How I used Suno AI to generate 70 different covers of my wife song

My wife, Fiamma D’Avino, surprised me at our wedding with a song she wrote herself.
Gaia Villò helped refine the lyrics and compose the music, and the performance was unforgettable.

Later, I discovered a video (in Italian) by Mark “The Hammer” that explained how various AI services can generate music.
I decided to try it out, but I already had the lyrics and a live recording, so I turned to Suno AI to create many different versions of our song.

What is Suno AI?

Suno AI offers a free plan that I used as much as possible.

When you sign up, no credit‑card is required, so you can start creating music immediately.
The platform also includes a gamified credit system that rewards you with a large batch of credits during the first 24 hours.

After that, you receive 50 credits each day.
In addition, performing actions such as publishing a song, creating a playlist, etc., unlocks a bonus of another 50 credits (that I did 5 times) each time you complete those actions.

Generating a single song (usually less of 1 minute to generate a song) costs 10 credits, so I was able to produce 70 different versions of my wedding song in just two days.
This is possible because each request generates two distinct tracks from the same prompt, effectively doubling the output.

Credits can also be spent on remixing or editing a track, but I was an exception—I used the credits solely for generating new songs (so I have no idea how works the remix/edit feature).

How I used it

Suno provides two separate fields: one for the lyrics (which I already had) and one for a textual prompt.
My first prompt was very simple:

“using these lyrics, create an electro‑swing song”

This produced a number of tracks, but for several genres the result was ambiguous or completely unexpected.
Because the model did not receive enough instructions, it sometimes generated music that I could not even classify.

I used this prompt in some cases:

“convert this song with this lyrics in the style of X”

Suno blocks the use of “banned words” such as specific artist names.
If an artist is not in the model’s knowledge base (so avoiding that filter), the system hallucinates and often produces a track in a genre that bears no resemblance to what I intended.

To work around the artist limitation, I used Suno’s chat again.
I asked the assistant to “generate a Suno prompt that uses these lyrics and this audio in the style of X.”
The assistant returned a detailed prompt that included often:

• tempo and BPM suggestions,
• instrumentation specifics (e.g., “brass section for swing”, “synth bass for electro”), and
• any additional mood or production notes.

Supplying this richer context dramatically improved the quality of the generated songs compared with my initial, very short prompts.

Suno includes an internal chat feature (free from credits). I selected a generated song and asked:

“What musical genre is this track in a couple of words?”

The answer helped me label every version in the playlist.
It is seems that now that Chat feature is not available anymore.
You can see the full playlist, together with the exact prompts I used for each track, here:

https://suno.com/playlist/64f5629a-4d70-483e-a917-7a6051102022

Even on the free tier, Suno allowed me to download each generated track as an MP3 file—a feature I hadn’t expected to be available without a paid subscription.

Thinking of is incredible what is possible to achieve the 4.5 model (the 5 version is only on the Pro plan).

In some of the generated songs the vocal track isn’t perfect—certain words are missing or cut off.
Suno AI’s Edit feature lets you separate the individual stems (vocals, drums, bass, etc.). When a vocal segment is incomplete you can re‑record that part with a singer and replace the stem, restoring the missing words.
Anyway I never used that feature because I am not a singer.

For many tracks the backing music is clearly recognisable and appears to be based on the original recording I uploaded.
In other cases the AI improvises more freely, producing a completely new accompaniment.

In a few cases I tried to recreate an Italian‑style version of the song by applying the same tricks I described earlier.
The model seems to struggle with non‑English music, so the results didn’t match the sound I had in mind.

Some of the generated versions turned out poorly because the musical arrangement and the lyrics simply didn’t fit the chosen genre.
Also there are some cases when for a real person it will be very difficult to sing because there is no time for the vocals to breath.

Generate a video for every song

The next step after creating the audio was to produce a video for each song.
The nice thing about Suno AI is that every MP3 it generates already contains a cover‑art image embedded in the file’s metadata.

Because I wanted to keep the whole process free, I turned to my Linux skills and wrote a Bash script (GitHub Copilot, ChatGPT, and Mistral) that assembles a video from the audio, the cover art, and the ID3 tags (artist, title, etc.).

• ffmpeg – Mistral suggested using ffmpeg, and I didn’t know that the tool could also create videos from a single image and an audio track.
• ID3 metadata – I have a huge personal MP3 collection (20 years old, I don’t use Spotify), so I’m used to editing ID3 tags. I cleaned up all the tags with Kid3, which made the artist and title information readily available for the script.

The script:

1. Scans a directory for MP3 files.
2. Extracts the embedded cover‑art image.
3. Reads the ID3 fields (artist, title).
4. Uses ffmpeg to combine the audio, the cover image, and a simple waveform visualisation into a single MP4 video.

You can view the script here:

https://github.com/Mte90/My-Scripts/blob/master/misc/video-generator.sh

The script took for every video something from 15-20 seconds to generate. With the help of those AI I was able to generate the ffmpeg ruleset I wanted for the final video.

PS: Suno offers a video‑generation feature in its Pro plan, but it consumes credits.

All the videos were uploaded to YouTube and are available in a single playlist:

https://www.youtube.com/playlist?list=PLTW7CYCRxoFGoQuOclyHup61y6q02CRjw

Copyright

It is public domain with no right (my wife approved).
For that reason all the songs are available on Suno with permissions to do a remix, just let me know if you are doing this.

PS: I used the genre to label every song to avoid any issues with artists.

Liked it? Take a second to support Mte90 on Patreon!