Generate Music from Text with AI

Enter a text prompt describing the music you want — genre, mood, tempo, instruments, and optionally full lyrics — and the AI writes a complete song around it, then returns two distinct variants so you have a choice on the very first run. You are describing a target rather than assembling parts by hand: say "upbeat indie-pop, jangly guitars, female lead, hopeful" and the model handles the arrangement, the melody, and the vocal delivery together. In Custom Mode you can go further and supply exact lyrics and precise style tags when you want the output to land a specific way instead of letting the AI interpret a loose brief.

Simple Mode takes one short description and lets the AI decide everything — it writes the lyrics, picks the arrangement, and chooses how the vocal sits. It is the fastest way to hear an idea. Custom Mode hands you the controls: you write the exact lyrics line by line and set comma-separated style tags, which Simple Mode does not send at all. The vocal gender row sits in Advanced and works in either mode. Use Simple Mode to explore and Custom Mode to nail down a song you already hear in your head — for example when you need a specific chorus hook or a particular instrument to lead.

Yes. Set the song mode to Instrumental (no singing) and the AI produces music with no sung vocal — just the arrangement. This is the right setting for background beds under a video edit, lo-fi study loops, game and app music, podcast intros, and beats you plan to write over later. Because instrumentals skip the vocal pass, they also tend to give you a cleaner canvas: you can generate an instrumental here and, if you decide it needs a topline afterwards, upload it to Add Vocals to have an AI singer perform lyrics over it.

Every generation returns two variants of your prompt rather than one, so a single 20-credit run gives you two takes on the same idea to compare. The two are genuine reinterpretations — different melodies, phrasing, and sometimes arrangement choices — not the same song twice, which means you often find that one of the pair already captures the feel you were after. If neither is quite right, you tweak the prompt or a style tag and run again; treating the prompt as something you nudge between runs gets you to the keeper faster than rewriting it from scratch each time.

Generate Music supports the full range of Music Factory engines, from the classic generations through the latest. The table below is a practical guide to choosing. In general the newest engine is the best default — it delivers the clearest vocals and the most coherent arrangements — while older engines can be worth trying when you want a rougher or more vintage character. Engine · Best for · Character. v5.5 · Exact song length, your own voice · Newest — the only engine that takes a length and a cloned voice. v5 · Most releases · Cleanest vocals, most coherent arrangement. v4.5+ / v4.5 All · High quality, more variation · Strong vocals, slightly looser feel. v4.5 / v4 · Fast drafts and demos · Solid all-rounders.

Think in layers and be concrete. Lead with genre and mood ("melancholy synthwave"), then name the vocal ("male falsetto lead"), then call out the instruments you expect to hear ("analog bass, gated snare, wide pads"), and finally the tempo or energy ("mid-tempo, building"). Naming real instruments and a clear vocal type gives the model a much sharper target than a single genre word. If the first result is too busy, remove an instrument from the prompt; if it feels thin, add one lead instrument rather than several at once. In Custom Mode you can also paste the exact lyrics, which is the surest way to control what gets sung.

The generator is not tied to one style — it handles pop, rock, EDM, hip-hop, jazz, classical, country, R&B, folk, metal, lo-fi, cinematic, and experimental music, plus blends of them. Because the input is a text description, you can also combine ideas the model has learned into hybrids, for example "trap drums under an orchestral string section" or "bossa nova rhythm with modern synths". The more specific and internally consistent your description is, the more reliably the arrangement holds together; vague or contradictory prompts (fast and slow, aggressive and gentle at once) tend to pull the result in several directions.

Each generation costs 20 credits and returns two track variants, and no subscription is required — you pay only when you generate. Plans start at $19.99 with credits included, and credits are consumed at the moment you start a generation. Because a single run gives you two takes, the effective cost per usable idea is often lower than it looks, since you frequently keep one of the pair. If you want to explore aggressively, generating a few short drafts and only refining the direction you like is more credit-efficient than perfecting one prompt in isolation.

Generate Music runs asynchronously rather than instantly — you start the job, it processes in the background, and both variants land in your Music Factory library when they are ready, typically a couple of minutes later depending on load. You do not need to keep the tab open the whole time; the results are saved so you can come back to preview and download them. That background model is also why you can queue an idea, start writing your next prompt, and come back to a finished pair rather than watching a spinner.

Yes — the generated track becomes the source for the rest of the suite. If it ends too soon for your video, continue it with Extend Track; if one verse or chorus is weak, rewrite just that span with Replace Section instead of regenerating the whole song; and if it has vocals and you want synced words for a lyric video, run it through Timestamped Lyrics. Because every stage stays inside your library, you keep the same key and tempo from the first draft through to the finished master without exporting and re-importing between tools.

Yes. Every track you generate carries a full commercial license, so you can use it in videos, ads, podcasts, games, streams, or paid client work, and you can publish it on YouTube, TikTok, Spotify, and other platforms as original AI-generated content. There are no Content ID claims to clear because the output is not lifted from an existing recording. Credits are consumed when you start the generation, and because outputs vary between runs it is worth listening through both variants before you build a project around one of them.

The common uses fall into a few groups: content backing (music beds for YouTube videos, TikToks, reels, and streams), production and games (soundtrack cues, level music, app and menu themes), and songwriting itself (demoing an idea, roughing out a topline, or exploring an arrangement before you record it for real). Because you get a full song with vocals rather than a loop, it also works for jingles, intros and outros, podcast theme music, and background tracks for presentations. If you need something specific — a 30-second upbeat bed, a moody instrumental for a trailer, a full pop song with a chorus — describing that target in the prompt gets you closer than a generic genre request, and the two variants per run give you a fallback if the first read is not quite right.

A prompt is a target, not a guarantee, and a few things commonly cause drift. Contradictory instructions pull the model in two directions — asking for "fast and relaxed" or "aggressive and gentle" forces a compromise. Very vague prompts leave too much open, so the model fills gaps its own way; adding concrete instruments and a clear vocal type narrows it. And because each run is a fresh interpretation, two generations of the same prompt will differ, which is by design so you get variety. If a result missed badly, tighten the prompt one change at a time — pin the vocal, name the lead instrument, or in Custom Mode paste the exact lyrics — rather than rewriting everything at once, so you can see which change fixed it.

Switch to Custom Mode and write the lyrics yourself, marking the sections you care about. Writing the actual chorus lines is the single most reliable way to control what gets sung, because the model performs the words you give it rather than inventing its own. Keep chorus lines short and repeatable — a hook is memorable partly because it is simple — and set style tags that describe how the chorus should lift ("big, open, layered harmonies") versus how the verses should sit ("restrained, intimate"). If the chorus still does not land after a couple of runs, that is exactly the case Replace Section was built for: keep the parts you like and regenerate just the chorus window.

A loop library gives you fixed clips you arrange and cut yourself, and every user has access to the same clips, so tracks built from popular packs can sound familiar. Generating from text writes an original song around your description instead — the melody, arrangement, and vocal are produced for your prompt, so the result is unique to that run and comes with a full commercial license rather than sample-clearance conditions. The trade-off is control style: with loops you assemble known pieces by hand; with generation you steer through a prompt and pick between variants, then refine specific parts with the rest of the suite. For original songs, custom lyrics, or bespoke content where a stock loop would not fit, generation is usually the faster path to something that is genuinely yours.