Harder to detect manually compared to image or video, but not necessarily harder to detect with another model. If it’s AI-generated MIDI (I have no idea if that’s the, or even a, way it’s done) there are probably patterns in the output similar to the way there are in generated text, but if it’s actually generating the audio itself then that should be pretty distinct at the finer-grained level that a model could analyze it at.
Harder to detect manually compared to image or video, but not necessarily harder to detect with another model. If it’s AI-generated MIDI (I have no idea if that’s the, or even a, way it’s done) there are probably patterns in the output similar to the way there are in generated text, but if it’s actually generating the audio itself then that should be pretty distinct at the finer-grained level that a model could analyze it at.