How to Make AI Songs?

Why Everyone's Making AI Songs Now (And What's Stopping You)

Six months ago, making a song required instruments, studio time, or at minimum a decade of piano lessons. Today, you type a sentence, and a full track with vocals, lyrics, and instrumentation comes back in under a minute. That shift is no longer experimental — it is a paid, mainstream behavior.

The numbers tell the story. Text-to-song with vocals now accounts for 62.1% of 2025 AI music generation revenue, according to market analysis, and Suno — the category leader — has reached 2 million paid subscribers and $300 million in annual recurring revenue. Over 100 million people have tried Suno, yet only 2 million pay, which reveals something important: curiosity is nearly universal, but knowing how to use these tools well is rare.

So what's actually stopping most professionals from crossing that gap? Usually three things:

• Tool confusion — Suno, Udio, Mureka, Eleven Music, and Google's Lyria 3 all generate songs from text, but they differ significantly in pricing, quality, and commercial rights.
• Legal uncertainty — the U.S. Copyright Office has made clear that works created by AI without human involvement cannot be copyrighted, leaving creators unsure what they actually own.
• Use-case mismatch — the tool that works for a novelty birthday song is not the tool for rights-cleared background music in a company video.

The market itself is settling these questions fast. Major labels have moved from lawsuits to licensing deals, with Universal Music Group settling with Udio and Warner settling with Suno in late 2025, according to industry reporting. Platforms like ElevenLabs now describe their outputs as cleared for commercial use under Merlin and Kobalt licences. The generative AI in music market is projected to grow from $440 million in 2023 to nearly $2.8 billion by 2030, per Grand View Research.

Trapital's analysis argues that production music and the musician creation stack are the most durable near-term markets — which matters if you're a business owner thinking about background tracks for content, ads, or meditation apps rather than chasing viral streams. The Velvet Sundown, an AI band that hit 1.4 million Spotify listeners before falling to roughly 141,000, shows how volatile the streaming-upload route can be.

The honest answer to "what's stopping you" is that the tools are ready but the judgment isn't automatic yet. Knowing which platform clears commercial rights, how to add enough human authorship to protect your work, and how to wire music generation into a broader content workflow separates dabblers from people getting real value. Agent skills can automate the rest of that workflow — SkillzBench is building a free directory where you can discover, verify, and benchmark skills before you trust them with your creative pipeline, and you can join the waitlist to be first in line.

The gap between "I could make a song" and making one well — and safely — has never been smaller. It just requires knowing the terrain, which is exactly what the rest of this guide covers.

The Text-to-Song Workflow: Platforms and Prompts That Work

Making an AI song in 2026 is refreshingly simple: you type what you want to hear, and a platform delivers vocals, instrumentation, and lyrics in return. That workflow — text-to-song with vocals — now accounts for 62.1% of AI music generation revenue, making it the clear mainstream path for creators, according to market research.

Suno leads the pack, with 2 million paid subscribers and roughly $300M in annual recurring revenue — over 100 million people have tried it. Its free tier makes experimentation cost-free, while Pro runs $10/month (or $8/month billed annually). Alternatives include Udio, Mureka, Eleven Music (cleared for commercial use via Merlin/Kobalt licences), and Google's Lyria 3 inside the Gemini app.

A good prompt specifies genre, mood, tempo, instrumentation, and vocal style. "An upbeat 1980s synth-pop track with female vocals and a driving chorus" beats "a pop song" every time. Iterate: regenerate, tweak the prompt, and refine the sections that work.

The tool matters less than matching it to your use case. Trapital's analysis identifies five segments, judging production music and the musician creation stack as the strongest near-term markets — "far more durable" than novelty tracks or direct streaming uploads, whose volatility is visible in The Velvet Sundown's fall from 1.4 million to roughly 141,000 Spotify monthly listeners.

• Production music — Amper Music for rights-cleared background tracks, Endel for adaptive soundscapes, Stable Audio 3.0 for tracks up to six minutes on licensed data.
• Monetizable streaming — Boomy lets creators publish and earn royalties on streaming platforms.
• Compositional control — AIVA, recognized as an official composer by France's SACEM, generates themes you then elaborate and orchestrate.
• Full songs fast — Suno, Udio, Mureka, Eleven Music, or Lyria 3 for complete vocal tracks from a single prompt.

One caution from Trapital's Dan Runcie: early impressions of AI music quality are outdated — the trajectory mirrors ChatGPT's leap from 2023 to 2026. If you're building agentic workflows around content generation, agent skills can automate the rest of this pipeline — browse and test free ones on SkillzBench (skillzbench.com), which benchmarks each skill's Lift in a sandbox so you know what actually helps before you commit.

The Legal Playbook: Copyright, Commercial Use and Voice Rules

The viral "Heart on My Sleeve" fake-Drake track racked up over 9 million TikTok views before Universal Music Group forced it down — a wake-up call that AI songmaking has real legal edges. Before you monetize anything, you need to understand three constraints that separate safe creators from defendants.

First, copyright requires a human touch. Under U.S. Copyright Office guidance, "works that are created by AI without human intervention or human involvement cannot be copyrighted," according to Harvard Law analysis of the issue. Works containing AI-generated material may still qualify where there's sufficient human authorship — for example, creatively selecting or arranging the AI output. If you want enforceable rights in your song, write your own lyrics, edit the arrangement, and document your creative decisions.

Second, commercial-use clearance varies sharply by platform. ElevenLabs launched Eleven Music in August 2025 with Merlin/Kobalt licences and describes its outputs as "cleared for commercial use," per market research on AI music generation. Other platforms can't make the same claim — the industry only shifted from litigation to licensing in late 2025, when UMG settled with Udio and Warner settled with Suno. Check each platform's terms before you upload a track to Spotify or license it to a client.

Third, voices are legally dangerous even when styles are not. Louis Tompros, an IP lecturer at Harvard Law, explains that style imitation alone is generally not infringement — "music in the style of someone else is not considered a derivative work." Voice imitation is different: the strongest legal argument against voice-cloned songs is right of publicity, rooted in the Bette Midler v. Ford precedent from 1988. Notably, the "Heart on My Sleeve" takedown may have stemmed from direct copying of Metro Boomin's producer tag, not the Drake-style mimicry itself.

Watch for these red flags in platform terms of service:

• Broad training licenses over anything you upload — some ToS grant platforms "a worldwide license to use, reproduce, modify, adapt, and process your music" for AI training, per the Incorporated Society of Musicians
• Ambiguous commercial-use language, where monetization rights depend on your subscription tier
• Voice-model features that let you imitate identifiable artists — a right-of-publicity minefield
• Missing disclosure requirements, such as SynthID watermarking, which regulators increasingly expect

The stakes are growing with the market: generative AI in music is forecast to expand from roughly USD 440 million in 2023 to nearly USD 2.8 billion by 2030, per Grand View Research. More money means more enforcement. As analyst Mark Mulligan puts it, "Copyright law was not designed for AI" — so build human authorship and clearance checks into your workflow from the start rather than retrofitting them after a takedown notice.

If you're building agentic workflows around music or content generation, agent skills can automate parts of this compliance work — SkillzBench (skillzbench.com) is building a free, verified directory of skills and a free MCP connector so you can test them with confidence.

From One Song to a Workflow: Practical Steps and Automation

Making your first AI song takes minutes — turning that into a repeatable, legally safe workflow is what separates a one-off novelty from a durable creative practice. With full songs with vocals accounting for 62.1% of 2025 AI music revenue, the tools are proven; the process around them is what you build yourself.

Start with authorship. Under the U.S. Copyright Office's March 15 guidance, works created by AI alone cannot be copyrighted, but songs with sufficient human authorship — such as writing your own lyrics or creatively selecting and arranging AI-generated material — may be protectable. So begin every project by writing your lyrics before you touch a prompt, and keep notes on which musical selections and arrangement choices you made. That documentation is your evidence of human contribution.

Next, handle clearance and voice risk before you monetize anything. Commercial-use terms vary by platform — ElevenLabs, for example, describes its Eleven Music outputs as cleared for commercial use under Merlin and Kobalt licences, while other platforms require paid tiers or make no such claim. Read the terms of service closely, since some now claim broad licences over uploaded music for AI training. And never imitate an identifiable artist's voice: style imitation is generally not infringement, but commercial voice imitation can violate right of publicity law under the Bette Midler v. Ford precedent — the viral fake-Drake track gained over 9 million TikTok views before being pulled.

Once one song is done safely, scale the pattern into a checklist you run every time:

• Write and save original lyrics first, timestamped, to establish human authorship.
• Log your creative selections — genre, structure, arrangement edits — in a project file.
• Verify the platform's commercial-clearance terms match your monetization plans.
• Screen prompts against imitating named artists' voices or signature sounds.
• Organize finished files, metadata, and documentation in a consistent folder structure.

Here's where automation earns its keep. Agent skills — instruction packages your AI agent reads — can handle the mechanical half of this workflow: generating structured prompt variations from your lyrics, naming and filing each song version, drafting publishing descriptions, and running your clearance checklist against new platforms. Rather than hand-picking skills and hoping they work, browse and test free skills on SkillzBench's free directory — SkillzBench is building version-pinned, security-scanned skills plus a free MCP connector that links any agent to the skills you choose, with everything free to use.

The market rewards this discipline. Suno alone reports 2 million paid subscribers and $300M ARR, but only a small fraction of its 100 million users pay — the winners treat AI songs as output from a real process, not a slot machine. Build the workflow once, and every song after the first gets easier.