Suno — Module 22.4

How This Lesson Fits the Module & Volume

ElevenLabs, Cartesia, and PlayAI speak language. Suno composes: text (or lyrics + style tags) → a full song with vocals, accompaniment, and structure. That is Vol. 16 audio / music generation, not TTS. Do not send a drum loop to Whisper and call it a transcript; do not use ElevenLabs to “sing a hit” and expect Suno-class arrangement.

Peer: Udio. Legal/product risk: Vol. 20 copyright (training data, output licensing, label disputes—educational, not legal advice). If you ship Suno inside an app, wrap HTTP behind Vol. 18 FastAPI like any other vendor.

Learning Objectives

By the end of this lesson, students should be able to:

Define Suno as a generative music product/API (songs with vocals), distinct from TTS vendors.
Contrast Suno vs Udio vs “TTS singing” vs traditional DAW + sample libraries.
Read qualitative pricing/strengths/weaknesses without inventing credit prices or chart positions.
Sketch a job-based generate → poll → download flow behind FastAPI.
List copyright, ToS, and brand-safety gates before using generated music commercially.
Know when a licensed library or human composer is the correct buy instead.

Definition

Suno is a generative music company and consumer product: users (and, where offered, API clients) prompt style, lyrics, or a simple description and receive complete songs—vocals + instrumental—typically as downloadable audio. Custom/advanced modes let you supply lyrics and genre tags. Treat model generations (v3, v4, …) as SKUs. Suno is not an STT engine, not a voice-clone IVR, and not a DAW.

Music Is Not Speech

Vol. 16.1 audio separated music, events, and speech. Suno lives on the music side: melody, harmony, lyrics, mix. Eval is listening tests + brand safety + license review—not WER. Latency is job-based (seconds to tens of seconds), not Cartesia TTFB. Cost is credits/subscriptions, not TTS characters.

Pricing / Strengths / Weaknesses (Qualitative)

Dimension	Suno	Udio	TTS “singing” (ElevenLabs etc.)	Licensed libraries / human
Pricing posture	Consumer plans + generation credits; API if offered is usage/credit—confirm live sheet and commercial ToS	Similar consumer credit/plan model	TTS characters; you still need arrangement	Sync licenses, work-for-hire, stock subscriptions
Strengths	Full songs with vocals from text; fast ideation; strong consumer UX; custom lyrics mode	Competitive song quality; vocal reputation in bake-offs (eval yourself)	Control over a known speaker identity	Clearer rights; professional mix; no model ToS surprise
Weaknesses	Copyright/training disputes with labels (follow counsel); output rights vary by plan; job latency; less DAW control; API maturity can lag the consumer app	Same legal/ToS class of risk; API often thinner than Suno	Not a band; no automatic drums/harmony	Slower; higher $ for original scores

Product Shapes

Simple prompt

“Upbeat indie song about shipping”
Good for moodboards
Weak for on-brand lyrics

Custom lyrics + style

You write words; model sings
Better for ads/courses
Still check ToS for commercial use

In-app API job

Create → poll → store URL
Idempotent job ids
Never stream keys to the browser

Reasonable uses

Internal scratch tracks
Placeholder scores in prototypes
User-generated music if ToS allows
Education demos with disclosure

Stop and call counsel

Shipping a “sounds like Artist X” button
Claiming you own training-data rights
Ignoring label litigation headlines
Using outputs in a film without reading the plan license

Python: Job-Shaped Generate (Conceptual)

Suno’s public API surface has changed over time (consumer app first). Do not copy unofficial Discord bots. Official HTTP: authenticate, create a generation job, poll until audio URLs exist, persist under your retention policy. Confirm current base URL, auth header, and commercial terms before production.

# suno_job.py — conceptual; pin live OpenAPI paths before shipping
import os, time, httpx

BASE = os.environ.get("SUNO_API_BASE", "https://api.suno.ai")  # verify
KEY = os.environ["SUNO_API_KEY"]

def generate_song(prompt: str, lyrics: str | None = None) -> dict:
    r = httpx.post(
        f"{BASE}/v1/generations",  # path may differ — read current docs
        headers={"Authorization": f"Bearer {KEY}"},
        json={"prompt": prompt, "lyrics": lyrics, "make_instrumental": False},
        timeout=30,
    )
    r.raise_for_status()
    job_id = r.json()["id"]
    for _ in range(60):
        s = httpx.get(f"{BASE}/v1/generations/{job_id}", headers={"Authorization": f"Bearer {KEY}"})
        s.raise_for_status()
        body = s.json()
        if body.get("status") in {"complete", "succeeded"}:
            return body  # audio URLs — download server-side, not via user browser+key
        if body.get("status") in {"failed", "error"}:
            raise RuntimeError(body)
        time.sleep(5)
    raise TimeoutError(job_id)

Related Lectures

Lecture	Role
Audio / speech / TTS	Modality split
Udio	Music peer
Copyright / compliance	Rights & ToS
FastAPI / authentication	Job API wrap
ElevenLabs	Not a substitute for songs

Common Misconception

“Suno is TTS with a beat.” Arrangement and vocals are generated as music, not read SSML. Second: consumer-plan downloads automatically include a film sync license. Third: prompting “in the style of [famous artist]” is a clever workaround—it is a legal and brand risk. Fourth: unofficial bots are the official API. Fifth: WER or BLEU scores a song. Sixth: Vol. 19 perplexity (the metric) measures Suno quality.

Knowledge Check

Short Answer: What does Suno generate that ElevenLabs typically does not? Answer: Full songs (vocals + accompaniment/arrangement), not just spoken TTS.
True/False: Suno is an STT vendor. Answer: False.
Multiple Choice: Commercial use of outputs should: (a) follow current plan ToS + counsel, (b) assume public-domain, (c) use WER. Answer: (a).
Short Answer: Name Suno’s main music peer in this module. Answer: Udio.
True/False: You should invent per-song dollar prices from this lecture. Answer: False.
Multiple Choice: API integration is usually: (a) async job poll + server-side download, (b) browser key + infinite stream, (c) Whisper timestamps. Answer: (a).
Short Answer: Which Vol. 20 lecture covers training-data and output rights literacy? Answer: Copyright.
True/False: “Sounds like Artist X” is a safe production feature. Answer: False — high legal/brand risk; escalate.
Multiple Choice: Eval Suno with: (a) listening tests + license review, (b) token perplexity only, (c) IoU. Answer: (a).
Short Answer: Which Vol. 16 lecture distinguishes music from speech? Answer: Audio (16.1).

Key Takeaways

Suno = text-to-song; not TTS, not STT, not a DAW replacement.
Credits/ToS/API paths change—no fake prices; read commercial terms.
Copyright and “style of artist” prompts are first-class product risks (Vol. 20).
Wrap generations as jobs behind FastAPI; keep keys server-side.
Next: Udio—the other major consumer music model.

Trainer’s Guide

Lab (no scrape, no artist impersonation): Students write lyrics for a 20-second course jingle (original words), generate via official Suno UI or documented API if available, and produce a one-page rights checklist: plan tier, commercial clause, retention, disclosure. Alternative if blocked: architecture-only job poller with mocked HTTP. Grade: no unofficial bots, no “in the style of” celebrity prompts.

Recap: Suno opens the music half of Module 22.4. Rights before vibes. Peer bake-off next: Udio.