How the AI music detector works
- Upload a song or a video: MP3, WAV, MP4, MOV and other common formats, up to 200 MB.
- Your browser cuts 30 seconds from the middle of the file and turns it into a small mono clip. The rest of the file never leaves your device.
- The clip goes to ACRCloud's AI music detection model, which looks for the patterns that music generators leave in the audio.
- In about 10 to 30 seconds you get an AI probability from 0 to 100% and, when the music looks generated, the tool it most likely came from.
Reading the result
The model gives one number: the chance that the clip was made by an AI music generator. ACRCloud draws its own line at 50%. We add a margin on both sides so a borderline score isn't presented as a firm answer:
- Likely AI-generated: 70% or more.
- Too close to call: 30% to 70%. Try a different stretch of the song, ideally one where nobody is talking over the music.
- Likely human-made: under 30%.
When a clip looks AI-generated, the result also shows the model's estimate for each generator, so you can see whether it points clearly at one tool or is split between several.
Which generators it knows
In October 2026 the model returned a probability for ten generators: Suno, Udio, Mureka, ElevenLabs, Lyria, Boomy, Sonauto, Riffusion, Seed-Music and MiniMax. Music from a newer or less common generator may still be flagged as AI, but the generator it names is less reliable in that case.
What we tested
We ran four 30-second clips from official store previews through the same model. Two were songs from acts reported to be AI-generated: The Velvet Sundown's "Dust on the Wind" came back at 98.8% and Breaking Rust's "Walk My Walk" at 86.9%, both attributed to Suno. Two were well-known human recordings: "Despacito" came back at 0.3% and "Billie Jean" at 0.7%. Four clips is far too few to measure accuracy, but it shows the scale of the numbers you can expect for clear cases.
When it can be wrong
- Heavily edited audio. Remixes, sped-up or slowed edits and mash-ups change the audio the model looks at.
- Voices or noise over the music. A video with talking, crowd noise or sound effects on top gives the model less of the music itself.
- Partly AI songs. A track can mix a generated backing with real vocals, or the other way round. One score for a 30-second clip can't tell you which parts were generated.
- Very new generators. A model trained on today's generators may miss what tomorrow's produce.
Before you use the song
Whether music is AI-generated and whether you're allowed to use it are separate questions. An AI-generated track can still belong to someone, and a human-made one usually does. To find out what a song is and who owns the recording, run the copyright detector on the same file, or find a song from a video.