Playing Around with Suno
You have to know the enemy by sight, so I periodically try to test the limits of AI slop. Today I generated a concept album in Suno. Let me disclaim right away: there is not a single real sound in the entire recording.
Look at this wonderfully pompous description:
NO DIAL TONE is an album about nostalgia. Naive 2000s synths, heavy electronicore guitars, and high screams come together into a story about returning to a place where everything is still in its proper place, but your presence there has already been erased.
More seriously, I was not interested in checking whether a song can be generated in one shot. I had done that before. I wanted to see whether, without any special knowledge, it is possible to assemble something that feels more or less like a complete statement: with a concept, development, recurring images, different genre roles for the tracks, and a relatively coherent stylistic line.
Gemini Flash 3.5 came up with the album idea and the lyrics. Then, in an agent with GPT-5.5, I refined everything into a form that could be fed to Suno: structure, prompts, BPM, genres, vocal delivery, and the overall dramaturgy. And I tried to make it varied, too. After about twenty attempts I got the nostalgic sound I like and propagated it across all tracks.
In the process I even started to believe that AI in this kind of case can be perceived not only as a garbage generator, but as a tool for self-expression at a slightly higher level of abstraction. You are not playing every part by hand, but you are still making artistic decisions: setting the frame, choosing the direction, rejecting things, refining them, arguing with the result, and gradually pulling something resembling your own statement out of the model. The next step is probably to go into studio mode and edit tracks separately.
No joke, I can already listen to this. If there were no vocals, it would be hard to distinguish from simply good music. The vocals still reveal their neural nature through pronunciation and manner, especially in Russian. So a lot of the work is about making the vocal another instrument rather than a thing pasted on top.
In places it is Attack Attack, in places early Eskimo Callboy and I See Stars, and in places something like a mix of Perturbator, Pendulum, and BMTH.
A curious experiment, in short. On top of that, I generated a visualizer with ffmpeg and Opus 4.8. Scary things can be done on your knees in a few hours of free time.