Is this song AI? Search, paste a link or upload a file to find out. Learn more.
How it works:
Our AI music detector analyzes songs two ways:
Spectral: we turn the song into an image and compare it against known AI output. This is our most reliable method.
Temporal: we look at tempo, phase, timing and loudness patterns. This one struggles with shorter songs or low-quality audio, so it plays a smaller role in the final output.
When both agree, it's a really strong signal.
What we detect:
Suno, Udio, Riffusion, ElevenLabs, MiniMax, Mureka, Treblo (formerly Sonauto), Google Lyria and FlowMusic. We retrain as new models appear. Current version: v4.3.
How accurate is it?
We've done our own testing as well as participated in 3rd party industry tests and our model consistently comes out as more than 99% accurate. Songs downloaded straight from an AI platform are nearly always caught. Edited stems, re-mastering and mixed human and AI production make it less certain.
Things that can make a human song come back as AI:
Compressed audio. Streaming links and low-bitrate MP3s strip detail the model relies on. This is why we suggest you upload a WAV or 320kbps MP3 instead.
Short clips. Under 60 seconds there's less to analyze, and the result is less reliable.
Heavy processing. Pitch correction, extreme compression, and ASMR or ambient textures can look synthetic.
AI mastering or mixing tools. Things like iZotope Ozone or LANDR use AI functionality that can leave artifacts behind.
A result is a probability, not proof. Please don't use it to accuse anyone. If you think a result is wrong, mark it Inaccurate and tell us about the song. That feedback goes into the next version.