Text to music, powered by MiniMax-Music3
Type a description of a piece of music and this tool composes it: a finished mix with instruments, arrangement, and, if you want them, sung vocals. It runs MiniMax-Music3, an open music generation model, on free shared GPUs, and it hands back a 44.1 kHz stereo WAV you can download immediately. There is no sign-up, no email, no watermark, and no spoken audio tag stamped over the result.
Generating music is a different job from the image and video models elsewhere in this collection. Nothing is being stitched together from a sample library; the model writes the audio itself, which is why a description naming a genre, an instrument, and a tempo gives you far more control than a one word request. If you want moving pictures to go with the track, the free AI video generator makes short clips from the same kind of text prompt.
Instrumental beds and songs with vocals
The instrumental switch decides whether anyone sings. Left on, you get an instrumental: background music for a video, a podcast intro, a game loop, a track for a slideshow, a bed under a voiceover. That is the mode most people want, because an instrumental sits underneath narration without fighting it for attention.
Switch it off and the model performs vocals. You can hand it your own words in the lyrics box, one line per line, and it will sing them, or you can leave the box empty and let the model write words that fit the description you gave. Vocal takes are less predictable than instrumentals. Pronunciation wanders on unusual names and invented words, and the phrasing is the model's choice rather than yours, so generating two or three takes and keeping the best one is normal practice.
Most of what people make here is short and functional: a bed under a product demo, a title sting, an intro and outro pair for a podcast, a loop for a game menu, a jingle for a shop video, a piece of scoring for a wedding slideshow. Because a new take costs nothing but the wait, it is worth generating several and choosing by ear rather than trying to nail the description on the first attempt. Save the ones you like as you go; the file lives on the temporary storage of a shared server and will not be there tomorrow.
Writing a description that gets the track you want
The model reads plain English, so write as though you were briefing a musician instead of typing keywords. Four things carry most of the weight: genre, instrumentation, tempo or energy, and mood. A brief like "slow, melancholic acoustic guitar ballad with brushed drums and a warm upright bass" gets you somewhere. "Sad song" does not.
Production language works too. Terms such as lo-fi, vinyl crackle, tape saturation, reverb heavy, sparse, four on the floor, and half time all land, because the model learned from descriptions written the way music is actually talked about. Describing what an artist sounds like is more useful than naming the artist. Every run uses a fresh random seed, so the same description gives a new take each time you press the button, and pressing it again is a legitimate way to steer.
Length, quality, and limits
You can aim for about fifteen or thirty seconds. Treat that as a target rather than a promise: the model shapes a piece toward the length you request and stops where the music stops, so a thirty second request often comes back a little under thirty seconds. The player under your result reports the real length of the file that arrived, not the number you asked for, so you always know what you are working with before you drop it into a timeline.
The download is a 44.1 kHz stereo WAV, CD sample rate and uncompressed, ready for a video editor or a DAW without a conversion step. Runs happen on shared free GPUs, so every visitor gets a daily allowance and a busy period can mean a short wait in the queue; connecting a free Hugging Face account raises that allowance and improves your place in line. Nothing you type is kept: the description and any lyrics travel to the GPU for the length of the run and the generated file expires from temporary storage on its own. If the audio you need is already inside a video you have, the free video to MP3 tool pulls it out in your browser without uploading anything.