Curious about this myself: What methods are being tried to improve AI-generated music? For example, what approaches are there to enhance the naturalness of the sound? Or how do people enrich AI outputs with a human touch? What steps do you usually follow? Also, there's this thing called "prompt engineering"—what should we pay attention to there?
How are people progressing with AI composing songs?
👁️ 4 views💬 2 replies❤️ 0 likes
2 Replies
Oh man, it's actually pretty cool to listen to AI-generated tracks, especially when you sample some voice and manually layer something live on top. But how do you really control the prompts? Definitely not just writing clichés like "create a sad melody in minor key"—are there any specific tricks to get a cleaner result?
AI-assisted music composition feels a lot like programming to me—it’s a mix of automation and manual touches.
One common thing we run into is "post-processing" to enhance the natural feel of AI-generated tracks. Raw outputs from AI can sometimes sound mechanical or uneven, so producers manually tweak different layers using tools like EQ, compressors, and reverb to add that organic touch. Similarly, it’s super common to edit MIDI data by hand, stretching out AI-generated rhythms to give them a human feel—kind of like optimizing a program’s output.
Prompt engineering is where things get interesting because it’s about guiding the AI with musical context. Instead of saying, "Jazz trio, 1970s sound, harmonious instruments," you might get better results with something like, "Heavy bassline, mid-tempo, bluesy vibe." It’s like calling a function in code—you’re fine-tuning the parameters to get the best output. Limiting prompts with constraints like "less noise, more harmony" also helps refine the results.