the licence. In the US, copyright protects human authorship, so a Suno AI song generated entirely by the model may get little or no copyright protection, even if Suno lets you sell it. Writing the lyrics yourself, arranging and editing the track, and adding your own recorded parts all strengthen your claim to the finished work. Distributors and streaming platforms also have their own AI disclosure rules. Read them before you upload.

Terms in this area have changed more than once, so check Suno’s current terms on the day you release, not a blog post from last year. That includes this one.

Why is Suno being sued?

In 2024, major record labels sued Suno for copyright infringement. Their argument was that you can’t build a model that produces convincing commercial-style music without training it on huge amounts of copyrighted recordings, and Suno didn’t license them. Suno’s position was that training on publicly available music is fair use, the US legal doctrine that allows some unlicensed uses of copyrighted work.

Since then, the situation has partly shifted from court cases to deals. Warner Music Group settled with Suno in late 2025 and announced a partnership based on licensed models. Other rights holders, including collecting societies in Europe, have brought their own cases. The overall direction is toward licensed training data and tighter rules on downloads and usage, and less toward shutting the product down.

For an ordinary user, the practical points are:

• The lawsuits are about how the model was trained, not about people who use it.
• Settlements can change the product. Download limits, plan features and model versions have all been adjusted after these deals.
• If you’re building a business on AI music, plan for the terms to keep changing, and keep records of which plan and model produced each track.

How to generate a Suno AI song with an API

If you want to make music inside an app, a game, a video tool or a batch workflow, the web app won’t work. You need an API. Unicorn API offers Suno’s models next to image, video and chat models, with one API key and one bill. You pay in credits: monthly plans give you credits at $0.01 each, and top-ups cost $0.015 per credit and never expire. See pricing for the plans.

Which Suno models are available

Model	What you send	Price
Ai Music API Sounds	A text prompt and model version (for example V5), with optional key, tempo, loop and lyric fields	2.5 credits per track
Ai Music API Mashup	Uploaded audio URLs plus style, title, lyrics, vocal gender and weighting controls	12 credits per track
Ai Music API Generate Lyrics	A prompt describing the song	0.4 credits
Ai Music API Boost Music Style	A short style description to expand	0.4 credits
Ai Music API Generate Persona	A name, description and the task and audio IDs of an earlier track	0 credits
Ai Music API Generate Music Cover	The task ID of an earlier music job	0 credits per image

Prices are per output and rounded up to whole credits. To see the exact cost of a request before you run it, call POST /v1/models/price. The API documentation lists the accepted values for each input field.

Your first track from code

Create a key on the API keys page and export it as UNICORN_API_KEY. Then submit a job:

curl -X POST https://api.unicornapi.net/v1/jobs \
  -H "Authorization: Bearer $UNICORN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "ai-music-api-sounds",
    "input": {
      "model": "V5",
      "prompt": "Indie folk, 90 bpm, fingerpicked acoustic guitar, brushed drums, warm and intimate"
    }
  }'

Music jobs are asynchronous. The response includes a job id, and you poll until the status is succeeded. Here’s a complete Python script that submits the job, waits for it, and saves the audio:

import os
import time
import requests

API = "https://api.unicornapi.net"
HEADERS = {"Authorization": f"Bearer {os.environ['UNICORN_API_KEY']}"}


def run_job(model, input_data, timeout=600):
    resp = requests.post(
        f"{API}/v1/jobs",
        headers=HEADERS,
        json={"model": model, "input": input_data},
    )
    resp.raise_for_status()
    job = resp.json()

    deadline = time.time() + timeout
    while job["status"] != "succeeded":
        if job["status"] == "failed":
            raise RuntimeError(f"Job {job['id']} failed: {job}")
        if time.time() > deadline:
            raise TimeoutError(f"Job {job['id']} still {job['status']}")
        time.sleep(5)
        job = requests.get(f"{API}/v1/jobs/{job['id']}", headers=HEADERS).json()
    return job


job = run_job("ai-music-api-sounds", {
    "model": "V5",
    "prompt": "Indie folk, 90 bpm, fingerpicked acoustic guitar, brushed drums, warm and intimate",
})

audio_url = job["outputs"][0]["url"]
with open("track.mp3", "wb") as f:
    f.write(requests.get(audio_url).content)

print("Saved", audio_url)

Once you have run_job, other models just need a different model and input. To draft lyrics first, call it with "ai-music-api-generate-lyrics" and {"prompt": "a country song about a dog waiting by the door"}. To make a mashup, pass style, title, prompt and an upload_url_list of source audio to "ai-music-api-mashup".

Worked example: a song with cover art and a music video

Say you’re building a feature that turns a user’s idea into a song they can share. A sensible pipeline is:

1. Generate draft lyrics (0.4 credits) and expand the user’s style description with Boost Music Style (0.4 credits).
2. Generate two Sounds takes so the user can choose one (2.5 credits each, 5 in total).
3. Make cover art with Generate Music Cover from the chosen job’s task ID (0 credits). If you want more control over the artwork, use Nano Banana 2 Lite at 4 credits per image or GPT Image 1.5, which handles text such as song titles well, at 3 to 22 credits depending on quality.
4. Animate the cover into a 10-second, 720p clip with Grok Imagine Video 1.5 (45 credits). If you need clips up to 30 seconds, Grok Imagine Video 1.0 handles longer durations at the same per-second rate.

The video step uses the same job endpoint:

curl -X POST https://api.unicornapi.net/v1/jobs \
  -H "Authorization: Bearer $UNICORN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "grok-imagine-video-1-5-preview",
    "input": {
      "image": "https://example.com/cover.png",
      "prompt": "Slow push-in on the album cover, porch light flickering, gentle dust in the air",
      "duration": 10,
      "resolution": "720p"
    }
  }'

At listed rates, the whole package comes to about 51 credits before rounding, or roughly 50 cents on a monthly plan. Almost all of that is the video. If you only need audio, the cost per Suno AI song is a few credits. For more on the video side, see the Grok AI video generator guide. If you’re also adding sound design, the AI sound effects generator guide covers that.

If you write code with an agent, you can also call these models from Cursor or Claude Code over MCP. See agents.

Common mistakes and trade-offs

These are the problems that cost people the most time and credits, whether they use the app or the API.

• Changing lyrics and style at the same time. If a take sounds wrong, change one thing per round. Otherwise you can’t tell which change fixed it, or which one broke it.
• Writing the style line as a paragraph. Long sentences dilute the signal. Five to eight short descriptors work better than a story about the vibe.
• Leaving out section tags. Untagged lyrics produce choruses in odd places and verses that run into each other.
• Judging from one generation. The two takes from a single request can be very different. Make at least two rounds before you decide a prompt doesn’t work.
• Assuming free-tier songs can be monetised. They can’t, and upgrading later doesn’t change that. Decide whether a project is commercial before you generate it.
• Ignoring copyright limits. A commercial licence from Suno is not the same as owning a copyright. If owning the work matters, add real human authorship to it.
• Polling too fast through the API. Music jobs take a while. Polling every few seconds, with a timeout, is enough and keeps your logs readable.

Choosing between the app and the API is mostly about volume and control. The app is better for writing one song you care about: Suno Studio’s editing tools and the immediate listen-and-tweak loop are hard to beat. The API is better when songs are part of something bigger, such as personalised tracks for users, background music for generated videos, or batch production, where you want predictable per-track prices and the same key for your image, video and chat models. Many teams use both: they find a good prompt in the app, then reuse it in code.

Whichever you choose, the process is the same: write a specific brief, add structure, generate several takes, pick one and refine it. If you start with a clear style line and tagged lyrics, your next Suno AI song will need far fewer rounds than your first. When you’re ready to automate it, browse the model catalog and run your first job.